A Smart Charging Pile Scheduling Method and System Based on Traffic Flow Prediction

Abstract

This invention provides a smart charging pile scheduling method and system based on traffic flow prediction. By constructing and integrating a teacher charging pile scheduling network and an auxiliary charging pile scheduling network, the intelligence and accuracy of charging pile scheduling are effectively improved. The two networks, trained separately using teacher and auxiliary sample traffic flow data sequences, not only fully consider the characteristics and differences of different traffic monitoring areas, but also further improve the adaptability and prediction accuracy to traffic flow changes through the student charging pile scheduling network generated through iterative learning. The final target charging pile scheduling network can quickly determine reasonable charging demand data based on the actual traffic flow data of any target traffic monitoring area, and achieve efficient charging pile scheduling accordingly. This effectively alleviates the problem of uneven distribution of charging pile resources and improves the convenience and user experience of charging services.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a smart charging pile scheduling method and system based on traffic flow prediction. Background Technology

[0002] With the increasing popularity of electric vehicles and the growing complexity of transportation systems, the rational scheduling of charging stations has become a key issue in improving urban traffic efficiency and user experience. Traditional charging station scheduling methods often rely on fixed rules or simple prediction models, making it difficult to accurately cope with dynamic changes in traffic flow and regional differences. Especially in different traffic monitoring areas of a city, the diversity of factors such as traffic flow patterns, vehicle types, and charging demands exacerbates the supply-demand imbalance of charging stations. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a smart charging pile scheduling method based on traffic flow prediction, the method comprising:

[0004] A teacher charging station scheduling network is obtained, which is generated by parameter learning based on teacher sample traffic flow data sequences; the teacher sample traffic flow data sequences correspond to multiple first traffic monitoring areas; each first traffic monitoring area includes multiple teacher sample traffic flow data; each teacher sample traffic flow data carries first sample charging demand data.

[0005] An auxiliary charging pile scheduling network is obtained, which is generated by parameter learning based on auxiliary sample traffic flow data sequences; the auxiliary sample traffic flow data sequences correspond to multiple second traffic monitoring areas; each second traffic monitoring area includes multiple auxiliary sample traffic flow data; each auxiliary sample traffic flow data carries second sample charging demand data; each second traffic monitoring area does not overlap with any first traffic monitoring area.

[0006] The system obtains a student charging station scheduling network based on the teacher charging station scheduling network, and obtains a student sample traffic flow data sequence based on the teacher sample traffic flow data sequence and the auxiliary sample traffic flow data sequence; the student sample traffic flow data sequence includes multiple student sample traffic flow data; each student sample traffic flow data carries third sample charging demand data;

[0007] Based on the student sample traffic flow data sequence, the network parameters of the student charging pile scheduling network are iteratively learned until the network convergence requirement is met, and then the target charging pile scheduling network is generated.

[0008] Based on the target charging pile scheduling network, decisions are made on the target traffic flow data of any target traffic monitoring area to generate target charging demand data, and charging piles are scheduled for traffic participants in the target traffic monitoring area according to the target charging demand data.

[0009] In another aspect, embodiments of the present invention also provide a smart charging pile scheduling system based on traffic flow prediction, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, this application embodiment effectively improves the intelligence and accuracy of charging pile scheduling by constructing and integrating a teacher charging pile scheduling network and an auxiliary charging pile scheduling network. The two networks, trained using teacher sample traffic flow data sequences and auxiliary sample traffic flow data sequences respectively, not only fully consider the characteristics and differences of different traffic monitoring areas, but also further improve the adaptability and prediction accuracy to traffic flow changes through the student charging pile scheduling network generated through iterative learning. The final generated target charging pile scheduling network can quickly determine reasonable charging demand data based on the actual traffic flow data of any target traffic monitoring area, and achieve efficient charging pile scheduling accordingly. This effectively alleviates the problem of uneven distribution of charging pile resources, improves the convenience and user experience of charging services, and promotes the overall operational efficiency and rational energy utilization of the transportation system. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the smart charging pile scheduling method based on traffic flow prediction provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of the hardware architecture of a smart charging pile scheduling system based on traffic flow prediction provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a smart charging pile scheduling method based on traffic flow prediction, provided in one embodiment of the present invention. The following is a detailed description of this smart charging pile scheduling method based on traffic flow prediction.

[0014] Step S110: Obtain the teacher charging station scheduling network, which is generated based on parameter learning of teacher sample traffic flow data sequences. The teacher sample traffic flow data sequences correspond to multiple first traffic monitoring areas. Each first traffic monitoring area includes multiple teacher sample traffic flow data. Each teacher sample traffic flow data carries first sample charging demand data.

[0015] In this embodiment, it is assumed that in the transportation system of a large city, the server needs to build a network for scheduling teacher charging stations. The city is divided into multiple primary traffic monitoring zones, such as the city center business district, several large industrial parks, and major residential areas. Each zone has corresponding traffic flow monitoring equipment that continuously collects traffic flow data.

[0016] In the city's central business district, the primary traffic monitoring area, traffic flow data includes information on various vehicle types (such as private cars, taxis, and buses) at different times of day (e.g., weekday morning rush hour, midday off-peak, evening rush hour, and weekends). Each teacher's sample traffic flow data also includes first-sample charging demand data. Taking taxis as an example, during weekday evening rush hour, taxi traffic in this area is high. Due to the long operating hours of taxis, their charging demand is more urgent. The first-sample charging demand data might indicate a need to add temporary charging stations or adjust the charging power of existing charging stations to meet the fast charging demand.

[0017] For large industrial parks, there are many logistics vehicles entering and leaving the park during weekdays, which may be electric trucks or electric forklifts. The first example of charging demand data in the teacher sample traffic flow data may indicate that charging stations need to be set up near the cargo loading and unloading area to facilitate logistics vehicles to charge while loading and unloading goods, avoiding long waiting times for charging that would affect logistics efficiency.

[0018] The server collects traffic flow data from multiple teacher samples across different primary traffic monitoring areas and then uses this data for parameter learning. This learning process may involve various algorithms, such as neural network algorithms. The server inputs this data into the neural network, where neurons continuously adjust their weights to fit the patterns in the data. After a long learning process, the server ultimately generates a teacher charging station scheduling network. This network can predict corresponding charging demand based on the input traffic flow data, providing a basis for subsequent charging station scheduling.

[0019] Step S120: Obtain the auxiliary charging pile scheduling network, which is generated based on parameter learning of auxiliary sample traffic flow data sequences. The auxiliary sample traffic flow data sequences correspond to multiple second traffic monitoring areas. Each second traffic monitoring area includes multiple auxiliary sample traffic flow data. Each auxiliary sample traffic flow data carries second sample charging demand data. Each second traffic monitoring area does not overlap with any of the first traffic monitoring areas.

[0020] In this embodiment, an auxiliary charging pile scheduling network also needs to be constructed. Besides the first traffic monitoring area mentioned earlier, the city also has other second traffic monitoring areas, such as newly developed areas on the city's outskirts and tourist attractions in the suburbs. These areas do not overlap with the first traffic monitoring area.

[0021] Taking newly developed areas on the city's outskirts as an example, the traffic flow here is relatively less than in the first traffic monitoring area, and the traffic composition is also different, likely consisting more of construction vehicles and a small number of commuter vehicles. The auxiliary sample traffic flow data in each second traffic monitoring area carries second sample charging demand data. Construction vehicles operate primarily during the day, and their charging demand is concentrated near their parking areas; these characteristics are reflected in the second sample charging demand data within the auxiliary sample traffic flow data.

[0022] In suburban tourist areas, the number of tour buses and tourists' private cars increases during peak seasons. Tour buses have unique charging needs, potentially requiring high-powered charging stations in parking lots to meet their rapid charging requirements for continued travel. The server collects traffic flow data from multiple auxiliary sample areas across these different secondary traffic monitoring zones and uses algorithms similar to those used to construct the teacher charging station scheduling network (such as neural network algorithms) for parameter learning, thereby generating an auxiliary charging station scheduling network. This network can predict charging demand based on traffic flow conditions in the secondary traffic monitoring areas, assisting in the entire charging station scheduling decision-making process.

[0023] Step S130: Obtain the student charging station scheduling network based on the teacher charging station scheduling network, and obtain the student sample traffic flow data sequence based on the teacher sample traffic flow data sequence and the auxiliary sample traffic flow data sequence. The student sample traffic flow data sequence includes multiple student sample traffic flow data. Each student sample traffic flow data carries third sample charging demand data.

