Genetically-Designed Time Delay Neural Networks for Multiple-interval
Urban Freeway Traffic Flow Forecasting
Ming Zhong
Dept. of Civil Engineering, University of New Brunswick
Fredericton, New Brunswick, Canada E3B 5A3
E-mail: ming@unb.ca
Satish Sharma
Faculty of Engineering, University of Regina
Regina, Saskatchewan, Canada, S4S 0A2
E-mail: satish.sharma@uregina.ca
Pawan Lingras
Dept. of Mathematics and Computing Science, Saint Mary’s University
Halifax, Nova Scotia, Canada B3H 3C3
E-mail: pawan.lingras@stmarys.ca
(Submitted on April 21 and July 7, 2006; Accepted on August 8, 2006)
Abstract— Previous research for short-term traffic prediction mostly forecasts only one
time interval ahead. Such a methodology may not be adequate for response to emergency
circumstances and road maintenance activities that last for a few hours or a longer period.
In this study, various approaches, including na?ve factor methods, exponential weighted
moving average (EWMA), autoregressive integrated moving average (ARIMA), and
genetically-designed time delay neural network (GA-TDNN) are proposed for predicting
traffic flow of continuous 12 hours ahead on a freeway near the City of Calgary, Canada.
Study results show that the ARIMA models outperform EWMA models, which in turn
superior to the factor methods. GA-TDNN results in only comparable accuracy with the
ARIMA model, and it seems not worth to develop such complicated models. However, the
adaptive nature of neural networks promises better accuracy as they are exposed to more
observations during field operation. Its non-parametric approach also guarantees a greater
portability and much faster computing speed for real-time applications.
Keywords— Time-delay neural network (TDNN), autoregressive integrated moving
average (ARIMA), traffic forecasting
1. Introduction
Traffic Management on urban freeways is crucial for providing mobility in city and surrounding areas. Urban freeways connect central business districts (CBD) with functional districts (such as residential, recreational, and business), suburbs and areas outside the cities. An urban freeway traffic management system (UFTMS) is to monitor traffic in real time, detect any impedances of mobility (e.g., accident or congestion), and activate proper actions (e.g., emergency response or ramp control) to keep traffic moving smoothly.
Successful UFTMS is largely dependent on the information provided from an advanced traveler information system (ATIS). An ATIS is required to monitor real-time traffic situations and be able to predict short-term evolutions. Then the information from ATIS will be passed to UFTMS to maximize the capacity of road networks.
201
It is evident from the literature that extensive research has been carried out for predicting short-term traffic on urban freeways [1-8]. However, two potential problems for real-world implementation can be identified from the previous research. The first one is that most of these research works forecast traffic only one time interval (e.g., five minutes or one hour) ahead. Such a methodology can not accommodate the need of real-time traffic control and operation in response to special situations, such as accidents, congestions, or road maintenance activity. These events could last for a few hours or a longer period, and models that can forecast the evolution of traffic situations for multiple intervals ahead are thus needed. The second problem is the lower accuracy of the previous research. The literature review indicates that average errors from the previous studies are mostly more than 7-8%. A few studies [5, 9] did try to provide non-parametric models for forecasting multiple-interval traffic. However, most of the mean absolute percent errors (MAPE) of forecasting in these studies are more than 9-10%. More accurate models are needed for successful implementation of an UFTMS.
In this study, various models, such as factor, time series analysis, and genetically designed neural networks, are used to forecast 12-hour traffic volumes ahead on an urban freeway in the City of Calgary, Canada. The advantages, disadvantages, and accuracy of these methods are compared, and recommendations are made for real-world ATIS implementations.
2. Study Data
Traffic volume data collected from 1996 to 2000 on a section of freeway near the city of Calgary, Alberta are used in this study. The study site locates on the Highway # 2, which is the main corridor between Calgary and Edmonton. This section of highway is also a part of CANAMEX Corridor, A Trans-North-American system for connecting Canada, the United States and the Mexico. The data used in this study is in the form of hourly volumes from individual directions. Figure 1 shows typical weekly pattern, which starts from a Saturday in the July, for the northbound and southbound respectively. It is interesting to notice that the southbound traffic (for going to the Calgary) shows both morning and evening peak, whereas the northbound traffic shows evening peak only. If the freeway solely serves the commuter traffic, there would be the morning peak only for the southbound (going to workplaces in the Calgary) and the evening peak for the northbound (going to home in the suburb). However, the evening peaks on the weekdays in the southbound are interesting. Such peak may result from the long-distance traffic from other major cities (e.g., Edmonton) or areas (e.g., Northwest Territories). The traffic generated from these places usually arrives in the Calgary before the sunset and result in the evening peaks. Morning peaks are usually more intense than those of the evenings in the southbound.
Figure 1 shows that there are large traffic volume differences between peak hours and non-peak hours. The differences for the northbound are over 1,700 vehicles per hour and over 1,500 vehicles per hour for the southbound. Traffic volumes are around or over 2,500 vehicles per hour for the morning peaks of the southbound and the evening peaks of the northbound. It is likely that congestion occurs in these periods due to capacity constraint or accidents. Therefore, 12-hour traffic forecasting covering both morning peak and evening peak (from 8:00 a.m. to 8:00 p.m.) is proposed and tested in this study. Such multi-interval traffic forecasting is essential for real-time traffic control and operation on the freeways like the one under study.
3. Study Models
3.1 Na?ve models
There are many intuitive short-term traffic prediction methods. Here we call them as na?ve approaches. These approaches simply take historical values as estimates or use historical values to calculate weighted average estimators. Stephanedes et al. [10] proposed a historical average approach for real-time traffic control. It was reported that the proposed algorithm was considerably better than the best of the past algorithms, such as those designed for control intervals on the order of 5-15 min (referred as the second generation algorithms) and those designed for a cycle-by-cycle basis (referred as the third generation algorithms). The virtue of such methods is their simplicity. However, the results are usually less accurate than more sophisticated models. Koutsopoulos and Xu [11] demonstrated that the use of historical data only as a basis of real-time traffic control was significantly inferior to the use of current and predictive information.
The following na?ve models are used to estimate hourly volumes at the study site:
(1)Last year value (LYV) model
The model simply takes the hourly volume from the same hour in the last year as estimate.
202
(a)
(b)
Figure 1. Short-term traffic patterns of study site. (a) Hourly patterns for the Northbound; (b) Hourly pattern for the Southbound.
