A multi-source data processing method and system

Abstract

The present application provides a multi-source data processing method and system, which relates to the technical field of data processing. The method comprises: first acquisition, modeling, model training and prediction steps; the system comprises: a first acquisition module, a modeling module, a model training module and a prediction module. The present application can improve the accuracy of the estimated arrival time in the navigation information.

Description

Technical Field

[0001] The present application relates to the technical field of data processing, and in particular to a multi-source data processing method and system. Background Art

[0002] With the acceleration of urbanization and the continuous increase in the number of cars, traffic congestion has become increasingly serious, which has had a significant impact on people's travel efficiency and quality of life. In order to meet this challenge, Intelligent Transportation Systems (ITS) came into being, aiming to achieve intelligent management and optimization of traffic information by integrating advanced information technology, communication technology, control technology and traffic engineering technology.

[0003] In intelligent transportation systems, dynamic path planning and navigation optimization are key technologies. Traditional navigation systems mainly rely on static map data and fixed route planning algorithms, ignoring the impact of surrounding vehicles on the vehicle, resulting in deviations between the estimated arrival time and the actual arrival time in navigation information. Summary of the invention

[0004] In order to improve the accuracy of predicted arrival time in navigation information, the present application provides a multi-source data processing method and system.

[0005] In the first aspect, the present application provides a multi-source data processing method, which adopts the following technical solution:

[0006] A multi-source data processing method comprises the following steps:

[0007] First collection: collect historical vehicle information and historical traffic information, record as first data; collect navigation information, use navigation information as the real label of first data; collect surrounding vehicle data, record as second data; use the difference between the time of arrival at the destination in the navigation information and the actual time of arrival at the destination as the real label of the second data; collect vehicle information, real-time traffic information and surrounding vehicle data of the vehicle, record as third data;

[0008] Modeling: establishing a prediction model, wherein the prediction model includes: a BI-LSTM sub-model and a Bayesian network sub-model;

[0009] Model training: includes the first training, the second training, the first update and the model output;

[0010] First training: the BI-LSTM sub-model is trained using the first data and the true label of the first data to obtain predicted navigation information, which is recorded as the first output data;

[0011] Second training: using the first output data, the second data and the real label of the second data to train the Bayesian network sub-model to obtain the time difference corresponding to the current navigation information;

[0012] First update: obtaining the time of arrival at the destination in the first output data, recorded as the first time, summing the first time and the time difference to obtain the predicted arrival time, and updating the predicted arrival time to the first output data to obtain the second output data;

[0013] Model output: obtaining navigation information with the earliest arrival time in the second output data, and using the navigation information as output data of the prediction model;

[0014] Prediction: Input the third data into the trained prediction model and output the navigation information of the vehicle.

[0015] By adopting the above technical solution, the present application trains the BI-LSTM sub-model through historical vehicle information and historical traffic information to obtain predicted navigation information, and then uses the predicted navigation information, surrounding vehicle data and the actual time deviation of arriving at the destination to train the Bayesian network sub-model, and combines the outputs of the BI-LSTM sub-model and the Bayesian network sub-model to obtain a more accurate estimated arrival time by summing them up. The present application can integrate the outputs of the two sub-models, take into account the influence of surrounding vehicles, and improve the accuracy of the overall prediction.

[0016] Optionally, after executing the first updating step and before executing the model output step, the method further includes:

[0017] First judgment: obtaining the position of the vehicle and judging whether the vehicle is in a state of waiting for a traffic light. If so, executing the second adjustment step; if not, executing the model output step;

[0018] Second adjustment: including second collection, second judgment, first calculation, third judgment and second calculation;

[0019] Second collection: obtaining the estimated passing time of the current traffic light in the second output data, which is recorded as the second time;

[0020] Second judgment: judging whether the second time is greater than the green light interval time in the direction of travel of the vehicle, if so, executing the first judgment step after a green light interval time; if not, executing the first calculation step;

[0021] First calculation: based on the position of the vehicle, the position of the stop line and the second time, calculate the acceleration of the vehicle and the speed at which the vehicle reaches the stop line;

[0022] The third judgment: judging whether the speed of the vehicle reaching the stop line is legal, if so, executing the model output step; if not, executing the second calculation step;

[0023] Second calculation: adding a green light interval time to the second output data to obtain new second output data.

[0024] By adopting the above technical solution, when a vehicle is waiting for a traffic light, by comparing the second time with the green light interval time, it is determined whether the vehicle needs to wait for the next green light, and the speed at the stop line calculated based on the vehicle position, the stop line position and the expected passing time helps to evaluate whether the vehicle can pass the current traffic light within a legal and safe speed range. When the speed at which the vehicle reaches the stop line is illegal, the second output data is adjusted by adding a green light interval time to improve the accuracy of time prediction. By providing accurate navigation information and traffic light passing strategies, the present application can provide users with a more intelligent and convenient driving experience.

[0025] Optionally, after performing the second adjustment step and before performing the model output step, the method further includes:

[0026] Fourth judgment: judging whether there is a congested road section according to the real-time traffic information, if so, executing the third updating step; if not, executing the model output step;

[0027] Third update: Taking the current position of the vehicle as the starting point, perform the prediction steps.