[0024] In this embodiment, the construction of the student charging station scheduling network and student sample traffic flow data sequences begins. First, the existing teacher charging station scheduling network is used. This network has already learned the relationship pattern between traffic flow and charging demand in the first traffic monitoring area. The server uses this pattern information to initialize the structure and parameters of the student charging station scheduling network. For example, if the teacher charging station scheduling network predicts taxi charging demand during the evening rush hour in the city center business district using specific neuron connection weights and thresholds, then the student charging station scheduling network will adopt a similar structure and initial parameters.

[0025] Next, the server obtains the student sample traffic flow data sequence based on the teacher sample traffic flow data sequence and the auxiliary sample traffic flow data sequence. Assume the server selects teacher sample traffic flow data from the teacher sample traffic flow data sequence. For example, it selects weekday morning rush hour data from the teacher sample traffic flow data of a city's central business district. The server performs multiple rounds of data optimization on the teacher sample traffic flow data sequence.

[0026] In the first round of data optimization, a teacher sample traffic flow data is randomly selected from the teacher sample traffic flow data sequence. For example, logistics vehicle traffic flow data during weekdays in a large industrial park is selected as the first candidate sample traffic flow data. Then, the first feature distance between each second candidate sample traffic flow data (other than the selected first candidate sample traffic flow data) in the teacher sample traffic flow data sequence and the first candidate sample traffic flow data is calculated. Taking the calculation of the first feature distance between the traffic flow data of private cars in residential areas on weekday evenings (as the second candidate sample traffic flow data) and the first candidate sample traffic flow data as an example, the server will obtain the first feature weighted value (e.g., the weighted value determined according to factors such as the type and load of logistics vehicles) and the first feature dispersion value (e.g., the degree of fluctuation of logistics vehicle traffic flow in different time periods) of the first candidate sample traffic flow data, and obtain the second feature weighted value (e.g., the weighted value determined according to factors such as the different models of private cars and the commuting habits of car owners) and the second feature dispersion value (e.g., the fluctuation of private car traffic flow in different time periods at night) of the second candidate sample traffic flow data, and determine the covariance value between the two (e.g., whether the traffic flow of logistics vehicles and private cars changes due to certain common factors). Then, the similarity of traffic volume is calculated based on the weighted values ​​of the first and second features (e.g., comparing the similarity of the overall traffic volume of logistics vehicles and private cars). The similarity of traffic change is calculated based on the dispersion values ​​of the first and second features (e.g., comparing the similarity of traffic changes in different time periods). The similarity of traffic distribution is calculated based on the dispersion values ​​of the first and second features and the covariance value (e.g., comparing the similarity of traffic distribution in different regions). Finally, the distance of the first feature is calculated based on the similarity of traffic volume, the similarity of traffic change, and the similarity of traffic distribution.

[0027] The traffic flow data of each second candidate sample are arranged in descending order according to the first feature distance. A candidate sample traffic flow data sequence is generated based on the traffic flow data of the second candidate sample of the first set size (e.g., the first 10) in the descending sort result and the traffic flow data of the first candidate sample.

[0028] If it is not the first round of data optimization, for each second candidate sample traffic flow data, obtain its second feature distance with each of the obtained candidate sample traffic flow data, and calculate the weighted feature distance of each second feature distance. For example, for a residential area private car traffic flow data that has not been selected (as the second candidate sample traffic flow data), calculate its second feature distance with the candidate sample traffic flow data obtained in previous rounds of optimization (such as the previously selected industrial park logistics vehicle traffic flow data, tourist attraction tour bus traffic flow data, etc.), comprehensively consider the weights of different rounds to obtain the weighted feature distance, and output the second candidate sample traffic flow data with the largest weighted feature distance as the first candidate sample traffic flow data.

[0029] The server loads candidate sample traffic flow data sequences into auxiliary sample traffic flow data sequences until the training size of the auxiliary sample traffic flow data sequences meets a first set size (e.g., reaching 1000 data points), then generates student sample traffic flow data sequences. Each student sample traffic flow data in this sequence carries third sample charging demand data, which reflects the charging demand information contained in the combined teacher sample traffic flow data and auxiliary sample traffic flow data.

[0030] Step S140: Based on the student sample traffic flow data sequence, iteratively learn the network parameters of the student charging pile scheduling network until the network convergence requirement is met, and then generate the target charging pile scheduling network.

[0031] In this embodiment, for each student sample traffic flow data point in the student sample traffic flow data sequence, for example, if the student sample traffic flow data pertains to traffic flow data in a newly developed area on the outskirts of the city during a certain period, this student sample traffic flow data is loaded into the teacher charging pile scheduling network to generate the first charging demand prediction data. Assuming that the teacher charging pile scheduling network predicts, based on the traffic flow of construction vehicles and commuter vehicles in the newly developed area, that a certain number of charging piles need to be added to a certain parking lot to meet the charging demand, this is the first charging demand prediction data.

[0032] Simultaneously, the student sample traffic flow data is loaded into the auxiliary charging pile scheduling network to generate second charging demand prediction data. Based on the sparse overall traffic flow characteristics of the newly developed area, the auxiliary charging pile scheduling network may predict that charging demand is relatively dispersed, eliminating the need for large-scale centralized installation of charging piles; this is the second charging demand prediction data.

[0033] The student sample traffic flow data is then loaded into the student charging station scheduling network to generate third charging demand prediction data. It is assumed that the initial charging demand prediction from the student charging station scheduling network differs from the predictions from the teacher charging station scheduling network and the auxiliary charging station scheduling network.

[0034] Then, the server determines the global network error parameters based on the first loss function value between the first and third charging demand prediction data, the second loss function value between the second and third charging demand prediction data, and the third loss function value between the third charging demand prediction data and the third sample charging demand data.

[0035] When calculating the first and second loss function values, we take construction vehicles in newly developed areas as the first charging pile dispatching subject and commuter vehicles as the non-charging pile dispatching subject as an example. The server obtains the first dynamic flow characteristics (such as vehicle speed, parking frequency, etc.) of the first charging pile dispatching subject (construction vehicles) and the second dynamic flow characteristics (such as commuter vehicle round-trip routes, parking time, etc.) of the non-charging pile dispatching subject (commuter vehicles) from the student sample traffic flow data, and obtains the third dynamic flow characteristics (such as the working mode of special vehicles) of the second charging pile dispatching subject (assuming it is another special vehicle type) from the student sample traffic flow data. We determine the third feature distance between the third charging demand prediction data and the first charging demand prediction data (for example, by comparing the differences in the predicted number and location of charging piles), and determine the fourth feature distance between the third charging demand prediction data and the second charging demand prediction data. Based on the first charging demand prediction data, obtain the number of first entities (e.g., the number of construction vehicles) corresponding to the first charging pile dispatching entity and the number of second entities (e.g., the number of commuter vehicles) corresponding to non-charging pile dispatching entities in the student sample traffic flow data. Based on the second charging demand prediction data, obtain the number of third entities (e.g., the number of special vehicles) corresponding to the second charging pile dispatching entity in the student sample traffic flow data. Calculate the fourth loss function value (e.g., calculated based on factors such as the degree to which the charging demand of construction vehicles is not met) between the first and third charging demand prediction data for the first charging pile dispatching entity, based on the third feature distance, the first dynamic flow characteristic, and the number of first entities. Also, calculate the fifth loss function value (e.g., calculated based on factors such as the impact of commuter vehicles on charging facilities) between the first and third charging demand prediction data for non-charging pile dispatching entities, based on the third feature distance, the third dynamic flow characteristic, and the number of second entities. Calculate the first loss function value based on the fourth and fifth loss function values, and calculate the second loss function value based on the fourth feature distance, the second dynamic flow characteristic, and the number of third entities.

[0036] Obtain the first weight value corresponding to the first loss function value (e.g., determined based on the importance of the first charging pile scheduling entity), obtain the second weight value corresponding to the second loss function value (e.g., determined based on the importance of the second charging pile scheduling entity), and obtain the third weight value corresponding to the third loss function value (e.g., determined based on the accuracy and importance of the third sample charging demand data). Determine the first error parameter of the student charging pile scheduling network based on the first weight value, the first loss function value, the second weight value, and the second loss function value, and determine the second error parameter of the student charging pile scheduling network based on the third weight value and the third loss function value. Determine the global network error parameter based on the first and second error parameters.

