Method and system for predicting highway toll loss

Abstract

This invention relates to the field of intelligent transportation technology, specifically to a method and system for predicting highway toll losses. The method includes: preprocessing historical toll data to obtain historical toll data under no traffic control influence and historical toll data under traffic control influence; based on the historical toll data under no traffic control influence, establishing toll prediction models for different vehicle types at each cross-section, and determining the predicted toll for each cross-section for different vehicle types; based on the historical toll data under traffic control influence and the predicted toll for each cross-section for different vehicle types, determining the toll revenue loss under historical traffic control influence; and based on the toll revenue loss under historical traffic control influence, establishing a toll loss prediction model to determine the predicted toll loss under future traffic control influence. This achieves a more objective and accurate prediction of toll losses under future traffic control influence.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a method and system for predicting highway toll losses. Background Technology

[0002] Highways are subject to traffic control measures due to extreme weather, accidents, construction, and routine maintenance. The most common measure is one-way closure. This causes many vehicles to change their routes or plans, such as detouring through closed sections, exiting the highway before reaching the closed section and using other roads to reach their destinations, or adopting other modes of transportation. All of these situations result in significant losses in highway toll revenue, which is a crucial source of funding for highway maintenance and improvement. Reduced toll revenue may prevent highway management departments from adequately investing in road maintenance and infrastructure upgrades, leading to deteriorating road conditions and increased safety hazards. Furthermore, reduced toll revenue also impacts the return on investment for highways, ultimately hindering the development of the highway network.

[0003] However, there is still a lack of research on predicting highway toll revenue losses under road control events. The main reason for this is the difficulty in predicting highway toll revenue losses. Specifically, predicting highway toll revenue losses involves the complex influence of multiple factors, including the degree of control measures, changes in traffic flow, road distance, toll rates, etc. Therefore, it is difficult to establish an accurate model for predicting highway toll revenue losses. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method and system for predicting highway toll losses, aiming to improve the accuracy of highway toll loss prediction.

[0005] Therefore, the present invention adopts the following technical solution: a method for predicting highway toll losses, comprising the following steps:

[0006] Step 102: Obtain the historical toll fees for different vehicle types at each section of the highway and record them as historical toll fee data;

[0007] Step 104: Preprocess the historical toll data to obtain historical toll data under no control and historical toll data under control.

[0008] Step 106: Based on the toll data under the historical uncontrolled toll period, establish toll prediction models for different vehicle types at each cross section, and determine the predicted toll for different vehicle types at each cross section.

[0009] Step 108: Based on the toll data under the influence of historical control measures and the predicted tolls for different vehicle types at each cross section, determine the toll revenue loss under the influence of historical control measures.

[0010] Step 110: Based on the toll revenue loss under the historical control measures, establish a toll loss prediction model to determine the predicted toll loss under future control measures.

[0011] The concept of this invention is as follows: First, the historical toll data for different vehicle types at various sections of a highway is preprocessed, dividing the historical toll data into toll data under the influence of historical traffic control and toll data under the influence of historical traffic control. Second, considering that different highway sections may have different characteristics and traffic flows, and that different vehicle types may exhibit different driving behaviors in the face of one-way closures on highway sections, toll prediction models are established for different vehicle types at each highway section based on the historical toll data under the influence of historical traffic control, in order to determine the predicted toll for different vehicle types at each section. Then, based on the toll data under the influence of historical traffic control and the predicted tolls for different vehicle types at each section determined by the toll prediction models, the toll revenue loss under the influence of historical traffic control can be determined. Finally, these toll revenue losses under the influence of historical traffic control are used as sample data to establish a toll loss prediction model, in order to determine the predicted toll loss under future traffic control.

[0012] By establishing toll prediction models for different vehicle types at various sections of highways based on historical toll data under no traffic control measures, this study fully considers the impact of one-way road closures on different highway sections and vehicle types. This allows for more accurate predictions of tolls at each section, providing personalized forecasts. Furthermore, by calculating toll revenue losses under historical traffic control measures and using this data as sample data to establish a toll loss prediction model, this study enables more objective and accurate predictions of toll losses under future traffic control measures. This provides a scientific basis for highway management departments to select control periods for future road closures, thereby reducing the adverse impact of road closure measures on highway toll revenue losses.

[0013] Preferably, in step 104, the historical toll data is preprocessed to obtain historical toll data under no control and historical toll data under control, including:

[0014] Obtain the control section range and control time range of historical control events;

[0015] Based on the historical toll data, the toll data for each historical control event is calculated for the control section range and the sum of tolls for each section in the two sections before and after the control section range within the control time range and the preset influence duration. The toll data for each historical control event is obtained, and the toll data for all historical control events constitute the toll data under the influence of historical control.

[0016] The toll data in the historical toll data that is not affected by the historical control measures shall be recorded as the toll data under the historical uncontrolled conditions.

[0017] Preferably, in step 106, based on the historical toll data under no-control conditions, toll prediction models are established for each cross-section for different vehicle types, and the predicted toll for each cross-section for different vehicle types is determined, including:

[0018] The Prophet algorithm is used to fit the historical toll data under no control conditions to obtain a time series combination model of the toll trend function, toll periodicity function, and toll impact function of holidays for each cross section for different vehicle types. This model is denoted as the toll prediction model for each cross section for different vehicle types. The input of the toll prediction model for each cross section for different vehicle types is the control time, and the output is the predicted toll for each cross section for different vehicle types.