[0028] By adopting the above technical solution, it is possible to quickly perceive changes in the traffic environment and foresee potential traffic congestion by judging whether there is a congested section through real-time traffic information, so as to take measures in advance to prevent the vehicle from getting stuck in serious traffic jams as much as possible. When a congested section is detected, the current position of the vehicle can be immediately used as the starting point to re-execute the prediction step, which reflects the high flexibility and adaptability of this method. Through re-prediction, this method can provide users with a better path, thereby saving time and energy. This method can perceive changes in the traffic environment in real time and automatically adjust the driving route, thereby improving traffic efficiency and reducing traffic congestion and delays.

[0029] Optionally, after executing the fourth judgment step and before executing the third update step, the method further includes:

[0030] Fifth judgment: judging whether the congested road section is a must-pass road section, if yes, executing the third calculation step; if no, executing the third update step;

[0031] The third calculation: obtain the traffic volume and road capacity of the congested road section and calculate the degree of traffic congestion. The calculation model is as follows:

[0032] ;

[0033] Among them, A is the degree of traffic congestion; B is the road capacity; C is the traffic volume;

[0034] Third collection: collecting the original passing time of the congested road section described in the navigation information, and recording it as the third time;

[0035] Fourth calculation: calculating the product of the third time and the traffic congestion degree to obtain a fourth time;

[0036] Fourth update: updating the fourth time into the second output data to obtain new second output data.

[0037] By adopting the above technical solution, by judging whether the congested road section is a must-pass road section, the present application can more accurately identify those congestion situations that have a significant impact on the journey. For the must-pass congested road sections, the present application will further calculate the congestion level and optimize the arrival time. The original passing time (third time) of the congested road section in the navigation information is collected, and the product of the traffic congestion level and the third time is calculated, and the arrival time is updated, which can provide users with the latest travel time information including the impact of congestion. The present application can more intelligently handle the situation of the must-pass congested road sections, and provide users with more accurate travel time predictions and more reasonable route planning suggestions.

[0038] Optionally, after executing the first updating step and before executing the model output step, the method further includes:

[0039] Setting the evaluation function: Setting the evaluation function of the i-th second output data, the calculation model of the evaluation function is as follows:

[0040] ;

[0041] in, is the evaluation function of the i-th second output data, is the driving time in the i-th second output data; is the jth acceleration in the i-th second output data; m is the number of accelerations; is the fuel consumption in the i-th second output data;

[0042] Fifth calculation: calculating an evaluation result of each second output data based on the evaluation function;

[0043] Fifth update: obtain the second output data corresponding to the maximum evaluation result, and record it as the third output data.

[0044] By adopting the above technical solution, the present application sets an evaluation function based on multiple factors such as driving time, acceleration and fuel consumption, which is used to evaluate the current navigation information, improve the accuracy and objectivity of decision-making, and calculate the evaluation result of each second output data based on the evaluation function, so as to convert the originally abstract driving scheme into a specific quantitative indicator for easy comparison and selection. Through calculation, the present application can quickly obtain the evaluation result of each driving scheme, thereby improving the efficiency of decision-making. The present application selects the second output data corresponding to the maximum evaluation result and uses it as the third output data, so that the user does not need to manually compare the pros and cons of different driving schemes. The present application can automatically select the best scheme for the user, thereby improving the user's driving experience and satisfaction. The present application can provide users with the best decision-making suggestions based on data support, thereby enhancing the user's trust and dependence.

[0045] Optionally, the evaluation function is also calculated using the following calculation model:

[0046] ;

[0047] in, is the average value of the acceleration in the i-th second output data.

[0048] By adopting the above technical solution, the present application modifies the average value of acceleration into the standard deviation of acceleration, which can reflect the degree of dispersion of acceleration in the navigation information, and the standard deviation of acceleration can quantify the fluctuation of acceleration during vehicle driving. A high standard deviation means a large change in acceleration, indicating a more aggressive driving behavior or complex driving conditions, such as frequent acceleration and deceleration; while a low standard deviation indicates a stable change in acceleration, a relatively mild driving behavior or a simple driving condition, which helps the present application to more accurately evaluate driving style and road condition information, provide more reliable input data, and after replacing the average value of acceleration with the standard deviation, it can recommend reasonable navigation routes for different driving environments and road condition changes.

[0049] Optionally, the step of outputting the model further includes:

[0050] First comparison: comparing the evaluation result of the third output data with the evaluation result of the second output data with the earliest arrival time to obtain a comparison result;

[0051] Sixth update: Based on the comparison results, set the output data of the prediction model.

[0052] By adopting the above technical solution, by comparing the evaluation results of the third output data (i.e., the driving plan with the best evaluation result) with the second output data with the earliest arrival time, the advantages and disadvantages of different driving plans can be evaluated more comprehensively. The optimal driving plan may not be the plan with the earliest arrival time, but it may have advantages such as lower fuel consumption and a smoother driving experience. Through comparison, it is possible to provide users with a driving suggestion that balances performance and time efficiency to meet the diverse needs of users. This application can provide users with the best decision-making suggestions based on data support, which helps users make more informed driving decisions and reduce unnecessary waiting time and energy consumption.