[0037] The neuron weights of the student charging station scheduling network are updated based on the global network error parameters. The updated student charging station scheduling network is then used as the student charging station scheduling network for the next round of network parameter learning. The server repeats this process until the network convergence requirements are met (e.g., the loss function value is lower than a certain set threshold, or the change in network parameters is less than a certain set value). At this point, the target charging station scheduling network is generated.

[0038] Step S150: Based on the target charging pile scheduling network, make a decision on the target traffic flow data of any target traffic monitoring area, generate target charging demand data, and schedule charging piles for the traffic participants in the target traffic monitoring area according to the target charging demand data.

[0039] In this embodiment, the server selects a target traffic monitoring area, such as a central urban business district. The target traffic flow data includes traffic information from various traffic participants (such as private cars, taxis, buses, etc.). The server inputs the target traffic flow data into the target charging pile scheduling network to generate target charging demand data.

[0040] Assuming target traffic flow data shows a significant increase in taxi traffic and peak bus operations during weekday evening rush hour, the target charging station dispatch network generates target charging demand data based on this data. This data may include the need to add temporary charging stations at locations with frequent taxi pick-up and drop-off (such as entrances to large shopping malls or near office buildings), and for buses, the need to adjust the charging power of charging stations at bus hubs to meet fast charging demands. It also predicts a significant increase in charging demand across the entire commercial area during the evening rush hour.

[0041] The server then preprocesses the target charging demand data for the target traffic monitoring area to extract key charging demand data. For example, the charging demand period is determined to be the evening peak on weekdays (17:00-19:00), the charging demand hotspots are the entrances of large shopping malls, near office buildings, and bus hubs, and the charging demand type ratio is 50% for taxis, 30% for buses, and 20% for private cars. The predicted charging demand is a total of 100 charging slots (calculated based on the charging demand of different types of vehicles).

[0042] Based on this key charging demand data, a charging resource assessment is conducted, generating a charging resource assessment report. The server queries the distribution of charging piles within the target traffic monitoring area and finds 20 charging piles near the shopping mall, 10 near the office building, and 30 near the bus hub. The availability status is that some charging piles are in use, while others are idle. Charging power varies, with both fast and slow charging piles available. Regarding expansion potential, the areas near the shopping mall and office building have limited space, while the bus hub has some expansion potential.

[0043] Based on the charging resource assessment report and key charging demand data, a charging pile scheduling strategy is constructed. The server determines the charging piles that need to be scheduled, for example, allocating 5 charging piles from other areas with low charging demand to the entrance of the shopping mall, and adjusting the charging power of 10 charging piles at the bus hub to fast charging mode; the scheduling time is to complete the scheduling before the evening peak; the scheduling location is to accurately place the allocated charging piles at the designated location at the entrance of the shopping mall; the status of the scheduled charging piles is that they are turned on and can be used normally.

[0044] According to the charging pile scheduling strategy, the server sends scheduling instructions to the charging pile management system. These instructions include the identifier of each charging pile, the scheduling time (e.g., scheduling starts at 16:30), the scheduling location (e.g., the specific coordinates at the mall entrance), and the charging status after scheduling (e.g., fast charging). Upon receiving the scheduling instructions, the charging pile management system controls the charging piles to perform corresponding operations, such as moving a designated charging pile to a specific location at the mall entrance, adjusting the 10 charging piles at the bus hub to fast charging power, or changing the charging status to usable.

[0045] After the charging pile scheduling operation is executed, the server monitors the scheduling effect of the charging piles in real time. The server collects real-time charging data of the charging piles (such as charging current, voltage, charging time, etc. for each charging pile), traffic flow data (such as changes in the flow of taxis, buses, and private cars during the evening peak hours), and user feedback data (such as whether taxi drivers find charging more convenient, whether buses can charge in time, etc.) to evaluate the actual scheduling effect of the charging piles. If it is found that although charging piles have been added at the entrance of the shopping mall, taxis cannot reach the charging piles in time due to traffic congestion, and the scheduling effect does not meet the expected conditions, the charging pile scheduling strategy is adjusted, such as replanning the placement of charging piles to avoid traffic congestion sections, and the scheduling operation is executed again.

[0046] Meanwhile, the server optimizes and learns charging pile scheduling strategies based on historical charging pile scheduling data (previous scheduling experience data in this or other areas), real-time charging data, traffic flow data, and user feedback data. Machine learning algorithms analyze the correlations and patterns in this data; for example, it discovers that the charging demand of buses during specific time periods is closely related to the bus route's operating schedule, and the charging demand of private cars is related to shopping mall opening hours and promotional activities. Key factors affecting the effectiveness of charging pile scheduling, such as traffic congestion and the rationality of charging facility layout, are extracted, and the charging pile scheduling strategy is intelligently adjusted based on these key factors to generate an optimized charging pile scheduling strategy. The optimized charging pile scheduling strategy is input into the charging pile scheduling system to update the system's scheduling parameters and strategy library, so that future charging pile scheduling decisions can more accurately and efficiently meet charging demand.

[0047] Based on the above steps, this embodiment of the application effectively improves the intelligence and accuracy of charging pile scheduling by constructing and integrating a teacher charging pile scheduling network and an auxiliary charging pile scheduling network. The two networks, trained using teacher sample traffic flow data sequences and auxiliary sample traffic flow data sequences respectively, not only fully consider the characteristics and differences of different traffic monitoring areas, but also further improve the adaptability and prediction accuracy to traffic flow changes through the student charging pile scheduling network generated through iterative learning. The final target charging pile scheduling network can quickly determine reasonable charging demand data based on the actual traffic flow data of any target traffic monitoring area, and achieve efficient charging pile scheduling accordingly. This effectively alleviates the problem of uneven distribution of charging pile resources, improves the convenience and user experience of charging services, and promotes the overall operational efficiency of the transportation system and the rationality of energy utilization.

[0048] In one possible implementation, the process of learning network parameters in each round specifically includes:

[0049] Step S141: For each student sample traffic flow data, the student sample traffic flow data is loaded into the teacher charging pile scheduling network to generate first charging demand prediction data. The student sample traffic flow data is loaded into the auxiliary charging pile scheduling network to generate second charging demand prediction data. The student sample traffic flow data is loaded into the student charging pile scheduling network to generate third charging demand prediction data.

[0050] Step S142: Determine the global network error parameters based on the first loss function value between the first charging demand prediction data and the third charging demand prediction data, the second loss function value between the second charging demand prediction data and the third charging demand prediction data, and the third loss function value between the third charging demand prediction data and the third sample charging demand data.

[0051] Step S143: Update the neuron weight information of the student charging pile scheduling network according to the global network error parameters, and use the updated student charging pile scheduling network as the student charging pile scheduling network corresponding to the next round of network parameter learning process.

[0052] In this embodiment, the operation is carried out using data from different traffic monitoring areas in the city as examples during each round of network parameter learning. For each student sample traffic flow data in the student sample traffic flow data sequence, such as the traffic flow data of a newly developed area on the edge of the city on a certain weekday, the server first loads it into the teacher charging pile scheduling network. In this area, the traffic flow of construction vehicles and commuter vehicles are important factors. The teacher charging pile scheduling network generates the first charging demand prediction data based on the relationship between traffic flow and charging demand in various areas of the city that it has learned before. For example, based on factors such as the working hours and parking patterns of construction vehicles and the round-trip routes of commuter vehicles, it predicts that a certain number of charging piles need to be set up near several construction sites in the newly developed area, and an appropriate number of charging piles are also needed in areas where commuter vehicles park more frequently to meet the charging demand.

[0053] Next, the server loads the same student sample traffic flow data into the auxiliary charging station scheduling network. Based on its learned relationships between different traffic characteristics and charging demand, the auxiliary charging station scheduling network generates second charging demand prediction data. Because the auxiliary charging station scheduling network takes into account the relatively sparse overall traffic flow in newly developed areas, it may predict that charging facilities do not need to be too concentrated, but rather dispersed across different small parking areas, and the number of charging stations may be less than predicted by the teacher charging station scheduling network.

[0054] The server then loads this student sample traffic flow data into the student charging station scheduling network to generate the third charging demand prediction data. The initial prediction data generated by the student charging station scheduling network may differ from the previous two because it is still in the process of learning and adjusting.