(2)Last month value (LMV) model
The LMV model simply takes the hourly volume from the same hour in the last month as estimate.
(3)Historical average (HA) model
The HA model uses the average of historical values from the last three years as estimate.
3.2 Time series analysis models
A time series is a chronological sequence of observations on a particular variable. Time series data are often examined in hopes of discovering a historical pattern that can be exploited in the forecast. Time series modeling is based on the assumption that the historical values of a variable provide an indication of its value in
the future [12].
203
204
(1) Exponential weighted moving average (EWMA) model
The historical values from the same hours of the last 12 weeks are used to compute an exponential-weighted-moving-average estimate with the Equation (1). The value of coefficient θ is estimated from the data.
n t n t t t t x x x x x
??????++?+?+?=13221)1(...)1()1()1(?θθθθθθθ (1) where x t-1 represents the most recent observation, and x t-n represents the oldest observation; θ dictates the degree of filtering, which is a constant such that 0<θ<1.
(2) Autoregressive integrated moving average (ARIMA) model
The traffic volume patterns from the last 8 days (same days of the week, e.g., Wednesdays) are used to develop ARIMA model and estimate 12 hourly volumes on the 9th day. The ARIMA(0,1,1)(0,1,1)12 are identified as the best ARIMA models and used to estimate 12-hour short-term traffic ahead on the freeway under study.
3.3 Genetically designed TDNN models
The variant of neural network used in this study is called time delay neural network (TDNN) [13]. It consists of three layers: input layer, hidden layer, and output layer. Each layer has one or more neurons. TDNN is particularly useful for time series analysis. The neurons in a given layer can receive delayed input from other neurons in the same layer. Figure 2 shows the TDNN used in this study. For example, the network receives a single input from the external environment. The remaining nodes in the input layer get their input from the neuron on the left delayed by one time interval. The input layer at any time will hold a part of the time series. Such delays can also be incorporated in other layers.
Many researchers have used GAs to determine neural network architectures [14-17]. A previous study [18] discussed the advantage and disadvantages of the different methods and an approached proposed by Lingras et al. is chosen for its simplicity and feasibility. First one week-long hourly volume time series (7 × 24 = 168 hourly volumes) before the predicted hour are presented to GAs for selecting 24 final inputs. The candidate input set is limited to 168 hourly volumes by assuming that the time series in this period contains all necessary information for short-term prediction of next hour. The number of final input variables is decided to be 24 because experiments indicated further increasing the number of final inputs led to little or no improvement on model’s accuracy. The selection criterion is based on that these 24 input variables have the maximum correlation with
predicted hourly volume, among all the combinations of selecting 24 from 168 variables (16824C ). The final input
set is then fed into the TDNN for prediction.
The GA population size is set to 110. GAs are allowed to evolve for 1000 generations. Single point crossover operator is used, and the crossover rate is set at 90%. For mutation, random replacement operator is used and the probability of mutation is set to 1%. These parameters were chosen after experimenting with Output layer
Figure 2 Time delay neural network structure
205
different values. The best chromosome selected after 1000 generations evolution was used as the final solution of the search. The connections selected by the genetic algorithm were used to design and implement the neural network models.
There are 168 neurons in the input layer of the TDNN models. However, only 24 of them have the connections with the hidden layer neurons. There are 12 neurons in the hidden layer and 1 neuron in the output layer.
In order to obtain accurate estimation, a TDNN was developed for each hour of each day of the week. The training observations for these models are from the same hours (e.g., 7:00-8:00 a.m.) on the same days of the week (e.g., Wednesday) in a given season (e.g., July and August). These individual models are fused together later for implementation.
All the models were trained and then tested. Depending on the model, the number of observations varied. The absolute percentage error (APE) was calculated as:
100 ×?=volume actual volume
estimated volume actual APE (2)
The key evaluation parameters consisted of the average and 95th percentile errors. These measures give a reasonable idea of the error distributions by including (e.g., when calculating average errors) or excluding (e.g., when calculating the 95th percentile errors) large errors caused by outliers or special events.
4. Results and Discussion
The models mentioned above were applied to the field data for short-term traffic flow forecasting. These models were used to predict 12-hour traffic volume ahead on both Wednesdays and Sundays in July and August. The results for Wednesdays and Sundays are essentially same, so that only the experiment results for Wednesdays are presented here. For comparison purpose, the average and 95th percentile errors for individual travel directions (northbound and southbound) are listed in the same tables.
It was found that the LYV model resulted in the largest errors for the northbound, but the smallest errors for the southbound. Simply taking the value from the last year can not provide good estimate for the northbound because it experienced a dramatic increase during the study period. The LYV model consistently underestimates traffic in the forecasting year. The largest average and 95th percentile error is 34.1% and 81.7% respectively for the morning peak hour 7:00-8:00 a.m. It also resulted in large errors for the evening peak hours (e.g., from 4:00 – 6:00 p.m.). The average errors during this period are more than 20%. In contrast, the southbound traffic was fairly stable during such period and LYV result in the average errors of 5-7%, and the 95th percentile errors of 11-15% (the results are not shown here).
Study results indicated that the HA model is helpful for providing stable predictions, compared to the LYV model. It resulted in similar accuracy for both travel directions. The average errors are usually between 9% and 12% and the 95th percentile errors range from 13% to 18%. The mean average errors over the 12 successive hours are 11% and the mean 95th percentile errors are around 16% for both travel directions. The LMV model seemed provide fairly reasonable estimates for both directions. Except for a few hours of the southbound traffic, average errors are usually from 5% to 8%. Such average errors indicate that there were small seasonal variations during the study period. However, fairly large 95th percentile errors (usually more than 15%) imply that simply taking a value from the last month can result in very bad estimates. The mean 95th percentile errors for the LMV model are 17-18% and larger than those of the HA model (the results are not shown here).
Table 1 shows the prediction errors of the EWMA and ARIMA models. Compared with the na?ve methods, the time series analysis models show higher accuracy. The average errors of these models are 4-5%, which are 2% or more less than those of the na?ve methods. The mean 95th percentile errors of the EWMA model are 12-13%, whereas those for the ARIMA model are only 7-8%. It can also be seen that the ARIMA model outperforms the EWMA model in most cases. The techniques employed in the ARIMA models allow considering both seasonal variation and the trend of time series and contributing to such improvements.