[0019] Preferably, in step 106, based on the historical toll data under no-control conditions, toll prediction models are established for each cross-section for different vehicle types, and the predicted toll for each cross-section for different vehicle types is determined, including:

[0020] Step 302: Divide the toll data under the historical uncontrolled toll system into a training set and a validation set;

[0021] Step 304: Preset several tuning parameters and the value range of each tuning parameter, and use the grid tuning method to arrange and combine the values ​​of the several tuning parameters to obtain several sets of parameter combinations;

[0022] Step 306: Using a set of parameter combinations, the Prophet algorithm is used to fit the training set to obtain the time series combination model of the toll trend function, toll periodic function and toll influence function of holidays for each cross section for different traffic types. This model is denoted as the toll prediction model for each cross section for different traffic types.

[0023] Step 308: Use the toll prediction model to predict tolls on the validation set, obtain the corresponding prediction results, and determine the error of the toll prediction model corresponding to the current parameter combination based on the validation set and the corresponding prediction results.

[0024] Step 310: Use the next set of parameter combinations and repeat steps 306 to 308 until all parameter combinations have been traversed, and obtain the error of the toll prediction model corresponding to all parameter combinations.

[0025] Step 312: Select the parameter combination with the smallest error value among the errors of the toll prediction models corresponding to all parameter combinations as the preferred parameter combination, and use the toll prediction model corresponding to the preferred parameter combination as the final toll prediction model for each cross section for different vehicle types.

[0026] Preferably, in step 108, based on the toll data under the influence of historical traffic control and the predicted tolls for different vehicle types at each cross-section, the toll revenue loss under the influence of historical traffic control is determined, including:

[0027] Input the control time corresponding to each historical control event in the toll data under the influence of the historical control into the toll prediction model for different vehicle types of the corresponding control section, and obtain the predicted toll for different vehicle types of the control section corresponding to each historical control event.

[0028] Calculate the sum of the predicted toll fees for different vehicle types at the control section corresponding to each historical control event to obtain the predicted toll fee data for each historical control event;

[0029] The difference between the predicted toll data for each historical control event and the toll data for the corresponding historical control event is calculated to obtain the toll revenue loss for each historical control event.

[0030] All the toll revenue losses from the aforementioned historical control events constitute toll revenue losses under the influence of historical control measures.

[0031] Preferably, in step 110, based on the toll revenue loss under the historical control measures, a toll loss prediction model is established to determine the predicted toll loss under future control measures, including:

[0032] The toll revenue loss caused by the historical control measures is divided into an original training set and a validation set.

[0033] Select some historical toll data under the influence of no control, and based on the toll data under the influence of no control and the predicted toll for different vehicle types at each section, determine the toll revenue loss under the influence of no control in some historical periods.

[0034] The toll revenue loss due to the lack of historical control measures is added to the original training set, and the newly constructed training set is designated as the first training set.

[0035] The first training set is augmented using the Bootstrap sampling method, and the augmented training set is designated as the second training set.

[0036] Based on the second training set, a toll loss prediction model is established to determine the predicted toll loss under the influence of future control measures.

[0037] Preferably, based on the second training set, a toll loss prediction model is established to determine the predicted toll loss under future control measures, including:

[0038] Feature extraction is performed on the second training set to obtain several feature data.

[0039] The feature data are ranked by importance, and the top-ranked feature data are selected as the input features of the second training set.

[0040] A toll loss prediction model is established, and the toll loss prediction model is trained using the input features of the second training set to obtain the trained toll loss prediction model. The trained toll loss prediction model is used to determine the predicted toll loss under the influence of future control measures.

[0041] Preferably, the establishment of the toll loss prediction model includes:

[0042] Construct a lightGBM regression model, train the lightGBM regression model using the second training set, and denote the trained lightGBM regression model as the first prediction model.

[0043] The first prediction model is used to predict toll fees on the validation set, and the corresponding prediction results are denoted as the first prediction results.

[0044] Based on the validation set and the first prediction result, the root mean square error of the first prediction model in the validation set is determined and denoted as the first root mean square error.

[0045] Construct an SVR regression model, train the SVR regression model using the second training set, and denote the trained SVR regression model as the second prediction model.

[0046] The second prediction model is used to predict toll fees on the validation set, and the corresponding prediction results are denoted as the second prediction results.

[0047] Based on the validation set and the second prediction results, the root mean square error of the second prediction model in the validation set is determined and denoted as the second root mean square error.

[0048] Based on the first root mean square error and the second root mean square error, a combined model of the first prediction model and the second prediction model is obtained, denoted as the toll loss prediction model. The input of the toll loss prediction model is the input features extracted from future control events, and the output is the predicted toll loss under the influence of future control.

[0049] Preferably, based on the first root mean square error and the second root mean square error, a combined model of the first prediction model and the second prediction model is obtained, denoted as the toll loss prediction model, specifically as follows:

[0050]

[0051] Among them, P Y R1 represents the predicted toll loss output by the toll loss prediction model, R2 represents the first root mean square error, P1 represents the prediction result obtained by inputting the input features extracted from future control events into the first prediction model, and P2 represents the prediction result obtained by inputting the input features extracted from future control events into the second prediction model.

[0052] A system for predicting highway toll losses includes:

[0053] The data acquisition module is used to acquire the historical toll fees for different vehicle types at various sections of the highway, and record them as historical toll fee data.

[0054] The data preprocessing module is used to preprocess the historical toll data to obtain historical toll data under no control and historical toll data under control.

[0055] The toll prediction module is used to establish toll prediction models for different vehicle types at each cross section based on the historical toll data under no-control conditions, and to determine the predicted toll for different vehicle types at each cross section.

[0056] The sample determination module is used to determine the loss of toll revenue under the influence of historical control measures based on the toll data under the influence of historical control measures and the predicted toll for different vehicle types at each cross section.