[0055] Subsequently, the server begins to determine global network error parameters based on the first loss function value between the first and third charging demand prediction data, the second loss function value between the second and third charging demand prediction data, and the third loss function value between the third charging demand prediction data and the third sample charging demand data. When calculating the first loss function value, construction vehicles are used as the primary subject for charging pile scheduling, while commuter vehicles are used as the non-charging pile scheduling subject. The server acquires the first dynamic flow characteristics of construction vehicles in the student sample traffic flow data, such as average vehicle speed and parking interval time, and the second dynamic flow characteristics of commuter vehicles, such as peak-hour concentrated parking areas. Simultaneously, it acquires the third dynamic flow characteristics of other potentially special vehicles (assuming they are test vehicles used for model validation), such as special charging time requirements. The third feature distance between the third and first charging demand prediction data is determined, for example, by comparing differences in charging pile location and quantity predictions. The fourth feature distance between the third and second charging demand prediction data is determined, which may be due to differences in the dispersion of charging facility layout. Based on the first charging demand prediction data, the number of construction vehicles (the first entity) and commuter vehicles (the second entity) in the student sample traffic flow data are obtained. Based on the second charging demand prediction data, the number of special vehicles (the third entity) is obtained. A fourth loss function value is calculated between the first and third charging demand prediction data for construction vehicles, based on the third feature distance, the first dynamic flow feature, and the number of the first entity. This loss function is calculated based on the impact of construction vehicles not being able to charge at the predicted charging station locations. A fifth loss function value is calculated between the first and third charging demand prediction data for commuter vehicles, based on the third feature distance, the third dynamic flow feature, and the number of the second entity. A first loss function value is calculated based on the fourth and fifth loss function values, and a second loss function value is calculated based on the fourth feature distance, the second dynamic flow feature, and the number of the third entity.

[0056] After determining the first, second, and third loss function values, the server obtains the first weight value corresponding to the first loss function value. This weight value is determined based on the importance of the first charging pile scheduling entity (construction vehicles) in the entire transportation system and its impact on charging demand. The second weight value corresponding to the second loss function value is also obtained, based on the relevant importance of the specific vehicles. The third weight value corresponding to the third loss function value is obtained, taking into account the importance of the accuracy of the third sample charging demand data to the entire learning process. The first error parameter of the student charging pile scheduling network is determined based on the first weight value, the first loss function value, the second weight value, and the second loss function value. The second error parameter of the student charging pile scheduling network is determined based on the third weight value and the third loss function value. Finally, the global network error parameter is determined based on the first and second error parameters.

[0057] Finally, the server updates the neuron weights of the student charging station scheduling network based on the global network error parameters. For example, if the global network error parameters indicate a significant deviation between the predicted charging demand data and the actual demand data (reflected by the teacher, auxiliary networks, and sample data), the server adjusts the connection weights between neurons. This allows the student charging station scheduling network to more accurately predict charging demand when processing similar student sample traffic flow data in the future. The updated student charging station scheduling network is then used as the corresponding network for the next round of network parameter learning, allowing for continuous optimization in subsequent learning rounds and gradually approaching an accurate charging demand prediction model.

[0058] In one possible implementation, step S130 includes:

[0059] Multiple teacher sample traffic flow data are obtained from the teacher sample traffic flow data sequence, and the multiple teacher sample traffic flow data are loaded into the auxiliary sample traffic flow data sequence to generate the student sample traffic flow data sequence.

[0060] In one possible implementation, the step of obtaining multiple teacher sample traffic flow data from the teacher sample traffic flow data sequence and loading the multiple teacher sample traffic flow data into an auxiliary sample traffic flow data sequence to generate a student sample traffic flow data sequence includes:

[0061] Step S131: Perform multiple rounds of data optimization on the teacher sample traffic flow data sequence to generate candidate sample traffic flow data sequences corresponding to each data optimization.

[0062] Step S132: Load the candidate sample traffic flow data sequence into the auxiliary sample traffic flow data sequence until the training scale in the auxiliary sample traffic flow data sequence meets the first set scale, then generate the student sample traffic flow data sequence.

[0063] Specifically, step S131 includes:

[0064] Step S1311: Obtain first candidate sample traffic flow data from the teacher sample traffic flow data sequence, and calculate the first feature distance between each second candidate sample traffic flow data in the teacher sample traffic flow data sequence and the first candidate sample traffic flow data based on the first candidate sample traffic flow data. The second candidate sample traffic flow data are other sample traffic flow data in the teacher sample traffic flow data sequence besides the first candidate sample traffic flow data.

[0065] Step S1312: Arrange the traffic flow data of each second candidate sample in descending order according to the first feature distance, and generate the candidate sample traffic flow data sequence based on the first second candidate sample traffic flow data of the second predetermined size in the descending sort result and the first candidate sample traffic flow data.

[0066] In this embodiment, firstly, the server obtains multiple teacher sample traffic flow data from the teacher sample traffic flow data sequence, and then loads these data into the auxiliary sample traffic flow data sequence to generate the student sample traffic flow data sequence.

[0067] This process involves multiple rounds of data optimization on the teacher sample traffic flow data sequence to generate candidate sample traffic flow data sequences for each round of optimization. Taking the urban traffic system as an example, the teacher sample traffic flow data sequence covers traffic flow data from multiple primary traffic monitoring areas in the city, including the city center business district, large industrial parks, and residential areas.

[0068] During the multi-round data optimization process, the first round begins with the server retrieving first-candidate sample traffic flow data from the teacher sample traffic flow data sequence. For example, the server selects traffic flow data from the morning rush hour on a weekday from the teacher sample traffic flow data in the city's central business district as the first-candidate sample traffic flow data. This data includes various traffic characteristics such as traffic flow information, driving speed, and parking frequency of different types of vehicles (e.g., private cars, taxis, buses) in the business district during that time period. It also carries first-candidate charging demand data, which reflects the charging needs of different vehicle types during that time period, such as the battery consumption of taxis after the morning rush hour and the location of charging stations they require.

[0069] Next, based on this first candidate sample traffic flow data, the server calculates the first feature distance between each second candidate sample traffic flow data point and the first candidate sample traffic flow data in the teacher sample traffic flow data sequence. Here, the second candidate sample traffic flow data refers to the traffic flow data points in the teacher sample traffic flow data sequence other than the first candidate sample traffic flow data. Taking the traffic flow data of a large industrial park during a weekday as an example, the server obtains the first feature weighted value of the first candidate sample traffic flow data. This weighted value is determined based on multiple characteristic factors of the traffic flow data during the morning rush hour in the city center's commercial district. For example, the weights are determined based on factors such as the proportion of private car trips, taxi occupancy rates, and bus occupancy rates, and then the first feature weighted value is derived by combining these factors. Simultaneously, the server obtains the first feature dispersion value of the first candidate sample traffic flow data, for example, based on the fluctuations in traffic flow during the morning rush hour in the commercial district at different time periods (such as 8:00-8:30, 8:30-9:00, etc.). For the traffic flow data of the second candidate sample, its second feature weighting value is also obtained. This weighting value is determined based on factors such as the different types of logistics vehicles, their load conditions, and the proportion of commuter vehicles in the industrial park. The second feature dispersion value is also determined based on the fluctuations in traffic flow at different times of the day in the industrial park. Then, the covariance value between the traffic flow data of the first and second candidate samples is determined. This may involve analyzing whether there are certain common influencing factors between the morning rush hour traffic in the commercial area and the daytime traffic in the industrial park, such as the degree of common influence of the overall urban weather conditions and large-scale events on the traffic flow of both.

[0070] For this second candidate sample traffic flow data, the similarity of traffic flow scale is calculated based on the weighted values ​​of the first and second features. For example, comparing the total vehicle flow scale during the morning peak hours in a commercial area with the total vehicle flow scale during the daytime in an industrial park, considering the different weights of vehicle types, the similarity of traffic flow scale is calculated. The similarity of traffic flow change is calculated based on the dispersion values ​​of the first and second features. For example, comparing the variation range of traffic flow during the morning peak hours in a commercial area with the variation range of traffic flow during the daytime in an industrial park. The similarity of traffic flow distribution is calculated based on the dispersion values ​​of the first and second features and the covariance value. For example, analyzing the similarity of traffic flow distribution in different areas of a commercial area (such as around shopping malls and office buildings) with the similarity of traffic flow distribution in different areas of an industrial park (such as production areas and office areas), and calculating the distance of the first feature after comprehensively considering the above factors.

[0071] The server performs this calculation on all second-candidate traffic flow data, and then arranges the second-candidate traffic flow data in descending order based on the first feature distance. Assuming the second set size is 10, the server generates a sequence of candidate traffic flow data based on the first 10 second-candidate traffic flow data in the descending order and the first-candidate traffic flow data. This sequence contains traffic flow data that are highly correlated with the first-candidate traffic flow data in multiple features (determined by the first feature distance).