The final input variables selected by GAs were used to train TDNN models based on a set of training data. The trained models were then used to forecast short-term traffic in the testing year. Table 2 shows the testing errors of genetically designed TDNN models. The average errors of the TDNN models are usually 4-5%, and the 95th percentile errors are around 9-10%. Study results show that the TDNN models have comparable or a slightly better accuracy than the ARIMA models. In terms of the average errors, the TDNNs outperform the ARIMA models in 6 out of 12 cases for the northbound and 7 cases for the southbound. It seems that the complexity involved in the GA-TDNN models can not be readily warranted given such kind of “improvement”. However, it
should be noted that TDNN models can continuously adapt themselves to new observations and have the potential for more accurate estimates during field operation. Therefore, it is possible that the TDNN models may result in better estimates than the ARIMA models after a certain period of implementation. Moreover, there is the least accuracy differences between two directions (the mean difference over 12 hours is about 3%) within all approaches, and thus indicate a good portability of TDNN models. Such a feature makes them very desirable for ATIS implementation in a large area (e.g., statewide) as onerous work of specifying specific models for individual locations may be avoided. In general, potential for more accurate forecasts, fast computing speed and greater portability make TDNN attractive for real-world ITS applications.
Table 1. Errors of Time Series Analysis Models
Prediction Errors
Hour Average 95th%
EWMA1ARIMA2 EWMA ARIMA
N S N S N S N S
(1) (2) (3)(4)(5) (6)(7)(8)(9)
07-08 5.0 9.5 5.4 5.3 9.812.99.99.3
08-09 3.9 7.0 4.6 3.1 10.112.38.7 5.1
09-10 5.4 5.57.5 4.3 14 13.513.2 6.6
10-11 5.0 4.3 4.5 1.7 15.814.97.4 2.9
11-12 5.9 4.8 4.0 5.1 16.916.1 5.99.3
12-13 4.5 5.1 2.7 5.0 13.117.1 4.79.6
13-14 4.9 5.7 5.0 6.2 10.815.810.512.2
14-15 2.9 4.3 5.1 3.0 8.915.310.5 6.2
15-16 3.4 3.1 3.8 2.9 8.98.87.0 4.0
16-17 4.5 3.1 2.5 4.0 9.58.6 4.4 6.4
17-18 5.5 3.4 3.57.8 11 12.6 6.815.0
18-19 5.4 5.0 6.9 4.0 15.113.713.3 6.9
Total-
Average 4.7 5.1 4.6 4.4 12 13.58.57.8
6. Conclusions
Urban freeways are important for providing fast mobility within cities and suburbs. It is critical to implement advanced traffic management systems (ATMS) on the freeways in or near large cities. The performance of ATMS largely relies on the information from corresponding ATIS. The major role of ATIS is to monitor real-time traffic situation and to predict a short-period evolution. The information obtained from ATIS is then fed into ATMS for decision-marking purposes. Proper traffic management and control can be implemented to avoid congestions and deal with accidents.
Literature review indicates that the short-term prediction models developed in previous research mostly forecast only one time interval ahead. Such models are not useful for the situation where operational strategy for a longer period (e.g., several hours) is needed (such as road construction or hazardous material incidents). In this study, various short-term traffic prediction models, including na?ve methods, time series analysis, and genetically designed TDNN models, are proposed and tested on the data collected from a section of freeway on CANAMEX (near the City of Calgary, Canada). These models are used to forecast traffic volumes of 12 successive hours ahead in both travel directions.
Na?ve methods used in this study include the model directly using the last year value (LYV), the model directly using the last month value (LMV), and that using the average value of last few years (HA). The virtue of such methods is their simplicity. However, the accuracy of these models is usually low. Simply taking a value from the last year or the last month as estimate could result in fairly large prediction errors. For example, the 95th percentile error for the LYV model is as high as 82% for the hour 7:00 – 8:00 a.m. in the northbound. Although the average mechanism of the HA model is helpful for reducing large estimation errors, most of the 95th percentile errors are still more than 15%.
206
Table 2. Errors for GA Designed TDNN Models
Errors
Prediction
Hour Average 95th%
N S N S
(1) (2) (3) (4) (5)
07-08 4.40 3.07 10.40 6.68
08-09 2.92 2.63 7.31 5.11
09-10 5.70 1.96 8.12 5.29
10-11 4.10 4.75 9.90 11.12
11-12 4.01 4.42 7.75 9.56
12-13 3.79 2.94 8.38 6.22
13-14 4.56 3.75 8.10 8.63
14-15 6.21 3.90 10.22 7.88
15-16 6.35 4.97 10.86 7.58
16-17 4.80 6.84 6.96 11.81
17-18 7.61 5.89 11.92 10.38
18-19 5.80 9.02 11.91 21.09
Total-
Average 5.02 4.51 9.32 9.28
Both the exponential weighted moving average (EWMA) and the autoregressive integrated moving average (ARIMA) are used to forecast multi-interval traffic on the study freeway. Compared with the na?ve methods, the time series analysis models are able to incorporate more historical data for predictions. For example, the EWMA models use the weighted averages of a series of historical values (12 cases in this study) as estimates. Moreover, for the ARIMA models, differencing is used to deal with the trend and seasonality of time series. Consequently, these models result in improved accuracy. The average errors are usually around 5-6%, and the 95th percentile errors for the ARIMA models are as low as 8%.
Study results show that the GA-TDNN models show comparable or only a slightly better accuracy than the ARIMA models. However, it should be noted that, due to their adaptive nature, TDNN models can continuously improve themselves when exposed to new observations. Therefore, they should provide better predictions than those reported in Table 2 during field operation. The potential of higher accuracy and good portability make them very attractive in future ITS implementation.
The genetically designed TDNN models used in this study seem very complicated and difficult for implementation. However, the techniques involved in these models can be easily integrated in a software package. These models can be developed offline based on the software package and historical data. The developed models can then be used online for real-time traffic forecasting and control. The experiments carried out for this study show that these models can provide both accurate and timely predictions based on a Linux server and workstation computing facility.
Acknowledgement
The authors are grateful towards NSERC, Canada for their financial support and Alberta Transportation for the data used in this study.
References
[1] Ahmed, S.A. and Cook, A.R.: Analysis of Freeway Traffic Time-Series Data by Using Box-Jenkins
Techniques. Transportation Research Record 722, Transportation Research Board, Washington D.C. (1979) 1-9.