[0053] Optionally, after executing the fifth updating step and before executing the model output step, the method further includes:

[0054] Sorting: sorting the second output data in descending order of the evaluation results to obtain a second output data sequence;

[0055] Filtering: filtering the first three second output data based on the second output data sequence, and recording them as the first navigation information, the second navigation information and the third navigation information in sequence;

[0056] Sixth calculation: calculating the distance between the remaining navigation information in the second output data sequence and the first navigation information, recorded as the first distance, calculating the distance between the remaining navigation information in the second output data sequence and the second navigation information, recorded as the second distance, and calculating the distance between the remaining navigation information in the second output data sequence and the third navigation information, recorded as the third distance;

[0057] Seventh update: updating the remaining navigation information based on the first distance, the second distance, and the third distance;

[0058] Eighth update: updating the first navigation information based on the remaining navigation information to obtain new first navigation information, updating the second navigation information to obtain new second navigation information, updating the third navigation information to obtain new third navigation information;

[0059] Iteration: Repeat the sorting to the eighth update step until the preset number of iterations is met;

[0060] Ninth update: using the new first navigation information after iteration as the fourth output data.

[0061] By adopting the above technical solution, the optimal driving plan can be clearly identified by sorting the second output data (i.e., the evaluation results of different driving plans). Calculating the distance between the remaining navigation information and the first three optimal navigation information helps to evaluate the similarities and differences between these navigation information. By updating the first navigation information / the second navigation information / the third navigation information, possible driving plans can be obtained in advance, and updated according to the new information, which helps to better adapt to the dynamically changing traffic environment. By evaluating the new navigation information and iterating according to the evaluation results, the driving plan can be continuously optimized and adjusted until the preset number of iterative executions is met. By repeatedly executing the sorting step to the eighth update, the present application can continuously iterate and optimize the navigation information, so that the present application can output a more accurate and reliable driving plan when facing a complex and changing traffic environment.

[0062] Optionally, in the model output step, the fourth output data is used as output data of the prediction model.

[0063] By adopting the above technical solution, the fourth output data is obtained after a series of complex processing and iterative optimization. Compared with directly outputting unprocessed raw data or preliminarily processed data, the fourth output data may have a lower prediction error because it has been more finely tuned. Since the fourth output data is obtained through iterative optimization, it can better adapt to changes and uncertainties in the data, thereby making the prediction model more robust when facing new data.

[0064] In a second aspect, the present invention provides a multi-source data processing system, which adopts the following technical solution:

[0065] A multi-source data processing system, comprising:

[0066] The first acquisition module is used to collect historical vehicle information and historical traffic information, which are recorded as first data; collect navigation information, and use the navigation information as the real label of the first data; the first acquisition module is also used to collect surrounding vehicle data, which are recorded as second data; the first acquisition module is also used to use the difference between the time of arrival at the destination in the navigation information and the actual time of arrival at the destination as the real label of the second data; the first acquisition module is also used to collect vehicle information, real-time traffic information and surrounding vehicle data of the vehicle, which are recorded as third data;

[0067] A modeling module, used to establish a prediction model, wherein the prediction model includes: a BI-LSTM sub-model and a Bayesian network sub-model;

[0068] A model training module is connected to the first acquisition module and the modeling module in communication, and the model training module includes a first training unit, a second training unit, a first updating unit and a model output unit;

[0069] A first training unit is used to train the BI-LSTM sub-model using the first data and the real label of the first data to obtain predicted navigation information, which is recorded as first output data;

[0070] A second training unit is used to train the Bayesian network sub-model using the first output data, the second data and the real label of the second data to obtain a time difference corresponding to the current navigation information;

[0071] A first updating unit is used to obtain the time of arrival at the destination in the first output data, recorded as the first time, sum the first time and the time difference to obtain the predicted arrival time, and update the predicted arrival time to the first output data to obtain the second output data;

[0072] a model output unit, configured to obtain navigation information with the earliest arrival time in the second output data, and use the navigation information as output data of the prediction model;

[0073] The prediction module is connected to the first acquisition module and the model output module for inputting the third data into the trained prediction model and outputting the navigation information of the vehicle.

[0074] By adopting the above technical solution, the present application comprehensively collects historical vehicle, traffic information and navigation data through the first acquisition module, and introduces surrounding vehicle data and actual arrival time difference as real labels, and uses the prediction model constructed by BI-LSTM and Bayesian network for phased training to optimize the accuracy of navigation information and arrival time prediction, and finally outputs the navigation information with the earliest arrival time as the prediction result, thereby realizing efficient, accurate and practical vehicle navigation services, and having good scalability to adapt to the increase of data sources and model optimization in the future.

[0075] In summary, the present application includes at least one of the following beneficial technical effects:

[0076] 1. This application trains the BI-LSTM sub-model through historical vehicle information and historical traffic information to obtain predicted navigation information, and then uses the predicted navigation information, surrounding vehicle data and the actual time deviation of arriving at the destination to train the Bayesian network sub-model. The outputs of the BI-LSTM sub-model and the Bayesian network sub-model are combined, and a more accurate estimated arrival time is obtained by summing. This application can integrate the outputs of the two sub-models, take into account the influence of surrounding vehicles, and improve the accuracy of the overall prediction.