[0072] For data optimization beyond the first round, the server still needs to obtain first candidate sample traffic flow data from the teacher sample traffic flow data sequence. At this point, for each second candidate sample traffic flow data, the server calculates its second feature distance with each of the obtained candidate sample traffic flow data. For example, for traffic flow data from a residential area on a weekday evening (as second candidate sample traffic flow data), the server calculates its second feature distance with the candidate sample traffic flow data obtained from previous optimization rounds (such as the morning rush hour traffic flow data from commercial areas and the daytime traffic flow data from industrial parks). This calculation process also needs to consider various factors, such as the feature weighting values, dispersion values, and covariance values ​​of traffic flow in different areas. Then, the weighted feature distance of each second feature distance is calculated; this weighted feature distance is obtained by comprehensively considering the weights of the optimization results from previous rounds. Finally, the server outputs the second candidate sample traffic flow data with the largest weighted feature distance as the first candidate sample traffic flow data.

[0073] After each round of data optimization generates candidate sample traffic flow data sequences, the server loads these sequences into the auxiliary sample traffic flow data sequences. The auxiliary sample traffic flow data sequences correspond to the second traffic monitoring area in the city, such as newly developed areas on the city's edge or tourist attractions in the suburbs. The server continues this loading process until the training size of the auxiliary sample traffic flow data sequences meets a first predetermined size. For example, if the first predetermined size is 1000 data points, when the number of data points in the auxiliary sample traffic flow data sequences reaches 1000, a student sample traffic flow data sequence is generated. Each data point in this student sample traffic flow data sequence carries third sample charging demand data. This data integrates the charging demand information from the teacher sample traffic flow data and the auxiliary sample traffic flow data, providing a rich and representative data foundation for the subsequent learning of the student charging station scheduling network.

[0074] In one possible implementation, if the first candidate sample traffic flow data is the sample traffic flow data corresponding to the first round of data optimization, then step S1311 includes:

[0075] Randomly select one teacher sample traffic flow data from the teacher sample traffic flow data sequence and output it as the first candidate sample traffic flow data.

[0076] If the first candidate sample traffic flow data is not the sample traffic flow data corresponding to the first round of data optimization, then step S1311 includes:

[0077] For each second candidate sample traffic flow data, the second feature distance between the second candidate sample traffic flow data and each obtained candidate sample traffic flow data is obtained, and the weighted feature distance of each second feature distance is calculated.

[0078] The traffic flow data of the second candidate sample with the largest weighted feature distance among the various second feature distances is output as the traffic flow data of the first candidate sample.

[0079] In one possible implementation, calculating the first feature distance between each second candidate sample traffic flow data and the first candidate sample traffic flow data in the teacher sample traffic flow data sequence includes:

[0080] Step S1311-1: Obtain the first feature weighting value and the first feature dispersion value of the first candidate sample traffic flow data, and obtain the second feature weighting value and the second feature dispersion value of each second candidate sample traffic flow data, and determine the covariance value between the first candidate sample traffic flow data and each second candidate sample traffic flow data.

[0081] Step S1311-2: For each second candidate sample traffic flow data, calculate the similarity of the traffic flow scale between the second candidate sample traffic flow data and the first candidate sample traffic flow data based on the first feature weighting value and the second feature weighting value.

[0082] Step S1311-3: Calculate the similarity of traffic flow changes between the second candidate sample traffic flow data and the first candidate sample traffic flow data based on the first feature dispersion value and the second feature dispersion value.

[0083] Step S1311-4: Calculate the similarity of traffic flow distribution between the second candidate sample traffic flow data and the first candidate sample traffic flow data based on the first feature dispersion value, the second feature dispersion value, and the covariance value.

[0084] Step S1311-5: Calculate the first feature distance between the second candidate sample traffic flow data and the first candidate sample traffic flow data based on the traffic flow scale similarity, the traffic flow change similarity, and the traffic flow distribution similarity.

[0085] In this embodiment, when the first candidate sample traffic flow data is the sample traffic flow data corresponding to the first round of data optimization, the server arbitrarily selects one teacher sample traffic flow data from the teacher sample traffic flow data sequence and outputs it as the first candidate sample traffic flow data. For example, in a teacher sample traffic flow data sequence containing traffic flow data from multiple areas of a city, this sequence covers traffic flow information from different areas such as the city center business district, large industrial parks, and residential areas, as well as the corresponding first sample charging demand data. The server may directly select the traffic flow data of a weekday morning rush hour from the traffic flow data of the city center business district as the first candidate sample traffic flow data. This morning rush hour data contains various traffic information, such as the number of vehicles of various types, such as private cars, taxis, and buses, the speed of vehicles on different road sections, the number of stops, and other traffic characteristics, as well as the first sample charging demand data generated based on these traffic conditions, such as the power consumption of taxis after operating during this period and information on which areas may need charging.

[0086] When the first candidate traffic flow data is not the traffic flow data corresponding to the first round of data optimization, the operation becomes relatively more complex. For each second candidate traffic flow data point, the server obtains the second feature distance between that second candidate traffic flow data point and all other obtained candidate traffic flow data points, and calculates the weighted feature distance of each second feature distance. Taking urban traffic data as an example, suppose several rounds of data optimization have already been performed, and some candidate traffic flow data points exist, such as the morning rush hour traffic flow data of the city center business district selected in the first round, and the daytime traffic flow data of a large industrial park selected in the second round. Now, new second candidate traffic flow data needs to be processed, such as the traffic flow data of a residential area on a weekday evening. The server will calculate the second feature distance between this residential area's evening traffic flow data and each previously obtained candidate traffic flow data point (morning rush hour in the business district, daytime in the industrial park, etc.). When calculating this distance, various factors need to be considered, such as the proportion of different vehicle types in the traffic flow data, vehicle speed distribution, and parking frequency. For each previous candidate traffic flow data point, the server assigns different weights based on its feature importance. For example, a commercial area with a high proportion of taxi traffic during the morning rush hour has a higher weight, while an industrial park with high weight for logistics vehicles during the day has a higher weight. Then, the server calculates the second feature distance between this data and the evening traffic flow data of residential areas. Next, it calculates the weighted feature distance of each second feature distance, taking into account the importance of each candidate traffic flow data point in previous optimization rounds. Finally, the second candidate traffic flow data point with the largest weighted feature distance is output as the first candidate traffic flow data point. For instance, after comparing traffic flow data from multiple residential areas at different times or from other areas, the server determines the weekday evening traffic flow data of a specific residential area as the new first candidate traffic flow data point.

[0087] When calculating the first feature distance between each second candidate sample traffic flow data and the first candidate sample traffic flow data in the teacher sample traffic flow data sequence, the first feature weighted value and first feature dispersion value of the first candidate sample traffic flow data are first obtained, and the second feature weighted value and second feature dispersion value of each second candidate sample traffic flow data are also obtained to determine the covariance value between the first candidate sample traffic flow data and each second candidate sample traffic flow data. For example, if the first candidate sample traffic flow data is the daytime traffic flow data of a large industrial park on a weekday, in which logistics vehicles account for a large proportion, its first feature weighted value will be determined based on factors such as the load type and number ratio of logistics vehicles, and its first feature dispersion value will be determined based on the traffic flow fluctuations of logistics vehicles entering and leaving the park at different times during the day. For a certain second candidate sample traffic flow data, such as the traffic flow data of a city center business district during the evening peak on a weekday, its second feature weighted value will be determined based on factors such as the proportion of private cars, taxis, and buses during the evening peak in the business district, and its second feature dispersion value will be determined based on the traffic flow fluctuations at different times during the evening peak. The degree of covariance considers whether there are common influencing factors between the two, such as the degree of common impact of large-scale events in the city on logistics and transportation in industrial parks and traffic flow in commercial areas.