[2] Lee, S.S. and Fambro, D.B.: Application of Subset Autoregressive Integrated Moving Average Model for
Short-Term Freeway Traffic Volume Forecasting. Transportation Research Record 1678, Transportation Research Board, Washington D.C. (1999) 179-189.
207
208[3] Lu, J.: Prediction of Traffic Flow by an Adaptive Prediction System. Transportation Research Record 1287, Transportation Research Board, Washington D.C. (1990) 54-61.
[4] Park, B., Messer, C.J., and Urbanik, T.: Short-Term Freeway Traffic Volume Forecasting Using Radial Basis Function Neural Network. Transportation Research Record 1651, Transportation Research Board, Washington D.C. (1998) 39-47.
[5] Smith, B.L., and Demetsky, M.J.: Short-Term Traffic Flow Prediction: Neural Network Approach. Transportation Research Record 1453, Transportation Research Board, Washington D.C. (1994) 98-104.
[6] Williams, B.M., Durvasula, P.K. and Donald, E.B.: Urban Freeway Traffic Flow Prediction: Application of Seasonal Autoregressive Integrated Moving Average And Exponential Smoothing Models. Transportation Research Record 1644, Transportation Research Board, Washington D.C. (1998) 132-141.
[7] Zhang, H., Ritchie, S.G., and Lo, Z.P.: Macroscopic Modeling of Freeway Traffic Using an Artificial Neural Network. Transportation Research Record 1588, Transportation Research Board, Washington D.C. (1997) 110-119.
[8] Shimizu, H., Yamagami, K. and Watanabe, E.: Applications of State Estimation Algorithms to Hourly Traffic Volume System. Proceedings of the Second World Congress on Intelligent Transport Systems , Yokohama, Japan, (1995) 72-77.
[9] Smith, B.: Meeting Real-Time Traffic Flow Forecasting Requirements with Imprecise Computations. Computer-Aided Civil and Infrastructure Engineering (18), (2003) 201-213.
[10] Stephanedes, Y.J., Michalopoulos, P.G., and Plum, R.A.: Improved Estimation of Traffic Flow for Real-
time Control. Transportation Research Record 795, Transportation Research Board, Washington D.C., (1981) 28-39.
[11] Koutsopoulos, H. and Xu, H.: An Information Discounting Routing Strategy for Advanced Traveler
Information Systems (ATIS). Transportation research Part C, Vol.1, No.3, (1992) 249-264.
[12] Box, G. and Jenkins, J.: Time Series Analysis: Forecasting and Control. San Francisco, Holden-Day (1970).
[13] Hecht-Nielsen, R.: Neurocomputing . Addison-Wesley Pub. Co, Don Mills, Ontario, (1990).
[14] Harp, S., Samad, T. and Guha, A.: Towards the Genetic Synthesis of neural networks. In: D. Shaffer (Ed.),
Proceedings of the Third International Conference on Genetic Algorithms . San Mateo, Morgan Kaufmann, (1989) 360-369.
[15] Miller, G., Todd, P. and Hedge, S.: Designing Neural Networks using Genetic Algorithms. In: D. Shaffer
(Ed.), Proceedings of the Third International Conference on Genetic Algorithms . San Mateo, Morgan Kaufmann, (1989) 53-60.
[16] Montana, D. and Davis, L.: Training Feedforward Networks using Genetic Algorithms. In: N. Sridhara
(Ed.), Proceedings of Eleventh International Joint Conference on Artificial Intelligence , (1989) 762-767.
[17] Hansen, J.V., McDonald, J.B. and Nelson, R.D.: Time Series Prediction with Genetic Algorithm Designed
Neural Networks: An Experimental Comparison with Modern Statistical Models. Computational Intelligence , 15(3), (1999) 171-184.
[18] Lingras, P.J., Sharma, S.C. and Zhong, M.: Prediction of Recreational Travel using Genetically Designed
Regression and Time Delay Neural Network Models. Transportation Research Record 1805, Transportation Research Board, Washington, D.C., U.S.A. (2002) 16-24.
Ming Zhong is an assistant professor at the Department of Civil Engineering, University
of New Brunswick, Canada and he is also a research fellow affiliated with the Institute of
Advanced Policy Research of the University of Calgary, Canada. He obtained his PhD at
the University of Regina, M.A.Sc. at the Beijing Jiaotong University, China, and his
B.Eng. at the Tongji University, China. His research interests include Land Use
Transport Interaction Modelling, Intelligent Transportation System, Traffic Data
Analysis, GIS and remote sensing applications in Transportation.
Pawan Lingras’ academic background includes Computer Science and Civil/ Array
Transportation Engineering. His research interests include theoretical and experimental work on Rough and Fuzzy Computing, Neural Networks, Genetic Algorithms, Time
Series Analysis, Data Mining, and Intelligent Transportation Systems.