[0057] The toll loss prediction module is used to establish a toll loss prediction model based on the toll revenue loss under the influence of historical control measures, and to determine the predicted toll loss under the influence of future control measures.

[0058] The beneficial technical effects of this invention include at least the following: It employs a method and system for predicting highway toll losses. By establishing toll prediction models for different vehicle types at each section of the highway based on historical toll data under no-control conditions, it fully considers the impact of different highway sections and different vehicle types under one-way road closure control measures. This allows for more accurate prediction of toll conditions at each section, providing personalized prediction results. Simultaneously, by calculating toll revenue losses under historical control conditions and using this as sample data to establish a toll loss prediction model, it enables a more objective and accurate prediction of toll losses under future control conditions. This provides a scientific basis for highway management departments to select control periods when implementing future road closures, thereby reducing the adverse impact of road closure control measures on highway toll revenue losses.

[0059] Other features and advantages of the present invention will be disclosed in detail in the following detailed description and accompanying drawings. Attached Figure Description

[0060] The invention will be further described below with reference to the accompanying drawings:

[0061] Figure 1 This is a flowchart of a method for predicting highway toll losses according to an embodiment of the present invention.

[0062] Figure 2 This is an example diagram illustrating the time series decomposition factors of the toll prediction model in an embodiment of the present invention.

[0063] Figure 3 This is a flowchart illustrating a method for establishing toll prediction models for different vehicle types at various cross-sections based on historical toll data under no-control conditions, as per an embodiment of the present invention.

[0064] Figure 4 This is a schematic diagram comparing the predicted results and actual values ​​of historical toll fees for the Tiantai-Baihe section according to an embodiment of the present invention.

[0065] Figure 5 This is a schematic diagram comparing the predicted results and actual values ​​of historical toll fees for the Shengzhou-Sanjie section according to an embodiment of the present invention.

[0066] Figure 6 This is a schematic diagram illustrating the method for establishing a toll loss prediction model according to an embodiment of the present invention.

[0067] Figure 7 This is a schematic diagram of the structure of a highway toll loss prediction system according to an embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of the present invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of the present invention.

[0069] In the following description, terms such as “inner,” “outer,” “upper,” “lower,” “left,” and “right” are used only to indicate orientation or positional relationship for the convenience of describing the embodiments and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0070] This application provides a method for predicting highway toll losses; please refer to the appendix. Figure 1 This includes the following steps:

[0071] Step 102: Obtain the historical toll fees for different vehicle types at each section of the highway and record them as historical toll fee data.

[0072] The vehicle types include, but are not limited to, passenger cars, freight cars, and special-purpose vehicles. Preferably, since the number of special-purpose vehicles actually traveling on highways is much smaller than that of passenger cars and freight cars, this embodiment only considers historical toll data for passenger cars and freight cars at various sections of the highway. Specifically, under the one-way closure control measures on highway sections, the driving behaviors of passenger cars and freight cars differ significantly. Passenger cars generally choose to detour around the closed section to find alternative routes or exits. This is because passenger cars are usually more flexible and adaptable, and can more easily choose other routes or exits to avoid the closed section. However, when faced with one-way closure control measures, freight cars generally choose to wait for the closure to be lifted. The main reason is that drivers consider the balance of costs and the daytime traffic restrictions on freight cars on main roads in many cities. Specifically, when there is still a long distance to the destination, it is more troublesome and costly for freight cars to detour around the highway closure via ordinary urban roads, so drivers tend to wait for the closure to be lifted when faced with one-way closure control measures.

[0073] For example, in this embodiment, the historical toll fees for passenger cars and freight cars at each section of the highway are obtained as follows: from the highway section traffic flow big data platform, the toll fee data of each section is counted at hourly intervals, the hourly section toll fees are divided into two groups: passenger cars and freight cars, and the sum of the toll fees for each group is calculated separately.

[0074] Step 104: Preprocess the historical toll data to obtain historical toll data under no control and historical toll data under control.

[0075] For example, the implementation method for preprocessing historical toll data is as follows: obtain the control section range and control time range of historical control events; based on the historical toll data, calculate the sum of tolls for each section within the control section range within the control time range for each historical control event, and obtain the toll data for each historical control event. The toll data of all historical control events constitute the toll data under the influence of historical control; then, the toll data other than the toll data under the influence of historical control is recorded as the toll data under the influence of historical control.

[0076] Step 106: Based on historical toll data under no-control conditions, establish toll prediction models for different vehicle types at each cross-section, and determine the predicted toll for different vehicle types at each cross-section.

[0077] Step 108: Based on the toll data under the influence of historical control measures and the predicted tolls for different vehicle types at each cross-section, determine the toll revenue loss under the influence of historical control measures.

[0078] Optionally, in this embodiment, the method for determining the toll revenue loss under the influence of historical control measures can be to compare the difference between the toll data under the influence of historical control measures and the predicted tolls for different vehicle types at each cross-section corresponding to the period without the influence of historical control measures, in order to determine the toll revenue loss under the influence of historical control measures. Alternatively, statistical methods can be used to analyze the toll data under the influence of historical control measures and the predicted tolls for different vehicle types at each cross-section, such as calculating the average value, variance, correlation, and other indicators, to obtain the toll revenue loss under the influence of historical control measures.

[0079] Step 110: Based on the toll revenue loss under historical control measures, establish a toll loss prediction model to determine the predicted toll loss under future control measures.