[0077] 2. By sorting the second output data (i.e., the evaluation results of different driving plans), the optimal driving plan can be clearly identified. Calculating the distance between the remaining navigation information and the first three optimal navigation information helps to evaluate the similarities and differences between these navigation information. By updating the first navigation information / second navigation information / third navigation information, possible driving plans can be obtained in advance and updated according to the new information, which helps to better adapt to the dynamically changing traffic environment. By evaluating the new navigation information and iterating according to the evaluation results, the driving plan can be continuously optimized and adjusted until the preset number of iterations is met. By repeatedly executing the sorting step to the eighth update, the present application can continuously iterate and optimize the navigation information, so that the present application can output a more accurate and reliable driving plan when facing a complex and changing traffic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 is a flow chart of Example 1 of the present application;

[0079] Figure 2 is a flow chart of Example 2 of the present application;

[0080] Figure 3 This is a flowchart of the sixth update from S61 setting the evaluation function to S342 of Example 3 of the present application;

[0081] Figure 4 It is a flowchart of S71 sorting to S34 model output in Example 3 of the present application. DETAILED DESCRIPTION

[0082] The following combination Figures 1 to 4 This application is described in further detail.

[0083] Embodiment 1: This embodiment discloses a multi-source data processing method, referring to Figure 1 The method includes: S1 first collection, S2 modeling, S3 model training and S4 prediction. First, relevant data is collected, then a prediction model is established, and then the prediction model is trained using historical data in the relevant data. Then, the real-time data in the relevant data is input into the trained prediction model to obtain a prediction result. The specific process of this embodiment is as follows:

[0084] S1 is the first collection, first collecting historical vehicle information and historical traffic information, and recording the historical vehicle information and historical traffic information as the first data. The historical vehicle information includes: vehicle speed, acceleration and fuel consumption; the historical traffic information includes: speed limit information, traffic light information and congestion information.

[0085] Then, navigation information is collected and used as a real label of the first data, wherein the navigation information includes: a navigation route and an estimated arrival time.

[0086] Then, surrounding vehicle data is collected and recorded as second data, wherein the surrounding vehicle data includes: driving speed, position information and driving direction of the surrounding vehicles.

[0087] Then, the difference between the estimated time to arrive at the destination in the navigation information and the actual time to travel to the destination is used as the true label of the second data.

[0088] Then, the vehicle information, real-time traffic information and surrounding vehicle data of the vehicle are collected and recorded as the third data. During driving, surrounding vehicles may change lanes (cars in other lanes change to the lane of the vehicle), the front vehicle or the front vehicle may brake suddenly, etc., all of which will affect the accuracy of navigation information.

[0089] S2 modeling, establish a prediction model, the prediction model includes: BI-LSTM sub-model and Bayesian network sub-model, the output of the BI-LSTM sub-model is connected to the input of the Bayesian network sub-model. The BI-LSTM sub-model can use the time series characteristics of historical data for learning and prediction, while the Bayesian network sub-model can consider the probabilistic relationship between different factors to improve the accuracy and robustness of the prediction.

[0090] S3 model training, including S31 first training, S32 second training, S33 first update and S34 model output.

[0091] S31 is the first training, in which the BI-LSTM sub-model is trained using the first data and the real label of the first data to obtain predicted navigation information, which is recorded as the first output data. The predicted navigation information can reflect the possible driving route and estimated arrival time of the vehicle.

[0092] S32 second training, using the first output data, the second data and the real label of the second data to train the Bayesian network sub-model to obtain the time difference corresponding to the current navigation information.

[0093] S33 is the first update, obtaining the time of arrival at the destination in the first output data, recorded as the first time, summing the first time and the time difference to obtain the predicted arrival time, updating the predicted arrival time to the first output data, and obtaining the second output data.

[0094] S34: Model output, obtaining navigation information with the earliest arrival time in the second output data, and using the navigation information as output data of the prediction model.

[0095] S4 predicts, inputs the third data into the trained prediction model, and outputs the navigation information of the vehicle.

[0096] This embodiment comprehensively collects historical vehicles, traffic information and navigation data through the first acquisition step, and introduces the surrounding vehicle data and the actual arrival time difference as the real label to build a prediction model including a BI-LSTM sub-model and a Bayesian network sub-model. After phased training, including using historical data to train the BI-LSTM sub-model, combining surrounding vehicle data to train the Bayesian network sub-model, and updating the predicted arrival time, the navigation information with the earliest arrival time is finally output as the prediction result. By inputting real-time vehicle, traffic and surrounding vehicle data into the trained model, users can be provided with efficient, accurate and real-time vehicle navigation services.

[0097] Example 2: Reference Figure 2 The difference between this embodiment and the first embodiment is that after executing the first update in S33 and before executing the model output in S34, the following is further included:

[0098] S51 is the first judgment, which obtains the position of the vehicle through the vehicle's GPS system, and obtains the status of the traffic light in the form of V2I through the vehicle's V2X system. Then, based on the position of the vehicle and the current status of the traffic light, it is determined whether the vehicle is in a state of waiting for the traffic light. If so, the second adjustment of S52 is executed; if not, the model output of S34 is executed.

[0099] S52 second adjustment includes S521 second collection, S522 second judgment, S523 first calculation, S524 third judgment and S525 second calculation.

[0100] S521 is a second acquisition, in which the estimated passing time of the current traffic light in the second output data is obtained and recorded as the second time.

[0101] S522 is the second judgment, judging whether the second time is greater than the green light interval time in the direction of travel of the vehicle. If so, it means that the vehicle cannot pass the current traffic light within one green light time, and it is necessary to execute S51 first judgment after a green light interval time; if not, it means that the vehicle can pass the current traffic light within one green light time, and execute S523 first calculation.