209
WHEN与WHILE用法区别 when, while这三个词都有"当……时候"之意，但用法有所不同，使用时要特别注意。 ①when意为"在……时刻或时期"，它可兼指"时间点"与"时间段"，所引导的从句的动词既可以是终止性动词，也可是持续性动词。如： When I got home, he was having supper.我到家时，他正在吃饭。 When I was young, I liked dancing.我年轻时喜欢跳舞。 ②while只指"时间段"，不指"时间点"，从句的动词只限于持续性动词。如：While I slept, a thief broke in.在我睡觉时,盗贼闯了进来。 辨析 ①when从句与主句动作先后发生时，不能与while互换。如： When he has finished his work, he takes a short rest.每当他做完工作后，总要稍稍休息一下。(when = after) When I got to the cinema, the film had already begun.当我到电影院时，电影已经开始了。(when=before) ②when从句动词为终止性动词时，不能由while替换。如： When he came yesterday, we were playing basketball.昨天他来时，我们正在打篮球。 ③当从句的谓语是表动作的延续性动词时，when, while才有可能互相替代。如：While / When we were still laughing, the teacher came in.正当我们仍在大声嬉笑时，老师进来了。 ④当从句的谓语动词是终止性动词，而且主句的谓语动词也是终止性动词 时,when可和as通用，而且用as比用when在时间上更为紧凑，有"正当这时"的含义。如： He came just as (or when) I reached the door.我刚到门那儿，他就来了。 ⑤从句的谓语动词如表示状态时，通常用while。如： We must strike while the iron is hot.我们应该趁热打铁。 ⑥while和when都可以用作并列连词。
when 和 where 引导定语从句的用法 定语从句既是英语语法的一个重点，同时又是一个难点。说它是难点，主要难在两点上：一点是如何正确判断什么样的汉语句子要译为英语的带定语从句的复合句；另一点是定语从句的引导词较多(包括关系代词who, that, which, as 和关系副词when, where, why)，而且其用法也较复杂。那么到底什么情况下用when和where来引导定语从句呢？它们又该怎么用呢？下面就举例说明： 一、when：当主句中的先行词(即主句中被后面定语从句修饰的词)是表示时间意义的名词时，它只能作定语从句的时间状语，放在定语从句句首。如果定语从句的引导词是作该定语从句的主语或宾语，则要改用关系代词that或which来引导。例如： The days when we used foreign oil are gone. 我们用洋油的日子一去不复返了。 I'll never forget the day when I was born. (=I'll never forget my birthday.) 我永远不会忘记我出生的日子。 It happened in November when the weather was wet and cold. 这事发生在天气又湿又冷的十一月。 In the years that (which) followed, Marx kept on studying English and using it. 在这之后的几年中，马克思继续学习和使用英语。(that作定语从句"that followed"的主语) The day (that) I always remember in all my life is my birthday. 我一生中最难忘的日子是我的生日。(that作定语从句"that I always remember in all my life"的宾语，that可以省略)
一、从句是如何出题的？ 1. 时态 2. 考连接词 3. 考语言顺序 二、学好从句的两个基本条件 1. 时态 2. 从句的三个必须:①必须是句子；②必须有连接词；③必须是陈述句 三、状语从句、宾语从句、定语从句重点 1.如何判断何种从句 2. 从句的时态 3. 从句的连接词与扩展 4. 经典单选、从句与选词、长句子分析 四、如何判断三种从句 1. 状语从句无先行词 2. 宾语(表语)从句无先行词有动词或词组 3. 定语从句先行词多为名词或代词 一、When引导的定语从句与时间状语从句的区分 1. when的译法不同。在时间状语中，when 翻译成“当……的时候” I want to be a teacher when I grow up. 当我长大的时候，我要做一名老师。在定语从句中，when不翻译。I won't forget the day when he says he loves me. 我不会忘记他说爱我的那一天。 2. 在时间状语中，when从句前面或后面是句子；定语从句中，when 从句不能位于句首，且通常when前为表示时间的名词day、year等。 3. when在从句的作用不同。在时间状语从句中，when是连词，只起连接主句和从句的作用，不做从句的任何成分。不过when引导的时间状语从句修饰主句的谓语，做主句的时间状语。 在定语从句中，when是关系副词，在从句中代替先行词做从句的时间状语，修饰从句的谓语。 例1 I will always remember the days when I lived with my
grandparents in the country. 例2 I always remember the days in the country when I see the photo of my grandparents. 点评：例1意为“我会永远记得跟我祖父母一起住在乡下的那些日子”，其中when 引导的是一个定语从句, 修饰the days, when在从句中作时间状语。例2意为“当我看到祖父母的照片时，总是会想起在乡下的那些日子”，其中when 引导的从句并不修饰前面的名词the country，因此可判定为时间状语从句。 例1中的when可用in which替代，即从句可改为...in which I lived with my grandparents in the country. 例2中从句前有名词，但根据句意可 知并不是从句所修饰的对象，也不能用“介词+ which”来替代。 二、判断关系代词与关系副词 方法一：用关系代词，还是关系副词完全取决于从句中的谓语动词。及物动词后面无宾语，就必须要求用关系代词；而不及物动词则要求用关系副词。例如： This is the mountain village where I stayed last year. 这是我去年呆过的山村。 I'll never forget the days when I worked together with you. 我永远不会忘记与你共事的日子。 判断改错： 1. This is the mountain village where I visited last year. 2. I will never forget the days when I spent in the countryside. 3. This is the mountain village （which）I visited last year.
when和while用法区别专项练习 讲解三例句： 1. The girls are dancing while the boys are singing. 2. Kangkang’s mother is cooking when he gets home. 3. When/While Kangkang’s mother is cooking, he gets home. 一、用when或者while填空 Margo was talking on the phone, her sister walked in. we visited the school, the children were playing games. · Sarah was at the barber’s, I was going to class. I saw Carlos, he was wearing a green shirt. Allen was cleaning his room, the phone rang. Rita bought her new dog; it was wearing a little coat. 7. He was driving along ________ suddenly a woman appeared. 8. _____ Jake was waiting at the door, an old woman called to him. 9. He was reading a book ______suddenly the telephone rang. 10. ______ it began to rain, they were playing chess. [ 二、用所给动词适当形式填空 11. While Jake __________ (look) for customers, he _______ (see) a woman. 12. They __________ (play) football on the playground when it _____ (begin) to rain. 13. A strange box ________ (arrive) while we _________ (talk). 14. John ____________ (sleep) when someone __________ (steal) his car. 15. Father still (sleep) when I (get) up yesterday morning. 16. Grandma (cook) breakfast while I (wash) my face this morning. 17. Mother (sweep) the floor when I (leave) home. ~ 18. I (read) a history book when someone (knock ) at the door. 19. Mary and Alice are busy (do) their homework. 20. The teacher asked us (keep) the windows closed. 21. I followed it (see) where it was going. 22. The students (play) basketball on the playground from 3 to 4 yesterday afternoon. / 三、完成下面句子，词数不限 1.飞机在伦敦起飞时正在下雨。 It when the plane in London. 2.你记得汶川大地震时你在做什么吗 Do you remember what you when Wenchuan Earthquake . 3.当铃声想起的时候，我们正在操场上玩得很开心。 We on the playground when the bell . 4.当妈妈下班回家时，你在做什么 % when Mum from work 5.当我在做作业时，有人敲门。 I was doing my homework, someone