[0080] The concept of this embodiment is as follows: First, the historical toll data for different vehicle types at various sections of the highway is preprocessed, dividing the historical toll data into toll data under the influence of historical traffic control and toll data under the influence of historical traffic control. Second, considering that different highway sections may have different characteristics and traffic flows, and that different vehicle types may exhibit different driving behaviors in the face of one-way closures on highway sections, toll prediction models are established for different vehicle types at each highway section based on the historical toll data under the influence of historical traffic control, in order to determine the predicted toll for different vehicle types at each section. Then, based on the toll data under the influence of historical traffic control and the predicted tolls for different vehicle types at each section determined by the toll prediction models, the toll revenue loss under the influence of historical traffic control can be determined. Finally, these toll revenue losses under the influence of historical traffic control are used as sample data to establish a toll loss prediction model, in order to determine the predicted toll loss under future traffic control.

[0081] By establishing toll prediction models for different vehicle types at various sections of highways based on historical toll data under no traffic control measures, this study fully considers the impact of one-way road closures on different highway sections and vehicle types. This allows for more accurate predictions of toll rates at each section, providing personalized forecasts. Furthermore, by calculating toll revenue losses under historical traffic control measures and using this data as sample data to establish a toll loss prediction model, this study enables more objective and accurate predictions of toll losses under future traffic control measures. This provides a scientific basis for highway management departments to select control periods for future road closures, thereby reducing the adverse impact of road closure measures on highway toll revenue losses.

[0082] In one embodiment of this specification, step 104 involves preprocessing historical toll data to obtain historical toll data under no control and historical toll data under control, including:

[0083] Obtain the control section range and control time range of historical control events;

[0084] Based on historical toll data, the toll data for each historical control event is calculated by summing the tolls for each section within the control section and the two sections before and after the control section within the control time range and the preset duration of influence. The toll data for each historical control event is obtained, and the toll data for all historical control events constitute the toll data under the influence of historical control.

[0085] Toll data other than those affected by historical traffic control measures will be recorded as toll data without the impact of historical traffic control measures.

[0086] The control section range and control time range of historical control events can be obtained from the historical control event records of the regional traffic information platform or application. Preferably, the duration of the impact is within 6 hours after the control ends.

[0087] Since one-way road closures not only affect traffic flow at the designated closure section and during the specified time period, but also often increase or decrease traffic flow at adjacent sections and in the period following the closure, thus reducing toll revenue, this embodiment comprehensively considers the controlled section range, the two sections before and after the controlled section range, the controlled time range, and the toll within the preset impact duration. This makes the toll data under historical control effects more comprehensive and detailed, providing a more accurate foundation for the establishment of subsequent prediction models.

[0088] Furthermore, since toll fees are subject to some abnormal factors (such as weather, other roads, etc.), the 3σ criterion can be used to remove outliers based on this embodiment, and the effective value (μ) range of toll fee data can be controlled within (μ-3σ, μ+3σ).

[0089] In one embodiment of this specification, step 106 involves establishing toll prediction models for different vehicle types at each cross-section based on historical toll data under no-control conditions, and determining the predicted toll for each cross-section for different vehicle types, including:

[0090] The Prophet algorithm is used to fit the toll data under historical conditions without traffic control to obtain a time series combination model of the toll trend function, toll periodicity function, and toll impact function of holidays for different vehicle types at each cross section. This model is denoted as the toll prediction model for different vehicle types at each cross section. The input of the toll prediction model for different vehicle types at each cross section is the control time, and the output is the predicted toll for different vehicle types at each cross section.

[0091] Among them, the Prophet algorithm is a time series forecasting algorithm that is based on an additive model and decomposes time series data into a combination of trend, seasonality and holiday effects.

[0092] For example, please refer to the appendix. Figure 2 The Prophet algorithm was used to organize the toll data under the historical lack of control into a format suitable for the Prophet algorithm input, namely, toll time series data, including the toll values ​​for different vehicle types at each cross-section for hourly intervals. The toll time series data under the historical lack of control was decomposed according to the following factors:

[0093] Fee(t)=g(t)+s(t)+h(t)+∈

[0094] Where g(t) represents the trend function of toll fees, s(t) represents the periodic function of toll fees, h(t) represents the function of toll fees affected by holidays, and ∈ represents the preset fluctuation error. For the trend term g(t), a piecewise logistic regression function is used, requiring pre-defined upper and lower bounds. For the periodic function s(t), a Fourier series is used for simulation, decomposing it into:

[0095]

[0096] When the cycle is based on years, P = 365.25, N = 10; when the cycle is based on weeks, P = 7, N = 3; a n and b n These are the correlation coefficients.

[0097] This embodiment uses the Prophet algorithm to detect and model the seasonal and holiday effects in historical toll data under no-control conditions. It establishes toll prediction models for different vehicle types at each cross-section to determine the predicted toll for different vehicle types at each cross-section. It fully considers the impact of different highway cross-sections and different vehicle types under the control measures of one-way closure on highway sections, and can more accurately predict the toll situation at each cross-section, providing personalized prediction results.

[0098] In one embodiment of this specification, in step 106, based on historical toll data under no-control conditions, toll prediction models are established for each cross-section for different vehicle types, and the predicted toll for each cross-section for different vehicle types is determined. Please refer to the appendix. Figure 3 ,include:

[0099] Step 302: Divide the toll data under the historical uncontrolled toll system into a training set and a validation set;

[0100] Step 304: Preset several tuning parameters and the value range of each tuning parameter, and use the grid tuning method to arrange and combine the values ​​of the several tuning parameters to obtain several sets of parameter combinations.