[0088] Next, for each second candidate sample traffic flow data, the similarity of traffic flow scale between the second candidate sample traffic flow data and the first candidate sample traffic flow data is calculated based on the weighted values ​​of the first and second features. For example, comparing the total traffic flow scale of logistics vehicles in an industrial park during the daytime with the total traffic flow scale of vehicles during the evening peak in a commercial area, the similarity of traffic scale is calculated considering the weighted values ​​of each vehicle type. The similarity of traffic change between the second and first candidate sample traffic flow data is calculated based on the dispersion values ​​of the first and second features. For example, comparing the variation range of logistics vehicle traffic flow in an industrial park during the daytime at different times with the variation range of traffic flow during the evening peak in a commercial area at the corresponding times. The similarity of traffic distribution between the second and first candidate sample traffic flow data is calculated based on the dispersion values ​​of the first and second features and the covariance value. For example, analyzing the similarity of logistics vehicle traffic distribution in different areas of an industrial park (production area, office area, etc.) with the evening peak traffic flow distribution in different areas of a commercial area (around shopping malls, around office buildings, etc.). Finally, the first feature distance between the second candidate sample traffic flow data and the first candidate sample traffic flow data is calculated based on the similarity of traffic volume, traffic change, and traffic distribution. This first feature distance comprehensively reflects the degree of difference between the two traffic flow data in multiple dimensions, providing an important basis for subsequent data screening and optimization. It helps the server to more accurately construct student sample traffic flow data sequences, thereby providing a high-quality data foundation for the learning and optimization of the charging pile scheduling network.

[0089] In one possible implementation, step S142 includes:

[0090] Step S1421: Obtain the first weight value corresponding to the first loss function value, obtain the second weight value corresponding to the second loss function value, and obtain the third weight value corresponding to the third loss function value.

[0091] Step S1422: Determine the first error parameter of the student charging pile scheduling network based on the first weight value, the first loss function value, the second weight value, and the second loss function value; and determine the second error parameter of the student charging pile scheduling network based on the third weight value and the third loss function value.

[0092] Step S1423: Determine the global network error parameters based on the first error parameters and the second error parameters.

[0093] In one possible implementation, the steps for determining the first loss function value and the second loss function value include:

[0094] The system obtains the first dynamic flow characteristics of the first charging pile scheduling entity and the second dynamic flow characteristics of the non-charging pile scheduling entity in the student sample traffic flow data, and obtains the third dynamic flow characteristics of the second charging pile scheduling entity in the student sample traffic flow data.

[0095] A third feature distance is determined between the third charging demand prediction data and the first charging demand prediction data, and a fourth feature distance is determined between the third charging demand prediction data and the second charging demand prediction data.

[0096] Based on the first charging demand prediction data, obtain the number of first entities corresponding to the first charging pile scheduling entity and the number of second entities corresponding to the non-charging pile scheduling entity in the student sample traffic flow data. Based on the second charging demand prediction data, obtain the number of third entities corresponding to the second charging pile scheduling entity in the student sample traffic flow data.

[0097] Based on the third feature distance, the first dynamic flow feature, and the first number of subjects, calculate a fourth loss function value for the first charging pile scheduling subject between the first charging demand prediction data and the third charging demand prediction data; and based on the third feature distance, the third dynamic flow feature, and the second number of subjects, calculate a fifth loss function value for the non-charging pile scheduling subject between the first charging demand prediction data and the third charging demand prediction data.

[0098] The first loss function value is calculated based on the fourth loss function value and the fifth loss function value, and the second loss function value is calculated based on the fourth feature distance, the second dynamic flow feature, and the third number of subjects.

[0099] In this embodiment, taking the scheduling of charging piles in urban transportation as an example, the determination of these weight values ​​is based on the importance considerations of different factors. Assume the first charging pile scheduling entity is taxis in the city center business district, the non-charging pile scheduling entity is private cars in the area, and the second charging pile scheduling entity is buses. If the accuracy of taxi charging demand prediction has a significant impact on the entire network, then the first weight value corresponding to the first loss function value will be relatively high. This weight value may be determined based on factors such as the proportion of taxis in urban transportation, their long operating hours, and frequent charging needs. For buses, if their charging demand pattern in the entire transportation system is relatively unique and has a certain influence on the overall charging facility layout, then the second weight value corresponding to the second loss function value will also be set according to its importance. The third weight value corresponding to the third loss function value considers more the importance of the accuracy of the third sample charging demand data itself to the entire learning process. For example, if the third sample charging demand data is based on accurate traffic flow statistics and actual charging records, its weight value may be high.

[0100] Next, the server determines the first error parameter of the student charging station scheduling network based on the first weight value, the first loss function value, the second weight value, and the second loss function value, and determines the second error parameter based on the third weight value and the third loss function value. For example, when calculating the first error parameter, if the first weight value is 0.4, the first loss function value is 0.3 (this value is obtained through subsequent detailed calculations), the second weight value is 0.3, and the second loss function value is 0.2, then the first error parameter may obtain a specific value through a certain calculation method (such as weighted summation). For the second error parameter, assuming the third weight value is 0.3 and the third loss function value is 0.4, the corresponding value is also obtained through specific calculations.

[0101] Then, the global network error parameter is determined based on the first and second error parameters. This global network error parameter comprehensively reflects the degree of difference between the entire student charging pile scheduling network and the teacher charging pile scheduling network, the auxiliary charging pile scheduling network, and the actual third sample charging demand data in terms of predicting charging demand. It is an important basis for adjusting the neuron weight information of the student charging pile scheduling network.

[0102] The determination of the first and second loss function values ​​also involves detailed steps. The server first acquires the first dynamic flow characteristics of the first charging pile dispatching entity (e.g., taxis) and the second dynamic flow characteristics of non-charging pile dispatching entities (e.g., private cars) from the student sample traffic flow data. It also acquires the third dynamic flow characteristics of the second charging pile dispatching entity (e.g., buses) from the student sample traffic flow data. The first dynamic flow characteristics of taxis may include factors such as average speed, empty-running rate, and changes in driving routes after picking up passengers at different time periods. The second dynamic flow characteristics of private cars may include information such as peak travel times, average parking time, and parking location distribution. The third dynamic flow characteristics of buses may involve factors such as the busyness of bus routes, stop times, and occupancy rates.

[0103] Then, the server determines the third characteristic distance between the third charging demand forecast data and the first charging demand forecast data, and the fourth characteristic distance between the third charging demand forecast data and the second charging demand forecast data. Taking a city center business district as an example, if the third charging demand forecast data predicts that 5 charging piles need to be set up in a certain area, while the first charging demand forecast data predicts that 8 charging piles need to be set up, then there is a difference in the dimension of the number of charging piles. There may also be differences in other aspects such as the predicted location of charging piles. The third characteristic distance is calculated by taking into account these factors. The calculation of the fourth characteristic distance is similar, comparing the differences between the third and second charging demand forecast data in terms of the number and location of charging piles.

[0104] Next, based on the first charging demand prediction data, the number of first entities corresponding to the first charging pile dispatching entity (taxi) and the number of second entities corresponding to non-charging pile dispatching entities (private cars) in the student sample traffic flow data are obtained. Based on the second charging demand prediction data, the number of third entities corresponding to the second charging pile dispatching entity (bus) in the student sample traffic flow data is obtained. For example, within a specific time period, according to the first charging demand prediction data, the number of taxis in the area is 100 (number of first entities), the number of private cars is 500 (number of second entities), and according to the second charging demand prediction data, the number of buses is 30 (number of third entities).

[0105] Then, based on the third feature distance, the first dynamic flow feature, and the number of the first subject, a fourth loss function value is calculated for the first charging pile dispatch subject (taxi) between the first and third charging demand prediction data. Similarly, a fifth loss function value is calculated for the non-charging pile dispatch subject (private car) between the first and third charging demand prediction data, based on the third feature distance, the third dynamic flow feature, and the number of the second subject. For example, if the third feature distance is large, the first dynamic flow feature of taxis shows a low empty-running rate (meaning longer operating time and more urgent charging demand), and the number of taxis (number of the first subject) is large, then the fourth loss function value may be large, reflecting a significant difference between the first and third charging demand prediction data in terms of taxi charging demand prediction. The calculation of the fifth loss function value for private cars is similar, considering factors such as the third feature distance, the second dynamic flow feature of private cars, and the number of private cars (number of the second subject).

[0106] Finally, the first loss function value is calculated based on the fourth and fifth loss function values, and the second loss function value is calculated based on the fourth feature distance, the second dynamic flow feature, and the third subject quantity. For example, if the fourth loss function value is 0.2 and the fifth loss function value is 0.1, then the first loss function value may be obtained as 0.3 through some calculation method (such as summation or weighted summation). For the second loss function value, a specific value is obtained by similar calculation based on the fourth feature distance, the third dynamic flow feature of the bus, and the number of buses (the third subject quantity). The accurate calculation of these loss function values ​​is crucial for determining the global network error parameters, which in turn affects the learning and optimization process of the student charging pile scheduling network, enabling it to more accurately predict charging demand and perform effective charging pile scheduling.