[0102] S523 is a first calculation, which is to calculate the acceleration of the vehicle and the speed at which the vehicle reaches the stop line based on the position of the vehicle, the position of the stop line (the position of the stop line is obtained based on the vehicle-mounted camera and the sensor) and the second time. In this step, the calculation model of the speed is as follows:

[0103] ;

[0104] ;

[0105] in, is the speed of the vehicle reaching the stop line; X is the distance from the vehicle to the stop line; T is the second time; is the acceleration of the vehicle; because the vehicle is in the state of waiting for the traffic light, the initial speed of the vehicle is 0.

[0106] S524 is the third judgment, judging whether the speed of the vehicle reaching the stop line is legal (meets the speed limit requirements of the current road section). If so, execute S34 model output; if not, execute S525 second calculation.

[0107] S525: second calculation, adding a green light interval to the second output data to obtain new second output data.

[0108] S53 is the fourth judgment, judging whether there is a congested road section according to the real-time traffic information. If so, executing S54 is the fifth judgment; if not, executing S34 model output.

[0109] S54 is a fifth judgment, determining whether the congested road section is a must-pass road section. If so, executing S55 a third calculation; if not, executing S59 a third update.

[0110] S55 performs the third calculation, obtaining the traffic volume and road capacity of the congested road section, and calculating the degree of traffic congestion. The calculation model is as follows:

[0111] ;

[0112] Among them, A is the degree of traffic congestion, B is the road capacity, and C is the traffic volume of the road section. When traffic is congested, the road capacity B is less than the traffic volume C, that is, A is greater than 1.

[0113] S56: third collection, collecting the estimated passing time of the congested road section described in the navigation information, and recording it as the third time.

[0114] S57: fourth calculation, calculating the product of the third time and the traffic congestion degree to obtain a fourth time.

[0115] S58: fourth updating, updating the fourth time into the second output data to obtain new second output data.

[0116] S59 is the third update. When there is a non-essential congested road section, the driving plan is adjusted and the route is replanned, that is, the current position of the vehicle is taken as the starting point, and the S4 prediction is executed.

[0117] This embodiment further includes, between the first updating step and the model output step: a first judging step, judging whether the vehicle is waiting for the traffic light through the GPS and V2X system, and if so, entering the second adjusting step (including collecting the estimated passing time, judging the possibility of passing, calculating the acceleration and speed, judging the legality of the speed, and increasing the green light interval time when necessary), and if not or the speed is legal, executing the model output step; at the same time, the fourth judging step checks whether there is congestion according to the real-time traffic information, and if so, entering the fifth judging step (i.e. judging whether the congested section is a must-pass section, calculating the degree of traffic congestion, collecting the estimated passing time, calculating the product and updating the data or adjusting the driving plan and replanning the route), and if not, directly executing the model output step.

[0118] Example 3: Reference Figure 3 The difference between this embodiment and the first embodiment is that after executing the first update in S33 and before executing the model output in S34, the following is further included:

[0119] S61 sets an evaluation function, which sets the evaluation function of the i-th second output data, aiming to quantify the quality of each navigation information. The calculation model of the evaluation function is as follows:

[0120] ;

[0121] in, is the evaluation function of the i-th second output data; is the driving time in the i-th second output data; is the jth acceleration in the i-th second output data; m is the number of accelerations; is the fuel consumption in the i-th second output data.

[0122] The evaluation function is also calculated using the following calculation model:

[0123] ;

[0124] in, is the average value of the acceleration in the i-th second output data. Replacing the average value in the original evaluation function with the standard deviation helps to evaluate the stability during driving. is the evaluation function of the i-th second output data; is the driving time in the i-th second output data; is the jth acceleration in the i-th second output data; m is the number of accelerations; is the fuel consumption in the i-th second output data.

[0125] S62: fifth calculation, calculating an evaluation result of each second output data based on the evaluation function.

[0126] S63 is the fifth update, obtaining the second output data when the evaluation result is the largest, recorded as the third output data, and the third output data represents the optimal driving plan under the current situation.

[0127] S34 model output, including: obtaining navigation information with the earliest arrival time in the second output data.

[0128] The S34 model output also includes: S341 first comparison and S342 sixth update.

[0129] S341 is a first comparison, in which the evaluation result of the third output data is compared with the evaluation result of the second output data with the earliest arrival time, to obtain a comparison result.

[0130] S342 Sixth update: based on the comparison result, set the output data of the prediction model. If the evaluation result of the third output data is better than the second output data with the earliest arrival time, then select the third output data as the final driving plan. Otherwise, select the second output data with the earliest arrival time.

[0131] For example, the vehicle is planning the optimal driving route from location A to location B.

[0132] Set the initial conditions:

[0133] Place A: Starting point, current location of the vehicle.

[0134] Place B: The end point, the destination where the vehicle needs to reach.

[0135] Second output data: multiple candidate solutions have been generated based on current traffic information and possible driving routes (each solution includes information such as driving time, acceleration, fuel consumption, etc.).

[0136] For example, the second output data includes the following three schemes:

[0137] Solution 1: Driving time 60 minutes, acceleration -0.5m / s 2 , 0.5m / s 2 、-1m / s 2 , fuel consumption is 8L.

[0138] Option 2: Driving time 65 minutes, acceleration 2m / s 2 , 0.5m / s 2 、-0.5m / s 2 , fuel consumption is 7L.

[0139] Option 3: Driving time 70 minutes, acceleration 0.5m / s 2 、-1m / s 2 , fuel consumption is 9L.