[0101] For example, step 304 is implemented as follows: Select the main tuning parameters of the toll prediction model and the value range of each tuning parameter as shown in Table 1:

[0102] Parameter tuning Range of values changepoint_range [i / 10 for i in range(3,10)] seasonality_prior_scale [0.05,0.1,0.5,1,5,10,15] holidays_prior_scale [0.05,0.1,0.5,1,5,10,15]

[0103] Table 1. Main tuning parameters and their value ranges

[0104] Here, `changepoint_range` specifies the range of trend change points; the example change point range is 0.3–1. `seasonality_prior_scale` controls the flexibility of the periodicity function; a higher value indicates that the periodicity is more prone to change. `holidays_prior_scale` controls the impact of the toll-rate-affected-by-holidays function on the toll-rate-prediction model. In fact, the correlation between tolls and holidays is significant; passenger vehicles enjoy free highway passage on certain special holidays. Therefore, a larger value should be assigned to `holidays_prior_scale` to increase the impact of holidays on the model. Next, a grid-based hyperparameter tuning method is used to permutate and combine the values ​​of several hyperparameters, exhaustively exploring all possible combinations to find the optimal model performance.

[0105] Step 306: Using a set of parameters, the Prophet algorithm is used to fit the training set to obtain the time series combination model of the toll trend function, toll periodic function and toll influence function of holidays for each cross section for different traffic types. This model is denoted as the toll prediction model for each cross section for different traffic types.

[0106] Step 308: Use the toll prediction model to predict tolls on the validation set, obtain the corresponding prediction results, and determine the error of the toll prediction model corresponding to the current parameter combination based on the validation set and the corresponding prediction results.

[0107] For example, under the condition of no historical toll control, toll prediction models were established for 60 sections of the Shangsan Expressway (the expressway section from Guzhu Interchange 1002 to Wutun Interchange 1201 in Shaoxing City) for different vehicle types, based on historical toll data for both passenger and freight vehicles. These toll prediction models were then used to predict tolls on the validation set (Tiantai-Baihe section and Shengzhou-Sanjie section), yielding the corresponding prediction results. A comparison diagram of the predicted historical tolls for the Tiantai-Baihe section (dashed curve in the figure) and the actual values ​​(solid curve in the figure) is attached. Figure 4 Please refer to the attached diagram for a comparison of the predicted historical toll rates for the Shengzhou-Sanjie section (dashed curve in the figure) and the actual rates (solid curve in the figure). Figure 5 As can be seen, the toll prediction models provided in this embodiment have high prediction accuracy for different traffic types at each cross-section.

[0108] Specifically, based on the validation set and the corresponding prediction results, the error of the toll prediction model corresponding to the current parameter combination is determined. This is done by inputting the time data of toll data for different vehicle types at each cross-section in the validation set into the toll prediction model for each cross-section for different vehicle types to obtain the predicted toll value. The absolute difference between this predicted toll value and the actual toll value for each vehicle type at each cross-section in the validation set is calculated, which is the error of the toll prediction model for each vehicle type at each cross-section. The average of the toll prediction model errors for each vehicle type at each cross-section is then calculated, which is the error of the toll prediction model corresponding to the current parameter combination.

[0109] On the other hand, in this embodiment, to avoid the potential randomness of errors on the validation set obtained after only one training iteration for a parameter combination, this embodiment can also employ sliding time series cross-validation to determine the error of the toll prediction model corresponding to the current parameter combination. Sliding time series cross-validation is a cross-validation technique particularly suitable for modeling and predicting time series data. It differs from traditional cross-validation methods (such as K-fold cross-validation) because the special nature of time series data requires consideration of the temporal order of the data. The basic concept of sliding time series cross-validation is to create multiple combinations of training and test sets by sliding a fixed-size time window.

[0110] Step 310: Use the next set of parameter combinations and repeat steps 306 to 308 until all parameter combinations have been traversed, and obtain the error of the toll prediction model corresponding to all parameter combinations.

[0111] Step 312: Select the parameter combination with the smallest error value among all parameter combinations for the toll prediction model, and use the toll prediction model corresponding to the preferred parameter combination as the final toll prediction model for each cross section for different vehicle types.

[0112] This embodiment, by optimizing the combination of parameter values, helps to obtain a more accurate toll prediction model for different traffic types at each cross-section.

[0113] In one embodiment of this specification, step 108, based on toll data under the influence of historical traffic control and the predicted tolls for different vehicle types at each cross-section, determines the toll revenue loss under the influence of historical traffic control, including:

[0114] Input the control time corresponding to each historical control event in the toll data under the influence of historical control into the toll prediction model for different vehicle types of the corresponding control section, and obtain the predicted toll for different vehicle types of the control section corresponding to each historical control event.

[0115] Calculate the sum of the predicted toll fees for different vehicle types at the control section corresponding to each historical control event to obtain the predicted toll fee data for each historical control event;

[0116] The difference between the predicted toll data for each historical control event and the toll data for the corresponding historical control event is calculated to obtain the toll revenue loss for each historical control event.

[0117] The loss of toll revenue from all historical control events constitutes the loss of toll revenue under the influence of historical control measures.

[0118] It is understandable that the actual value of toll revenue loss in historical control is unknowable in practice. Therefore, this embodiment creatively uses the difference between the predicted toll data of each historical control event and the toll data of the corresponding historical control event to restore the actual toll revenue loss of each historical control event, which has high practical value.

[0119] On the other hand, in this embodiment, the control time corresponding to each historical control event in the toll data under the influence of historical control includes the preset influence duration, and the corresponding control section of each historical control event includes two sections before and after the control section range.

[0120] In one embodiment of this specification, step 110 involves establishing a toll loss prediction model based on historical toll revenue losses under the influence of toll control measures, and determining the predicted toll loss under future toll control measures, including:

[0121] The toll revenue loss due to historical control measures is divided into the original training set and the validation set.