[0107] In one possible implementation, step S150 specifically includes:

[0108] Step S151: Preprocess the target charging demand data of the target traffic monitoring area to extract key charging demand data, which includes charging demand time periods, charging demand hotspot areas, charging demand type ratios, and predicted charging demand amounts.

[0109] Step S152: Based on the key charging demand data, perform a charging resource assessment and generate a charging resource assessment report. The charging resource assessment report includes the distribution, availability, charging power, and expansion potential of charging piles within the target traffic monitoring area.

[0110] Step S153: Based on the charging resource assessment report and the key charging demand data, construct a charging pile scheduling strategy. The charging pile scheduling strategy includes the charging piles to be scheduled, the scheduling time, the scheduling location, and the status of the charging piles after scheduling.

[0111] Step S154: According to the charging pile scheduling strategy, a scheduling instruction is sent to the charging pile management system to control the charging pile to move or adjust its charging status. The scheduling instruction includes the charging pile's identifier, scheduling time, scheduling location, and the charging status after scheduling. After receiving the scheduling instruction, the charging pile management system controls the charging pile to perform corresponding operations, including moving to a designated location, adjusting the charging power, or changing the charging status.

[0112] Step S155: After the charging pile scheduling operation is executed, the scheduling effect of the charging pile is monitored in real time. Specifically, the actual scheduling effect of the charging pile is evaluated by collecting real-time charging data, traffic flow data and user feedback data of the charging pile. If the scheduling effect does not meet the expected conditions, the charging pile scheduling strategy is adjusted and the scheduling operation is executed again.

[0113] In step S156, based on historical charging pile scheduling data, real-time charging data, traffic flow data, and user feedback data, the charging pile scheduling strategy is optimized and learned. The correlation and regularity in the historical charging pile scheduling data, real-time charging data, traffic flow data, and user feedback data are analyzed through machine learning algorithms to extract key factors affecting the charging pile scheduling effect. The charging pile scheduling strategy is then intelligently adjusted according to the key factors to generate an optimized charging pile scheduling strategy.

[0114] Step S157: Input the optimized charging pile scheduling strategy into the charging pile scheduling system and update the scheduling parameters and scheduling strategy library of the charging pile scheduling system.

[0115] In this embodiment, taking the central business district of an city as the target traffic monitoring area as an example, the target charging demand data includes charging demand information of various traffic participants (such as taxis, private cars, buses, etc.) in the area at different times. During the preprocessing process, the server accurately extracts key charging demand data from this data. Regarding charging demand periods, it may be analyzed that charging demand increases significantly during the evening peak hours (17:00-19:00) on weekdays, because traffic flow is high, vehicle operating time is long, and battery consumption is rapid. Charging demand hotspots may be concentrated near the entrances of large shopping malls, office buildings, and bus hubs. This is because shopping malls and office buildings are densely populated places, taxis and private cars frequently pick up and drop off passengers, and vehicle dwell time is relatively short but charging demand is concentrated; bus hubs are places where buses stop and turn around, and buses need to be charged in time after operating for a period of time in order to continue running. Regarding the proportion of charging demand types, analysis suggests that taxis account for 30% of the charging demand, due to the large number of taxis operating in commercial areas and their long operating hours; private cars account for 40%, because there are a large number of private cars and some owners charge their vehicles while shopping or working in commercial areas; and buses account for 30%, depending on the bus routes and the number of buses. The predicted charging demand is calculated based on factors such as traffic flow, vehicle type, and operating hours. For example, during the evening peak hours, a total of 100 charging spots are needed, with 30 for taxis, 40 for private cars, and 30 for buses.

[0116] Next, the server assesses charging resources based on this key charging demand data and generates a charging resource assessment report. The server queries the existing charging pile distribution within the target traffic monitoring area and finds 20 charging piles around the shopping mall, 15 near office buildings, and 30 at bus hubs. Regarding availability, communication with the charging pile management system reveals that some charging piles are in use while others are idle. For example, around the shopping mall, 10 are in use and 10 are idle; near office buildings, 8 are in use and 7 are idle; and at bus hubs, 20 are in use and 10 are idle. In terms of charging power, the charging piles vary, with both fast and slow charging piles available. Fast charging piles are mainly distributed at bus hubs to meet the rapid charging needs of buses, while the ratio of fast to slow charging piles around shopping malls and office buildings is relatively balanced. Regarding expansion potential, shopping malls and office buildings have limited expansion potential due to building layout and land resource constraints, while there is some space around bus hubs where the number of charging piles can be increased.

[0117] Then, the server constructs a charging pile scheduling strategy based on the charging resource assessment report and key charging demand data. Based on the predicted charging demand and the existing distribution and availability of charging piles, it determines which charging piles need to be scheduled. For example, five charging piles are moved from areas with low charging demand to the entrance of a shopping mall to meet the increased charging needs of private cars and taxis; the charging power of ten charging piles at a bus hub is adjusted to fast charging mode to improve the charging efficiency of buses. The scheduling time is set to be completed before the evening peak (16:00-16:30) to ensure that the charging piles can operate normally when the peak charging demand arrives. The scheduling location is specified as accurately placing the allocated charging piles in designated vacant locations at the entrance of the shopping mall, and for the adjustment of charging piles within the bus hub, the specific location of the pile is specified. After scheduling, the status of the charging piles is set to be turned on and usable normally, and for charging piles with adjusted charging power, it is ensured that they operate at the set fast charging power.

[0118] Afterwards, the server sends scheduling instructions to the charging pile management system according to the constructed charging pile scheduling strategy. The scheduling instructions contain detailed information, such as the identifier of each charging pile to be scheduled (a unique identifier for each charging pile in the management system used to accurately identify each charging pile); the scheduling time is specified as starting at 16:00; the scheduling location is accurate to the specific coordinates at the mall entrance or the specific charging pile location within the bus hub; and the charging status after scheduling is clearly set as either "on and fast charging" (for charging piles with adjusted power) or "normal charging" (for allocated charging piles). Upon receiving the scheduling instructions, the charging pile management system controls the charging piles to perform corresponding operations according to the instructions. For charging piles that need to be moved, they are moved to the designated location; for charging piles that need to have their charging power adjusted, their charging power is adjusted to the specified power; and for charging piles that need to change their charging status, their status is changed to the set status.

[0119] After the charging pile scheduling operation is executed, the server monitors the scheduling effect of the charging piles in real time. The server collects real-time charging data from the charging piles, including charging current, voltage, and charging time for each charging pile, to understand the actual charging status. It also collects traffic flow data, such as changes in the flow of taxis, private cars, and buses within the target traffic monitoring area during evening rush hour, to determine whether charging demand is effectively met. Furthermore, it collects user feedback data, such as whether taxi drivers find charging more convenient and faster, whether private car owners can easily find available charging piles, and whether buses can charge in time to ensure normal operation. Based on this data, the actual scheduling effect of the charging piles is evaluated. If it is found that despite adding charging piles at the mall entrance, traffic congestion prevents taxis from reaching charging piles in time, or bus charging times are still too long, affecting operation—meaning the scheduling effect has not met expectations—the server adjusts the charging pile scheduling strategy. For example, it might re-plan the placement of charging piles to avoid congested areas, or increase the number of fast-charging piles at bus hubs, and then execute the scheduling operation again.

[0120] Meanwhile, the server optimizes and learns charging pile scheduling strategies based on historical charging pile scheduling data, real-time charging data, traffic flow data, and user feedback data. Machine learning algorithms are used to deeply analyze the correlations and patterns in this data. For example, historical charging pile scheduling data reveals that traffic flow and charging demand in commercial areas change peculiarly on certain dates (such as holidays or promotional days); real-time charging data shows fluctuations in charging time and power requirements for different types of vehicles; traffic flow data analyzes the impact of traffic congestion on vehicle arrival at charging piles; and user feedback data reveals user expectations and dissatisfaction regarding charging pile location and charging speed. The server extracts key factors affecting charging pile scheduling effectiveness, such as traffic congestion, the rationality of charging facility layout, and the charging characteristics of different vehicle types, and intelligently adjusts the charging pile scheduling strategy based on these factors. For example, it dynamically adjusts the layout of charging piles based on traffic congestion and optimizes charging power allocation based on the charging characteristics of different vehicle types, generating an optimized charging pile scheduling strategy. Finally, the optimized charging pile scheduling strategy is input into the charging pile scheduling system to update the scheduling parameters and scheduling strategy library of the charging pile scheduling system, so as to meet the charging demand more accurately and efficiently in future charging pile scheduling decisions, thereby improving the operating efficiency of the entire transportation system and user satisfaction.