[0140] S61 Setting the evaluation function

[0141] An evaluation function is set for each candidate solution, which comprehensively considers driving time, acceleration change and fuel consumption.

[0142] ;

[0143] in, is the evaluation function of the i-th second output data; is the driving time in the i-th second output data; is the jth acceleration in the i-th second output data; m is the number of accelerations; is the fuel consumption in the i-th second output data.

[0144] S62 Fifth calculation, the evaluation results of each solution are as follows:

[0145] Solution 1: .

[0146] Option 2: .

[0147] Option 3: .

[0148] S63 Fifth update: The second output data corresponding to the solution 2 with the largest evaluation result is used as the third output data.

[0149] S34 model output: obtain the navigation information with the earliest arrival time in the second output data, that is, the navigation information with the shortest time, that is, plan one.

[0150] S341 is the first comparison, which compares the evaluation result of the third output data (Scheme 2) with the scheme 1 which takes the shortest time, and finds that the evaluation result of Scheme 2 is higher.

[0151] S342 is the sixth update, based on the comparison result in the first comparison in S341, setting the output of the prediction model to scheme 2.

[0152] This embodiment sets an evaluation function for quantifying the quality of each navigation information, which is calculated based on driving time, acceleration (including average value and standard deviation to evaluate stability) and fuel consumption. Then, the evaluation result of each second output data is calculated based on the evaluation function, and the data when the evaluation result is the largest is obtained as the third output data, representing the optimal driving plan. In addition, it also includes a model output step, in which the evaluation results of the third output data and the second output data with the earliest arrival time are compared, and the output data of the prediction model is set based on the comparison result: if the third output data is better, it is selected as the final driving plan; otherwise, the data with the earliest arrival time is selected.

[0153] Reference Figure 4 In other embodiments, after executing S62 fifth update and before executing S34 model output, it also includes:

[0154] S71 sorting: sorting the second output data in descending order of the evaluation results to obtain a second output data sequence. After the sorting is completed, an ordered second output data sequence will be obtained, which reflects the order of each navigation information according to its merits.

[0155] S72 screening, based on the second output data sequence, screening the first three second output data, and recording them as first navigation information, second navigation information and third navigation information in sequence. These three navigation information are ranked in the top three in the evaluation result, so they are considered to be relatively optimal navigation solutions.

[0156] S73 sixth calculation, using the Hamming algorithm or the Euclidean distance algorithm, calculates the distance between the remaining navigation information in the second output data sequence and the first navigation information, recorded as the first distance, calculates the distance between the remaining navigation information in the second output data sequence and the second navigation information, recorded as the second distance, and calculates the distance between the remaining navigation information in the second output data sequence and the third navigation information, recorded as the third distance.

[0157] S74 seventh update, updating the remaining navigation information based on the first distance, the second distance and the third distance.

[0158] S75 eighth update: update the first navigation information at the next moment based on the remaining navigation information to obtain new first navigation information, update the second navigation information at the next moment to obtain new second navigation information, update the third navigation information at the next moment to obtain new third navigation information.

[0159] S76 iteration, repeatedly executing S71 sorting to S75 eighth update until the preset number of iterations is met.

[0160] S77 is the ninth update, and the new first navigation information after iteration is used as the fourth output data.

[0161] S34 model output, using the fourth output data as output data of the prediction model.

[0162] For example, there are four second output data, namely:

[0163] Solution 1: Driving time 60 minutes, acceleration -0.5m / s 2 , 0.5m / s 2 、-1m / s 2 , fuel consumption is 8L.

[0164] Option 2: Driving time 65 minutes, acceleration 2m / s 2 , 0.5m / s 2 、-0.5m / s 2 , fuel consumption is 7L.

[0165] Option 3: Driving time 70 minutes, acceleration 0.5m / s 2 、-1m / s 2 , fuel consumption is 9L.

[0166] Option 4: Driving time 55 minutes, acceleration 1m / s 2 、-1m / s 2 , fuel consumption is 6L.

[0167] The evaluation results of each plan are as follows:

[0168] Solution 1: ;

[0169] Option 2: ;

[0170] Option 3: ;

[0171] Option 4: .

[0172] S71 sorting, the second output data sequence is: plan four, plan two, plan three, plan one.

[0173] S72 screening, the first navigation information is solution four, the second navigation information is solution two, and the third navigation information is solution three.

[0174] The remaining navigation information is solution one.

[0175] Then, scheme 1, scheme 2, scheme 3, and scheme 4 are encoded respectively to obtain four vectors , , , , , , , They respectively represent the position of the first navigation information, the position of the second navigation information, the position of the third navigation information and the position of the fourth navigation.

[0176] S73 Sixth calculation: the model for calculating the distance is as follows:

[0177] ;

[0178] ;

[0179] Wherein, G is the distance between the remaining navigation information in the second output data and the i-th navigation information; E is a random number with a value of (0,2); is the position of the i-th navigation information in the d-th iteration; d is the current iteration number; is the position of the j-th navigation information in the d-th iteration; is the position of the j-th navigation information in the d+1-th iteration; H is a random number with a value of (-e, e), e is a constant with an initial value of 2, and its value decreases linearly from 2 to 0 as the number of iterations increases.

[0180] ;

[0181] in, is the first distance, is the second distance, is the third distance; E is a random number with a value of (0,2); , , , They respectively represent the position of the first navigation information, the position of the second navigation information, the position of the third navigation information and the position of the fourth navigation.