[0122] Select some historical toll data under the influence of no control, and based on the historical toll data under the influence of no control and the predicted toll for different vehicle types at each section, determine the toll revenue loss under the influence of no control in some historical periods.

[0123] The toll revenue losses due to historical lack of control were added to the original training set, and the newly constructed training set was designated as the first training set.

[0124] The Bootstrap sampling method was used to augment the first training set, and the augmented training set was designated as the second training set.

[0125] Based on the second training set, a toll loss prediction model is established to determine the predicted toll loss under the influence of future traffic control measures.

[0126] It is understandable that the sample size of toll data under historical control is much smaller than that of toll data without historical control. Therefore, this embodiment increases the sample richness of the training set by adding some samples of toll revenue loss under historical uncontrolled conditions to the original training set, which helps to improve the model's generalization ability and accuracy. At the same time, the training set is expanded by using the Bootstrap sampling method, which increases the sample size of the training set, helps to reduce the variance in the model training process, improves the model's stability and reliability, and thus better addresses the technical problem of the small sample size of toll data under historical control, making the model more representative and predictive.

[0127] In one embodiment of this specification, a toll loss prediction model is established based on a second training set to determine the predicted toll loss under future traffic control measures, including:

[0128] Feature extraction is performed on the second training set to obtain several feature data.

[0129] The feature data are ranked by importance, and the top-ranked features are selected as the input features of the second training set.

[0130] A toll loss prediction model is established. The toll loss prediction model is trained using the input features of the second training set to obtain the trained toll loss prediction model. The trained toll loss prediction model is used to determine the predicted toll loss under the influence of future control measures.

[0131] For example, feature extraction is performed on the second training set, resulting in the following feature data:

[0132] 1. Number of sealed sections;

[0133] 2. The sum of passenger vehicle tolls for the closed sections during the control period is predicted using Model A;

[0134] 3. During the period when the control measures are implemented, the sum of toll fees for trucks and special-purpose vehicles at the closed section is predicted using Model A;

[0135] 4. The distance between the first section of a one-way road closure and the starting point of the Shangsan Expressway;

[0136] 5. The distance between the last section of the one-way road closure and the end point of the Shangsan Expressway;

[0137] 6. Number of days the road closure will last;

[0138] 7. Road closure duration;

[0139] 8. Did any similar road closures occur within the two days prior to the incident control period?

[0140] 9. The number of hubs within the event control scope;

[0141] 10. Number of lanes in the closed section;

[0142] 11. Number of tunnels in the closed section.

[0143] Alternatively, the methods for ranking feature data by importance include, but are not limited to, the following:

[0144] 1. Random Forest: Random forest is an ensemble learning algorithm that evaluates the importance of features by constructing multiple decision trees and ensembling their predictions. The importance of a feature in a random forest can be measured using feature importance scores in the decision trees (such as Gini importance or average reduced impurity).

[0145] 2. Gradient Boosting Tree: Gradient boosting trees are also an ensemble learning algorithm that improves model performance by progressively training a decision tree. Gradient boosting trees calculate the contribution of each feature to model performance, and these contributions can be used to evaluate the importance of the features.

[0146] 3. Model-based Feature Selection: These methods use machine learning models with built-in feature selection capabilities to evaluate the importance of features. For example, using models such as logistic regression, support vector machines, or neural networks, the importance of features can be determined by the magnitude or weights of the model parameters.

[0147] 4. Univariate Feature Selection: These methods assess the importance of features by calculating a statistical index (such as chi-square test, F-test, or mutual information) for each feature. The features are then ranked according to their scores, and the feature with the highest score is selected.

[0148] This embodiment sorts the feature data by importance, selects the top-ranked feature data as the input features of the training set, and removes some weakly correlated or redundant features to find a better feature subset as the training set for the toll loss prediction model, which helps to accelerate the model training convergence.

[0149] In one embodiment of this specification, a toll loss prediction model is established; please refer to the appendix. Figure 6 ,include:

[0150] Step 602: Construct a lightGBM regression model. Train the lightGBM regression model using the second training set, and denote the trained lightGBM regression model as the first prediction model.

[0151] LightGBM (Light Gradient Boosting Machine) is a framework for implementing the GBDT algorithm. It supports highly efficient parallel training and has advantages such as faster training speed, lower memory consumption, better accuracy, and support for distributed processing to quickly handle massive amounts of data.

[0152] Step 604: Use the first prediction model to predict toll fees on the validation set and obtain the corresponding prediction results, which are denoted as the first prediction results.

[0153] Step 606: Based on the validation set and the first prediction result, determine the root mean square error of the first prediction model in the validation set, denoted as the first root mean square error.

[0154] Step 608: Construct the SVR regression model. Train the SVR regression model using the second training set, and denote the trained SVR regression model as the second prediction model.

[0155] SVR (Support Vector Regression) is an application of Support Vector Machines (SVM) in regression problems. Similar to the SVM classification model, SVR is also a non-probabilistic algorithm. It maps the data to a high-dimensional space using a kernel function and finds the optimal hyperplane in this space that maximizes the margin between the hyperplane and the training data, thus obtaining the regression model. SVR is a widely used basic regression model in industry, with advantages such as stable calculation results and strong robustness to outliers.

[0156] Step 610: Use the second prediction model to predict toll fees on the validation set and obtain the corresponding prediction results, which are denoted as the second prediction results.

[0157] Step 612: Based on the validation set and the second prediction result, determine the root mean square error of the second prediction model in the validation set, denoted as the second root mean square error.

[0158] Step 614: Based on the first root mean square error and the second root mean square error, a combined model of the first prediction model and the second prediction model is obtained, denoted as the toll loss prediction model. The input of the toll loss prediction model is the input features extracted from future control events, and the output is the predicted toll loss under the influence of future control.