过去进行时过去进行时表示过去某一时刻或者某段时间正在进行或发生的动作，常和表过去的时间状语连用，如： 1. I was doing my homework at this time yesterday. 昨天的这个时候我正在做作业。 2. They were waiting for you yesterday. 他们昨天一直在等你。 3. He was cooking in the kitchen at 12 o'clock yesterday. 昨天12点，他正在厨房烧饭。 过去进行时的构成： 肯定形式：主语+was/were+V-ing 否定形式：主语+was not (wasn't)/were not (weren't)+V-ing 疑问形式：Was/Were+主语+V-ing。 基本用法： 1. 过去进行时表示过去某一段时间或某一时刻正在进行的动作。常与之连用的时间状语有，at that time/moment, (at) this time yesterday (last night/Sunday/week…), at+点钟+yesterday (last night / Sunday…)，when sb. did sth.等时间状语从句，如： 1）What were you doing at 7p.m. yesterday? 昨天晚上七点你在干什么？ 2）I first met Mary three years ago. She was working at a radio shop at the time. 我第一次遇到玛丽是在三年前，当时她在一家无线电商店工作。 3）I was cooking when she knocked at the door. 她敲门时我正在做饭。 2. when后通常用表示暂短性动词，while后通常用表示持续性动词，而while所引导的状语从句中，谓语动词常用进行时态，如： When the car exploded I was walking past it. = While I was walking past the car it exploded. 3. when用作并列连词时，主句常用进行时态，从句则用一般过去时，表示主句动作发生的过程中，另一个意想不到的动作发生了。如： I was walking in the street when someone called me. 我正在街上走时突然有人喊我。 4. when作并列连词，表示“（这时）突然”之意时，第一个并列分句用过去进行时，when引导的并列分句用一般过去时。如：
Period____8______ in Unit ___6_________ （Period_____20_____Week 4-6）Subject _Grammar (B) Teaching aims To use the Past Continuous Tense with while and when correctly Teaching interesting points of this class: The differences of when and while in Past Continuous Tense Teaching steps: I. Cooperation and intercourse 1. Check the preparation 2. Discuss in groups II. Question and explanation Check the exercises on page 101, and explain the differences between when and while. when和while的区别是： when只能用于一般时态 while可以用于进行时态 when conj. 在...的时候, 当…的时候 when 在绝大多数情况下，所引导的从句中，应该使用非延续性动词（也叫瞬间动词） 例如：I'll call you when I get there. 我一到那里就给你打电话 I was about to go out when the telephone rang.我刚要出门，电话铃就响了。 但是，when 却可以be 连用。 例如：I lived in this village when I was a boy. 当我还是个孩子的时候我住在这个村庄里。 When I was young, I was sick all the time. 在我小时候我总是生病 while 当...的时候 While he was eating, I asked him to lend me $2. 当他正在吃饭时,我请他借给我二美元。While I read, she sang. 我看书时，她在唱歌。 while 的这种用法一般都和延续性动词连用 while 可以表示“对比‘，这样用有的语法书认为是并列连词 Some people like coffee, while others like tea. 有些人喜欢咖啡, 而有些人喜欢茶。 as 当...之时，一边......一边....... I slipped on the ice as I ran home. 我跑回家时在冰上滑了一跤 He dropped the glass as he stood up. 他站起来时，把杯子摔了。 She sang As she worked. 她一边工作一边唱歌。 III. Rumination and evaluation when, as, while这三个词都可以引出时间状语从句，它们的差别是：when 从句表示某时刻或一段时间as 从句表示进展过程，while 只表示一段时间 When he left the house, I was sitting in the garden. 当他离开家时，我正在院子里坐着。 When he arrived home, it was just nine o'clock.
when 和while的用法区别 ①when是at or during the time that, 既指时间点，也可指一段时间； while是during the time that，只指一段时间，因此when引导的时间状语从句中的动词可以是终止性动词，也可以是延续性动词，而while从句中的动词必须是延续性动词。 ②when 说明从句的动作和主句的动作可以是同时，也可以是先后发生；while 则强调主句的动作在从句动作的发生的过程中或主从句两个动作同时发生。 ③由when引导的时间状语从句，主句用过去进行时，从句应用一般过去时；如果从句和主句的动作同时发生，两句都用过去进行时的时候，多用while引导，如： a. When the teacher came in, we were talking. 当此句改变主从句的位置时，则为： While we were talking, the teacher came in. b. They were singing while we were dancing. ④when和while 还可作并列连词。when表示“在那时”；while表示“而，却”，表对照关系。如： a. The children were running to move the bag of rice when they heard the sound of a motor bike. 孩子们正要跑过去搬开那袋米，这时他们听到了摩托车的声音。 b. He is strong while his brother is weak. 他长得很结实，而他弟弟却很瘦弱。 when，while，as引导时间状语从句的区别 when，while，as显然都可以引导时间状语从句，但用法区别非常大。 一、when可以和延续性动词连用，也可以和短暂性动词连用；而while和as只能和延续性动词连用。 ①Why do you want a new job when youve got such a good one already？（get 为短暂性动词）你已经找到如此好的工作，为何还想再找新的？ ②Sorry，I was out when you called me．（call为短暂性动词）对不起，你打电话时我刚好外出了。 ③Strike while the iron is hot．（is为延续性动词，表示一种持续的状态）趁热打铁。 ④The students took notes as they listened．（listen为延续性动词）学生们边听课边做笔记。 二、when从句的谓语动词可以在主句谓语动作之前、之后或同时发生；while和as从句的谓语动作必须是和主句谓语动作同时发生。 1．从句动作在主句动作前发生，只用when。 ①When he had finished his homework，he took a short rest．（finished先发生）当他完成作业后，他休息了一会儿。 ②When I got to the airport，the guests had left．（got to后发生）当我赶到飞机场时，客人们已经离开了。 2．从句动作和主句动作同时发生，且从句动作为延续性动词时，when，while，as都可使用。
While和When在过去进行时中两者的区别如下： ①when是at or during the time that, 既指时间点，也可指一段时间；while是during the time that，只指一段时间，因此when引导的时间状语从句中的动词可以是终止性动词，也可以是延续性动词，而while 从句中的动词必须是延续性动词。 ②when 说明从句的动作和主句的动作可以是同时，也可以是先后发生；while 则强调主句的动作在从句动作的发生的过程中或主从句两个动作同时发生。 ③由when引导的时间状语从句，主句用过去进行时，从句应用一般过去时；如果从句和主句的动作同时发生，两句都用过去进行时的时候，多用while引导，如： a. When the teacher came in, we were talking. 当此句改变主从句的位置时，则为： While we were talking, the teacher came in. b. They were singing while we were dancing. ④when和while 还可作并列连词。when表示“在那时”；while表示“而，却”，表对照关系。如： a. The children were running to move the bag of rice when they heard the sound of a motor bike. 孩子们正要跑过去搬开那袋米，这时他们听到了摩托车的声音。 b. He is strong while his brother is weak. 他长得很结实，而他弟弟却很瘦弱。 when; while 当……时候 while能用when代替; 但是when却不一定能用while代替. while+从句, 动作一定会延续 when+延续性动词/瞬间动词; when he arrived