[0182] S74 seventh update: updating the remaining navigation information based on the first distance, the second distance and the third distance, the process is as follows:

[0183] ;

[0184] ;

[0185] in, is the intermediate parameter; , , Respectively represent the position of the first navigation information, the position of the second navigation information, and the position of the third navigation information; H is a random number with a value of (-e, e), e is a constant, and its initial value is 2. As the number of iterations increases, its value decreases linearly from 2 to 0; is the first distance, is the second distance, is the third distance; is the position of the fourth navigation information in the d+1th iteration process.

[0186] S75 eighth update, based on renew , , , the process is: calculate , , , Based on the evaluation results, update Plan 4, Plan 2, and Plan 3.

[0187] S76 iteration, repeatedly executing S71 sorting to S75 eighth update until the preset number of iterations is met.

[0188] S77 is the ninth update, and the new first navigation information after iteration is used as the fourth output data.

[0189] S34 model output, using the fourth output data as output data of the prediction model.

[0190] In this embodiment, the second output data is first sorted to obtain an ordered sequence reflecting the quality of the navigation information. Then, the top three navigation information are selected as the optimal solution. Then, the distance between the remaining navigation information and the top three navigation information is calculated. Based on these distances, the remaining navigation information is updated. The top three navigation information at the next moment is further updated. The iteration is performed until a preset number of times is reached. Then, the new first navigation information after iteration is used as the fourth output data and output by the model. In this embodiment, the navigation information can be continuously optimized to obtain better navigation information.

[0191] Embodiment 4: This embodiment discloses a multi-source data processing system, the system comprising:

[0192] The first acquisition module is used to collect historical vehicle information and historical traffic information, which are recorded as the first data; collect navigation information and use the navigation information as the real label of the first data; the first acquisition module is also used to collect surrounding vehicle data, which is recorded as the second data; the first acquisition module is also used to use the difference between the time of arrival at the destination in the navigation information and the actual time of arrival at the destination as the real label of the second data; the first acquisition module is also used to collect the vehicle information, real-time traffic information and the surrounding vehicle data of the vehicle, which are recorded as the third data.

[0193] The modeling module is used to establish a prediction model, and the prediction model includes: a BI-LSTM sub-model and a Bayesian network sub-model.

[0194] The model training module is communicatively connected with the first acquisition module and the modeling module, and the model training module includes a first training unit, a second training unit, a first updating unit and a model output unit.

[0195] The first training unit is used to train the BI-LSTM sub-model using the first data and the true label of the first data, and the trained BI-LSTM sub-model can predict the navigation information, that is, the first output data.

[0196] The second training unit is used to train the Bayesian network sub-model using the first output data, the second data and the true label of the second data, and the trained Bayesian network sub-model can predict the time difference corresponding to the current navigation information.

[0197] The first updating unit is used to obtain the time of arrival at the destination in the first output data, recorded as the first time, sum the first time and the time difference to obtain the predicted arrival time, and update the predicted arrival time to the first output data to obtain the second output data.

[0198] The model output unit is used to select the navigation information with the earliest arrival time from the second output data and use it as the output data of the prediction model.

[0199] The prediction module is connected to the first acquisition module and the model output module for inputting the third data into the trained prediction model and outputting the navigation information of the vehicle.

[0200] This embodiment integrates a first acquisition module to collect historical vehicles, traffic information and navigation data (including the time difference between the real label and the surrounding vehicle data), and constructs a prediction model including BI-LSTM and Bayesian network sub-models through the modeling module. The model training module uses these data to train the model, first predicting the navigation information through BI-LSTM, then combining the Bayesian network to predict the time difference to update the arrival time of the navigation information, and finally outputs the navigation with the earliest arrival time as the model prediction result. The prediction module receives vehicle and traffic data in real time, and inputs the trained model to provide instant navigation information.

[0201] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A multi-source data processing method, characterized in that: include: First collection: collect historical vehicle information and historical traffic information, which are recorded as first data; Collecting navigation information and using the navigation information as a true label of the first data; Collect surrounding vehicle data and record it as second data; The difference between the time of arrival at the destination in the navigation information and the actual time of arrival at the destination is used as the true label of the second data; Collect vehicle information, real-time traffic information and data of vehicles around the vehicle, and record them as third data; Modeling: establishing a prediction model, wherein the prediction model includes: a BI-LSTM sub-model and a Bayesian network sub-model; Model training: includes first training, second training, first update, setting evaluation function, fifth calculation, fifth update and model output; First training: the BI-LSTM sub-model is trained using the first data and the true label of the first data to obtain predicted navigation information, which is recorded as the first output data; Second training: using the first output data, the second data and the real label of the second data to train the Bayesian network sub-model to obtain the time difference corresponding to the current navigation information; First update: obtaining the time of arrival at the destination in the first output data, recorded as the first time, summing the first time and the time difference to obtain the predicted arrival time, and updating the predicted arrival time to the first output data to obtain the second output data; Setting the evaluation function: Setting the evaluation function of the i-th second output data, the calculation model of the evaluation function is as follows: ； in, is the evaluation function of the i-th second output data, is the driving time in the i-th second output data; is the jth acceleration in the i-th second output data; m is the number of accelerations; is the fuel consumption in the i-th second output data; The evaluation function is also calculated using the following calculation model: ； in, is the average value of the acceleration in the i-th second output data; Fifth calculation: calculating an evaluation result of each second output data based on the evaluation function; Fifth update: obtaining the second output data corresponding to the maximum evaluation result, recorded as the third output data; Model output: obtaining navigation information with the earliest arrival time in the second output data, and using the navigation information as output data of the prediction model; The step of outputting the model further includes: First comparison: comparing the evaluation result of the third output data with the evaluation result of the second output data with the earliest arrival time to obtain a comparison result; Sixth update: Based on the comparison results, set the output data of the prediction model; Prediction: Input the third data into the trained prediction model and output the navigation information of the vehicle.