[0159] Optionally, the implementation methods of this embodiment for obtaining a combined prediction model based on the first root mean square error and the second root mean square error include, but are not limited to, the following:

[0160] 1. Weighted Average Method: Based on the first and second root mean square errors, different weights can be assigned to the two models, and their prediction results are then weighted and averaged to obtain the final combined prediction result. The weights can be determined based on model performance, performance on the validation set, or the experience of domain experts.

[0161] 2. Stacking: Stacking is an ensemble learning method that takes the prediction results of the first and second prediction models as input, and then uses a meta-model (usually a linear regression model or other machine learning model) to retrain these prediction results to obtain the final combined model.

[0162] 3. Feature fusion method: The intermediate layer features of the first and second prediction models are fused together and then input into a new model for training, thereby realizing the construction of a combined model.

[0163] This embodiment calculates the first root mean square error and the second root mean square error, and constructs a LightGBM_SVR fusion algorithm regression model as a toll loss prediction model based on these errors. This improves the robustness and generalization ability of the model, while giving full play to the advantages of the lightGBM regression model and the SVR regression model in terms of training efficiency, accuracy and stability. This further improves the accuracy and stability of the prediction, making the final toll loss prediction model more credible and practical.

[0164] In one embodiment of this specification, a combined model of the first prediction model and the second prediction model is obtained based on the first root mean square error and the second root mean square error, denoted as the toll loss prediction model, specifically as follows:

[0165]

[0166] Among them, P Y R1 represents the predicted toll loss output by the toll loss prediction model, R2 represents the first root mean square error, P1 represents the prediction result obtained by inputting the input features extracted from future control events into the first prediction model, and P2 represents the prediction result obtained by inputting the input features extracted from future control events into the second prediction model.

[0167] On the other hand, embodiments of this application also provide a highway toll loss prediction system similar to the foregoing technical concept; please refer to the appendix. Figure 7 ,include:

[0168] Data acquisition module 1 is used to acquire the historical toll fees for different vehicle types at various sections of the highway, and record them as historical toll fee data;

[0169] Data preprocessing module 2 is used to preprocess historical toll data to obtain historical toll data under no control and historical toll data under control.

[0170] Toll prediction module 3 is used to establish toll prediction models for different vehicle types at each cross section based on historical toll data under no control conditions, and to determine the predicted toll for different vehicle types at each cross section.

[0171] Sample determination module 4 is used to determine the loss of toll revenue under the influence of historical control measures based on toll data under the influence of historical control measures and the predicted toll for different vehicle types at each cross section.

[0172] The toll loss prediction module 5 is used to establish a toll loss prediction model based on the toll revenue loss under the influence of historical control measures, and to determine the predicted toll loss under the influence of future control measures.

[0173] The above description is merely a preferred embodiment disclosed in this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of protection involved in this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0174] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

Claims

1. A method for predicting highway toll losses, characterized in that, Includes the following steps: Step 102: Obtain the historical toll fees for different vehicle types at each section of the highway and record them as historical toll fee data; Step 104: Preprocess the historical toll data to obtain historical toll data under no control and historical toll data under control. Step 106: Based on the toll data under the historical uncontrolled toll period, establish toll prediction models for different vehicle types at each cross section, and determine the predicted toll for different vehicle types at each cross section. Step 108: Based on the toll data under the influence of historical control measures and the predicted tolls for different vehicle types at each cross section, determine the toll revenue loss under the influence of historical control measures. Step 110: Based on the toll revenue loss under the historical control measures, establish a toll loss prediction model to determine the predicted toll loss under future control measures. In step 108, based on the toll data under the influence of historical traffic control and the predicted tolls for different vehicle types at each cross-section, the toll revenue loss under the influence of historical traffic control is determined, including: Input the control time corresponding to each historical control event in the toll data under the influence of historical control into the toll prediction model for different vehicle types of the corresponding control section, and obtain the predicted toll for different vehicle types of the control section corresponding to each historical control event. Calculate the sum of the predicted toll fees for different vehicle types at the control section corresponding to each historical control event to obtain the predicted toll fee data for each historical control event; The difference between the predicted toll data for each historical control event and the toll data for the corresponding historical control event is calculated to obtain the toll revenue loss for each historical control event. All the toll revenue losses from the aforementioned historical control events constitute toll revenue losses under the influence of historical control measures.

2. The method for predicting highway toll loss as described in claim 1, characterized in that, In step 104, the historical toll data is preprocessed to obtain historical toll data under no control and historical toll data under control, including: Obtain the control section range and control time range of historical control events; Based on the historical toll data, the toll data for each historical control event is calculated for the control section range and the sum of tolls for each section in the two sections before and after the control section range within the control time range and the preset influence duration. The toll data for each historical control event is obtained, and the toll data for all historical control events constitute the toll data under the influence of historical control. The toll data in the historical toll data that is not affected by the historical control measures shall be recorded as the toll data under the historical uncontrolled conditions.

3. The method for predicting highway toll loss as described in claim 1, characterized in that, In step 106, based on the historical toll data under no-control conditions, toll prediction models are established for each cross-section for different vehicle types to determine the predicted toll for each cross-section for different vehicle types, including: The Prophet algorithm is used to fit the historical toll data under no control conditions to obtain a time series combination model of the toll trend function, toll periodicity function, and toll impact function of holidays for each cross section for different vehicle types. This model is denoted as the toll prediction model for each cross section for different vehicle types. The input of the toll prediction model for each cross section for different vehicle types is the control time, and the output is the predicted toll for each cross section for different vehicle types.