1、When 和While 的区别 when,while 都有“当……时候”的意思. when 既可表示某一点时间,也可以表示某一段时间. （一）在when 引导的时间状语从句中,其谓语动词可以是延续性的,也可以是非延续性的,可与主句中的谓语动词同时发生,也可在其后发生. 例如： 1、I was just reading a book when she came into my room. 她走进我房间时,我正在看书. （二）、while 只能表示某一段时间,不能表示某一点时间.在while 引导的时间状语从句中,其谓语动词只能是延续性的,而且也只能与主句中的谓语动词同时发生或存在. 例如： 1、While Jim was mending his bike, Lin Tao came to see him. 正当吉姆修自行车时,林涛来看他. Remark: 由when 引导的时间状语从句，主句用过去进行时，从句应用一般过去时；如果从句和主句的动作同时发生，两句都用过去进行时的时候，多用while 引导，如： a. When the teacher came in, we were talking. 2、倒装知识点 完全倒装 表示地点的副词here, there 置于句首, 且主语是名词(不是代词) 时 表示时间、方向的副词或介词短语置于句首, 且主语是名词(不是代词) 时 作表语的形容词、分词、介词短语、such 置于句首时 部分倒装 “ only+状语” 置于句首, 主句需要部分倒装 具有否定意义或半否定意义的副词以及含否定词的介词短语置于句首作状语时“so 或neither + 助动词/情态动词/be 动词+主语”表示“……也/也不” so/such...that...句型 以had/were/should 开头省略if 的虚拟条件句
过去进行时 过去进行时表示过去某一时刻或者某段时间正在进行或发生的动作，常和表过去的时间状语连用，如： 1. I was doing my homework at this time yesterday. 昨天的这个时候我正在做作业。 2. They were waiting for you yesterday. 他们昨天一直在等你。 3. He was cooking in the kitchen at 12 o'clock yesterday. 昨天12点，他正在厨房烧饭。 过去进行时的构成： 肯定形式：主语+was/were+V-ing 否定形式：主语+was not (wasn't)/were not (weren't)+V-ing 疑问形式：Was/Were+主语+V-ing。 基本用法： 1. 过去进行时表示过去某一段时间或某一时刻正在进行的动作。常与之连用的时间状语有，at that time/moment, (at) this time yesterday (last night/Sunday/week…), at+点钟+yesterday (last night / Sunday…)，when sb. did sth.等时间状语从句，如： 1）What were you doing at 7p.m. yesterday? 昨天晚上七点你在干什么？ 2）I first met Mary three years ago. She was working at a radio shop at the time. 我第一次遇到玛丽是在三年前，当时她在一家无线电商店工作。 3）I was cooking when she knocked at the door. 她敲门时我正在做饭。 2. when后通常用表示暂短性动词，while后通常用表示持续性动词，而while所引导的状语从句中，谓语动词常用进行时态，如： When the car exploded I was walking past it. = While I was walking past the car it exploded. 3. when用作并列连词时，主句常用进行时态，从句则用一般过去时，表示主句动作发生的过程中，另一个意想不到的动作发生了。如： I was walking in the street when someone called me. 我正在街上走时突然有人喊我。 4. when作并列连词，表示“（这时）突然”之意时，第一个并列分句用过去进行时，when引导的并列分句用一般过去时。如： 1）I was taking a walk when I met him. 我正在散步，突然遇见了他。 2）We were playing outside when it began to rain. 我们正在外边玩，这时下起雨来了。 过去进行时和一般过去时的用法比较： 1）过去进行时表示过去正在进行的动作，而一般过去时则表示一个完整的动作。 例如：They were writing letters to their friends last night. 昨晚他们在写信给他们的朋友。（没有说明信是否写完） They wrote letters to their friends last night.
as when while 的区别和用法 as when while的用法 一、as的意思是“正当……时候”,它既可表示一个具体的时间点，也可以表示一段时间。as可表示主句和从句的动作同时发生或同时持续，即“点点重合”“线线重合”；又可表示一个动作发生在另一个动作的持续过程中，即“点线重合”， 但不能表示两个动作一前一后发生。如果主句和从句的谓语动词都表示持续性的动作，二者均可用进行时，也可以一个用进行时，一个用一般时或者都用一般时。 1、As I got on the bus，he got off. 我上车，他下车。（点点重合）两个动作都是非延续性的 2、He was writing as I was reading. 我看书时，他在写字。（线线重合）两个动作都是延续性的 3、The students were talking as the teacher came in. 老师进来时，学生们正在讲话。（点线重合）前一个动作是延续性的，而后一个动作时非延续性的 二、while的意思是“在……同时（at the same time that )”“在……期间(for as long as, during the time that）”。从while的本身词义来看，它只能表示一段时间，不能表示具体的时间点。在时间上可以是“线线重合”或“点线重合”，但不能表示“点点重合”。例如： 1、He was watching TV while she was cooking. 她做饭时，他在看电视。（线线重合） 2、He was waiting for me while I was working. 我工作的时候，他正等着我。（线线重合） 3、He asked me a question while I was speaking. 我在讲话时，他问了我一个问题。（点线重合）
when和while的用法和区别 while和when都是表示同时，到底句子中是用when还是while主要看从句和主句中所使用的动词是短暂性动作（瞬时动词）还是持续性动作。 1、若主句表示的是一个短暂性的动作，而从句表示的是一个持续性动作时，两者都可用。如：He fell asleep when [while] he was reading. 他看书时睡着了。 I met him when [while] I was taking a walk in the park. 我在公园散步时遇到了他。 2、若主、从句表示两个同时进行的持续性动作，且强调主句表示的动作延续到从句所指的整个时间，通常要用while。如： Don’t talk while you’re eating. 吃饭时不要说话。 I kept silent while he was writing. 在他写的时候，我默不作声。 3、若从句是一个短暂性动作，而主句是一个持续性动作，可以when 但不用while。如：When he came in, I was listening to the radio. 他进来时，我在听收音机。 It was raining hard when we arrived. 我们到达时正下着大雨。 4、若主、从句表示的是两个同时(或几乎同时)发生的短暂性动作，一般要用when。如： I thought of it just when you opened your mouth. 就在你要说的时候，我也想到了。 至于什么是短暂性动作，什么是持续性动作，其实有个很简单的规律。就是如果是进行时态，一般是持续性的。如果是过去式，一般是短暂性动作。 对于，填写when还是while的问题，通常首先看主句和从句中的时态，再根据以上4个规律来判断填写那个单词。 When和While的区别 通常when 后面接一般过去时while 后面接过去进行时 ①when是at or during the time that, 既指时间点，也可指一段时间，while是during the time that，只指一段时间，因此when引导的时间状语从句中的动词可以是终止性动词，也可以是延续性动词，而while从句中的动词必须是延续性动词。 ②when 说明从句的动作和主句的动作可以是同时，也可以是先后发生；while 则强调主句的动作在从句动作的发生的过程中或主从句两个动作同时发生。 ③由when引导的时间状语从句，主句用过去进行时，从句应用一般过去时；如果从句和主句的动作同时发生，两句都用过去进行时的时候，多用while引导，如： a. When the teacher came in, we were talking. 当此句改变主从句的位置时，则为： While we were talking, the teacher came in. b. They were singing while we were dancing. ④when和while 还可作并列连词。when表示“在那时”；while表示“而，却”，表对照关系。如： a. The children were running to move the bag of rice when they heard the sound of a motor bike.