2. The multi-source data processing method according to claim 1, characterized in that: After executing the first updating step and before executing the model output step, the method further includes: First judgment: obtaining the position of the vehicle and judging whether the vehicle is in a state of waiting for a traffic light. If so, executing the second adjustment step; if not, executing the model output step; Second adjustment: including second collection, second judgment, first calculation, third judgment and second calculation; Second collection: obtaining the estimated passing time of the current traffic light in the second output data, which is recorded as the second time; Second judgment: judging whether the second time is greater than the green light interval time in the direction of travel of the vehicle, if so, executing the first judgment step after a green light interval time; if not, executing the first calculation step; First calculation: based on the position of the vehicle, the position of the stop line and the second time, calculate the acceleration of the vehicle and the speed at which the vehicle reaches the stop line; The third judgment: judging whether the speed of the vehicle reaching the stop line is legal, if so, executing the model output step; if not, executing the second calculation step; Second calculation: adding a green light interval time to the second output data to obtain new second output data.

3. The multi-source data processing method according to claim 2, characterized in that: After performing the second adjustment step and before performing the model output step, the method further includes: Fourth judgment: judging whether there is a congested road section according to the real-time traffic information, if so, executing the third updating step; if not, executing the model output step; Third update: Taking the current position of the vehicle as the starting point, perform the prediction steps.

4. The multi-source data processing method according to claim 3, characterized in that: After executing the fourth determination step and before executing the third update step, the method further includes: Fifth judgment: judging whether the congested road section is a must-pass road section, if yes, executing the third calculation step; if no, executing the third update step; The third calculation: obtain the traffic volume and road capacity of the congested road section and calculate the degree of traffic congestion. The calculation model is as follows: ； Among them, A represents the degree of traffic congestion; B represents the road capacity; C represents the traffic volume; Third collection: collecting the original passing time of the congested road section described in the navigation information, and recording it as the third time; Fourth calculation: calculating the product of the third time and the traffic congestion degree to obtain a fourth time; Fourth update: updating the fourth time into the second output data to obtain new second output data.

5. The multi-source data processing method according to claim 1, characterized in that: After executing the fifth updating step and before executing the model output step, the method further includes: Sorting: sorting the second output data in descending order of the evaluation results to obtain a second output data sequence; Filtering: filtering the first three second output data based on the second output data sequence, and recording them as the first navigation information, the second navigation information and the third navigation information in sequence; Sixth calculation: calculating the distance between the remaining navigation information in the second output data sequence and the first navigation information, recorded as the first distance, calculating the distance between the remaining navigation information in the second output data sequence and the second navigation information, recorded as the second distance, and calculating the distance between the remaining navigation information in the second output data sequence and the third navigation information, recorded as the third distance; Seventh update: updating the remaining navigation information based on the first distance, the second distance, and the third distance; Eighth update: updating the first navigation information based on the remaining navigation information to obtain new first navigation information, updating the second navigation information to obtain new second navigation information, updating the third navigation information to obtain new third navigation information; Iteration: Repeat the sorting to the eighth update step until the preset number of iterations is met; Ninth update: using the new first navigation information after iteration as the fourth output data.

6. The multi-source data processing method according to claim 5, characterized in that: In the model output step, the fourth output data is used as output data of the prediction model.

7. A multi-source data processing system, the system being used to execute the method according to any one of claims 1 to 6, characterized in that: include: A first collection module, used for collecting historical vehicle information and historical traffic information, recorded as first data; Collecting navigation information and using the navigation information as a true label of the first data; The first acquisition module is further used to collect surrounding vehicle data, which is recorded as second data; the first acquisition module is further used to use the difference between the time of arrival at the destination in the navigation information and the actual time of arrival at the destination as the true label of the second data; The first acquisition module is also used to collect vehicle information, real-time traffic information and data of vehicles around the vehicle, which are recorded as third data; A modeling module, used to establish a prediction model, wherein the prediction model includes: a BI-LSTM sub-model and a Bayesian network sub-model; A model training module is connected to the first acquisition module and the modeling module in communication, and the model training module includes a first training unit, a second training unit, a first updating unit and a model output unit; A first training unit is used to train the BI-LSTM sub-model using the first data and the real label of the first data to obtain predicted navigation information, which is recorded as first output data; A second training unit is used to train the Bayesian network sub-model using the first output data, the second data and the real label of the second data to obtain a time difference corresponding to the current navigation information; A first updating unit is used to obtain the time of arrival at the destination in the first output data, recorded as the first time, sum the first time and the time difference to obtain the predicted arrival time, and update the predicted arrival time to the first output data to obtain the second output data; a model output unit, configured to obtain navigation information with the earliest arrival time in the second output data, and use the navigation information as output data of the prediction model; The prediction module is connected to the first acquisition module and the model output module for inputting the third data into the trained prediction model and outputting the navigation information of the vehicle.

Images