4. The method for predicting highway toll loss as described in claim 3, characterized in that, In step 106, based on the historical toll data under no-control conditions, toll prediction models are established for each cross-section for different vehicle types to determine the predicted toll for each cross-section for different vehicle types, including: Step 302: Divide the toll data under the historical uncontrolled toll system into a training set and a validation set; Step 304: Preset several tuning parameters and the value range of each tuning parameter, and use the grid tuning method to arrange and combine the values ​​of the several tuning parameters to obtain several sets of parameter combinations; Step 306: Using a set of parameter combinations, the Prophet algorithm is used to fit the training set to obtain the time series combination model of the toll trend function, toll periodic function and toll influence function of holidays for each cross section for different traffic types. This model is denoted as the toll prediction model for each cross section for different traffic types. Step 308: Use the toll prediction model to predict tolls on the validation set, obtain the corresponding prediction results, and determine the error of the toll prediction model corresponding to the current parameter combination based on the validation set and the corresponding prediction results. Step 310: Use the next set of parameter combinations and repeat steps 306 to 308 until all parameter combinations have been traversed, and obtain the error of the toll prediction model corresponding to all parameter combinations. Step 312: Select the parameter combination with the smallest error value among the errors of the toll prediction models corresponding to all parameter combinations as the preferred parameter combination, and use the toll prediction model corresponding to the preferred parameter combination as the final toll prediction model for each cross section for different vehicle types.

5. The method for predicting highway toll losses as described in claim 1, characterized in that, In step 110, based on the toll revenue loss under the historical control measures, a toll loss prediction model is established to determine the predicted toll loss under future control measures, including: The toll revenue loss caused by the historical control measures is divided into an original training set and a validation set. Select some historical toll data under the influence of no control, and based on the toll data under the influence of no control and the predicted toll for different vehicle types at each section, determine the toll revenue loss under the influence of no control in some historical periods. The toll revenue loss due to the lack of historical control measures is added to the original training set, and the newly constructed training set is designated as the first training set. The first training set is augmented using the Bootstrap sampling method, and the augmented training set is designated as the second training set. Based on the second training set, a toll loss prediction model is established to determine the predicted toll loss under the influence of future control measures.

6. The method for predicting highway toll losses as described in claim 5, characterized in that, Based on the second training set, a toll loss prediction model is established to determine the predicted toll loss under future traffic control measures, including: Feature extraction is performed on the second training set to obtain several feature data. The feature data are ranked by importance, and the top-ranked feature data are selected as the input features of the second training set. A toll loss prediction model is established, and the toll loss prediction model is trained using the input features of the second training set to obtain the trained toll loss prediction model. The trained toll loss prediction model is used to determine the predicted toll loss under the influence of future control measures.

7. The method for predicting highway toll losses as described in claim 5, characterized in that, The establishment of the toll loss prediction model includes: Construct a lightGBM regression model, train the lightGBM regression model using the second training set, and denote the trained lightGBM regression model as the first prediction model. The first prediction model is used to predict toll fees on the validation set, and the corresponding prediction results are denoted as the first prediction results. Based on the validation set and the first prediction result, the root mean square error of the first prediction model in the validation set is determined and denoted as the first root mean square error. Construct an SVR regression model, train the SVR regression model using the second training set, and denote the trained SVR regression model as the second prediction model. The second prediction model is used to predict toll fees on the validation set, and the corresponding prediction results are denoted as the second prediction results. Based on the validation set and the second prediction results, the root mean square error of the second prediction model in the validation set is determined and denoted as the second root mean square error. Based on the first root mean square error and the second root mean square error, a combined model of the first prediction model and the second prediction model is obtained, denoted as the toll loss prediction model. The input of the toll loss prediction model is the input features extracted from future control events, and the output is the predicted toll loss under the influence of future control.

8. The method for predicting highway toll loss as described in claim 7, characterized in that, Based on the first root mean square error and the second root mean square error, a combined model of the first prediction model and the second prediction model is obtained, denoted as the toll loss prediction model, which is as follows: ， in, This represents the predicted toll loss output by the toll loss prediction model. This represents the first root mean square error. This represents the second root mean square error. This represents the prediction result obtained by inputting the extracted input features of future control events into the first prediction model. This represents the prediction result obtained by inputting the input features extracted from future control events into the second prediction model.

9. A system for predicting highway toll losses, characterized in that, include: The data acquisition module is used to acquire the historical toll fees for different vehicle types at various sections of the highway, and record them as historical toll fee data. The data preprocessing module is used to preprocess the historical toll data to obtain historical toll data under no control and historical toll data under control. The toll prediction module is used to establish toll prediction models for different vehicle types at each cross section based on the historical toll data under no-control conditions, and to determine the predicted toll for different vehicle types at each cross section. The sample determination module is used to determine the loss of toll revenue under the influence of historical control measures based on the toll data under the influence of historical control measures and the predicted toll for different vehicle types at each cross section. The toll loss prediction module is used to establish a toll loss prediction model based on the toll revenue loss under the influence of historical control measures, and to determine the predicted toll loss under the influence of future control measures. The sample determination module is used to perform the following steps: Input the control time corresponding to each historical control event in the toll data under the influence of historical control into the toll prediction model for different vehicle types of the corresponding control section, and obtain the predicted toll for different vehicle types of the control section corresponding to each historical control event. Calculate the sum of the predicted toll fees for different vehicle types at the control section corresponding to each historical control event to obtain the predicted toll fee data for each historical control event; The difference between the predicted toll data for each historical control event and the toll data for the corresponding historical control event is calculated to obtain the toll revenue loss for each historical control event. All the toll revenue losses from the aforementioned historical control events constitute toll revenue losses under the influence of historical control measures.

Images