when 的用法 一、when 用作副词。 1. 用作疑问副词，引导特殊疑问句。(什么时候，何时 [at what time ])例如： ①When will you come to see me? ②When are they going to visit the Great Wall? 2. 用作连接副词，通常用来引导名词性从句[主语从句、表语从句、同位语从句、宾语从句]及起名词作用的“when ＋动词不定式”结构。(什么时候，何时 [at which; on which ])例如： ①When he comes is not known. [主语从句] ②The morning is when I am busiest. [表语从句] ④I don't know when the plane takes off. [宾语从句] ⑤I don't know when to leave for London. [宾语] 3. 用作关系副词，引导限定性定语从句和非限定性定语从句。(在…的时候 [at/on/in/during which])例如： ①Do you still remember the days when we stayed in America? ②The day will come soon when the Chinese astronauts will go to the moon. ③It happened ten years ago, when I was a child. ④We will go to the countryside at the beginning of June, when the summer harvest will start. 二 . when 用作连词。 1. 用作从属连词，意为“当……的时候[at the time when ]”，引导时间状语从句。例如： ①They learned a lot from the peasants when they stayed in the village. ②It was snowing when he arrived at the station. 【点津】如果 when 引导的从句中的主语和主句中的主语一致，且从句中的谓语动词是“be ＋分词”或从句主语是 it ，则 be 动词及其主语常可省略。例如： ③When（ he was ）asked why he was late, he made no answer. ④I'll tell him about it when（ it is ）possible. 2. 用作从属连词，意为“一……就……”，引导时间状语从句。[immediately after]例如： ①We will stand up when the teacher comes into the classroom. 老师一进教室我们就起立。 ②Fire the rockets when I give the signal. 3. 用作从属连词，意为“还没 / 刚刚/刚一……就”，引导时间状语从句。[immediately after]例如： ①I had hardly opened the door when he came in. 我刚一开门，他就进来了。 ②I had not been reading for half an hour when I heard someone call my name. 【点津】hardly…when和no sooner…than的结构要注意三点：意思为“一A就B“；A句通常用完成时态；hardly 和 no sooner 位于句首时要注意部分倒装。 4. 用作从属连词，意为“倘若，如果”，表示条件。例如： ①Turn off the switch when anything goes wrong with the machine. 如果机器发生故障，就把电源关上。 ②He will be likely to recover when he is operated on. 5. 用作从属连词，意为“既然，尽管”，表示让步。例如： ①Why use metal when you can use plastic?既然能用塑料,为什么用金属? ②They kept trying when they knew it was hopeless. 尽管他们知道那件事没有希望，可是他们还在不断地努力。6. 用作并列连词，意为“在那时，届时;就在这时”，表示时间。这时主句中可以用过去进行时，过去完成时或“ was/were about to do sth. ”结构。 ①Last night I was about to go to bed when the phone rang. ②I was cooking in the kitchen when someone knocked at the door. ③He had just finished the book when supper was served. 7. 用作并列连词，意为“虽然、然而、可是”，表示转折。例如： ①He usually walks to work when he might take a bus. 虽然他可 以坐公共汽车上班，但他却常常步行上班。 ②I had only twenty dollars when I needed thirty to buy the dictionary.我需要 30 美元买那本字典，可是我只有20美元。 8. 用作并列连词，意为“而、却”，表示对比。例如： How can he say that everything is fine when it's obvious that it is not? 他怎能说一切都好呢？情况显然不是那样。 三 . when 用作代词[which time]。 when 作为代词常常位于介词之后，意为“那时，什么时候”。例如： ①Since when have you been studying Japanese? 【点津】since when 作引导词时是“介词（since）+关系代词（when）”的结构，when意为which time。since when常引导非限制性定语从句，从句应用完成时态。 ②I came here in 1949, since when I have been engaged in this work. 我1949年到这里，从那时起我就担任这项工作。 ③We came back on Tuesday, since when we have been working in the repair shop. ④We came a week ago, since when the weather has been bad. 四、when 还可用作名词，前面常常用定冠词 the。 the when 表示事 件发生的时间,常常与 the where, the how 并列使用。 He told the police the when and the how of the accident. 他告 诉警察事故发生的时间及发生的原委。 ［巩固练习］ 1. I remember ______ this used to be a quiet village. A. when B. how C. where D. what 2. Why do you want a new job _____ you've got such a good one already? A. that B. where C. which D. when 3. It was an exciting moment for these football fans this year, ______ for the first time in years their team won the World Cup. A. that B. while C. which D. when 4. We are living in an age _______ many things are done on computer. A. which B. that C. where D. when 5. The reporter said that the UFO ______ east to west when he saw it. A. was travelling B. travelled C. had been travelling D. was to travel 6. ______ got into the room _________ the telephone rang. A. He hardly had; then B. Hardly had he; when C. He had not; than D. Not had he; when 7. ---- Can I join your club, Dad? ---- You can when you _______ a bit older. A. get B. will got C. are getting D. will have got 8. I shall never forget those days ________ I lived in the countryside with the farmers, ________ had a great effect on my life. A. that; which B. when; which C. which; that D. when; who 9. The film brought the hours back to me ______ I was taken good care of in that faraway village. A. until B. that C. when D. where 10. He was about to tell me the secret ______ someone patted him on the shoulder.