[0121] Figure 2 The diagram illustrates the hardware structure of a smart charging pile scheduling system 100 based on traffic flow prediction, provided by an embodiment of the present invention, for implementing the aforementioned smart charging pile scheduling method based on traffic flow prediction. Figure 2As shown, the smart charging pile scheduling system 100 based on traffic flow prediction may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0122] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions used by the smart charging pile scheduling system 100 based on traffic flow prediction to perform or use in order to accomplish the exemplary methods described in this invention.

[0123] In the specific implementation process, one or more processors 110 execute computer-executable instructions stored in machine-readable storage medium 120, so that processor 110 can execute the smart charging pile scheduling method based on traffic flow prediction as described in the above method embodiment. Processor 110, machine-readable storage medium 120 and communication unit 140 are connected through bus 130. Processor 110 can be used to control the sending and receiving actions of communication unit 140.

[0124] The specific implementation process of processor 110 can be found in the various method embodiments executed by the smart charging pile scheduling system 100 based on traffic flow prediction. The implementation principle and technical effect are similar, and will not be repeated here.

[0125] Furthermore, this embodiment of the invention also provides a readable storage medium, which has computer-executable instructions pre-set in it. When the processor executes the computer-executable instructions, the above-mentioned smart charging pile scheduling method based on traffic flow prediction is implemented.

[0126] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A smart charging pile scheduling method based on traffic flow prediction, characterized in that, The method includes: A teacher charging station scheduling network is obtained, which is generated by parameter learning based on teacher sample traffic flow data sequences; the teacher sample traffic flow data sequences correspond to multiple first traffic monitoring areas; each first traffic monitoring area includes multiple teacher sample traffic flow data; each teacher sample traffic flow data carries first sample charging demand data. An auxiliary charging pile scheduling network is obtained, which is generated by parameter learning based on auxiliary sample traffic flow data sequences; the auxiliary sample traffic flow data sequences correspond to multiple second traffic monitoring areas; each second traffic monitoring area includes multiple auxiliary sample traffic flow data; each auxiliary sample traffic flow data carries second sample charging demand data; each second traffic monitoring area does not overlap with any first traffic monitoring area. The system obtains a student charging station scheduling network based on the teacher charging station scheduling network, and obtains a student sample traffic flow data sequence based on the teacher sample traffic flow data sequence and the auxiliary sample traffic flow data sequence; the student sample traffic flow data sequence includes multiple student sample traffic flow data; each student sample traffic flow data carries third sample charging demand data; Based on the student sample traffic flow data sequence, the network parameters of the student charging pile scheduling network are iteratively learned until the network convergence requirement is met, and then the target charging pile scheduling network is generated. Based on the target charging pile scheduling network, a decision is made on the target traffic flow data of any target traffic monitoring area to generate target charging demand data, and charging piles are scheduled for the traffic participants in the target traffic monitoring area according to the target charging demand data. The step of scheduling charging piles for traffic participants in the target traffic monitoring area based on the target charging demand data specifically includes: The target charging demand data of the target traffic monitoring area is preprocessed to extract key charging demand data, which includes charging demand time periods, charging demand hotspot areas, charging demand type ratios, and predicted charging demand amounts. Based on the key charging demand data, a charging resource assessment is conducted to generate a charging resource assessment report. The charging resource assessment report includes the distribution, availability, charging power, and expansion potential of charging piles within the target traffic monitoring area. Based on the charging resource assessment report and the key charging demand data, a charging pile scheduling strategy is constructed. The charging pile scheduling strategy includes the charging piles to be scheduled, the scheduling time, the scheduling location, and the status of the charging piles after scheduling. According to the charging pile scheduling strategy, a scheduling instruction is sent to the charging pile management system to control the charging pile to move or adjust its charging status. The scheduling instruction includes the charging pile's identifier, scheduling time, scheduling location, and the charging status after scheduling. After receiving the scheduling instruction, the charging pile management system controls the charging pile to perform corresponding operations, including moving to a designated location, adjusting the charging power, or changing the charging status. After the charging pile scheduling operation is executed, the scheduling effect of the charging pile is monitored in real time. Specifically, the actual scheduling effect of the charging pile is evaluated by collecting real-time charging data, traffic flow data and user feedback data. If the scheduling effect does not meet the expected conditions, the charging pile scheduling strategy is adjusted and the scheduling operation is executed again. Furthermore, based on historical charging pile scheduling data, real-time charging data, traffic flow data, and user feedback data, the charging pile scheduling strategy is optimized and learned. The correlation and regularity in the historical charging pile scheduling data, real-time charging data, traffic flow data, and user feedback data are analyzed through machine learning algorithms to extract key factors affecting the charging pile scheduling effect. The charging pile scheduling strategy is then intelligently adjusted according to the key factors to generate an optimized charging pile scheduling strategy. The optimized charging pile scheduling strategy is input into the charging pile scheduling system to update the scheduling parameters and scheduling strategy library of the charging pile scheduling system.

2. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 1, characterized in that, Each round of network parameter learning specifically includes: For each student sample traffic flow data, the student sample traffic flow data is loaded into the teacher charging pile scheduling network to generate first charging demand prediction data; the student sample traffic flow data is loaded into the auxiliary charging pile scheduling network to generate second charging demand prediction data; and the student sample traffic flow data is loaded into the student charging pile scheduling network to generate third charging demand prediction data. Based on the first loss function value between the first charging demand prediction data and the third charging demand prediction data, the second loss function value between the second charging demand prediction data and the third charging demand prediction data, and the third loss function value between the third charging demand prediction data and the third sample charging demand data, the global network error parameters are determined. The neuron weight information of the student charging pile scheduling network is updated based on the global network error parameters, and the updated student charging pile scheduling network is used as the student charging pile scheduling network for the next round of network parameter learning.

3. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 1, characterized in that, The step of obtaining the student sample traffic flow data sequence based on the teacher sample traffic flow data sequence and the auxiliary sample traffic flow data sequence includes: Multiple teacher sample traffic flow data are obtained from the teacher sample traffic flow data sequence, and the multiple teacher sample traffic flow data are loaded into the auxiliary sample traffic flow data sequence to generate the student sample traffic flow data sequence.

4. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 3, characterized in that, The step of obtaining multiple teacher sample traffic flow data from the teacher sample traffic flow data sequence and loading the multiple teacher sample traffic flow data into the auxiliary sample traffic flow data sequence to generate the student sample traffic flow data sequence includes: The teacher sample traffic flow data sequence is subjected to multiple rounds of data optimization to generate candidate sample traffic flow data sequences corresponding to each data optimization. The candidate sample traffic flow data sequence is loaded into the auxiliary sample traffic flow data sequence until the training scale in the auxiliary sample traffic flow data sequence meets the first set scale, then the student sample traffic flow data sequence is generated. Specifically, the step of performing multiple rounds of data optimization on the teacher sample traffic flow data sequence to generate candidate sample traffic flow data sequences corresponding to each data optimization includes: First candidate sample traffic flow data is obtained from the teacher sample traffic flow data sequence, and based on the first candidate sample traffic flow data, a first feature distance is calculated between each second candidate sample traffic flow data in the teacher sample traffic flow data sequence and the first candidate sample traffic flow data. The second candidate sample traffic flow data are other sample traffic flow data in the teacher sample traffic flow data sequence besides the first candidate sample traffic flow data. The traffic flow data of each second candidate sample are arranged in descending order according to the first feature distance, and the candidate sample traffic flow data sequence is generated based on the traffic flow data of the second candidate sample of the second set size that is first in the descending sort result and the traffic flow data of the first candidate sample.

5. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 4, characterized in that, If the first candidate sample traffic flow data is the sample traffic flow data corresponding to the first round of data optimization, then obtaining the first candidate sample traffic flow data from the teacher sample traffic flow data sequence includes: Randomly select one teacher sample traffic flow data from the teacher sample traffic flow data sequence and output it as the first candidate sample traffic flow data; If the first candidate sample traffic flow data is not the sample traffic flow data corresponding to the first round of data optimization, then obtaining the first candidate sample traffic flow data from the teacher sample traffic flow data sequence includes: For each second candidate sample traffic flow data, the second feature distance between the second candidate sample traffic flow data and each obtained candidate sample traffic flow data is obtained, and the weighted feature distance of each second feature distance is calculated. The traffic flow data of the second candidate sample with the largest weighted feature distance among the various second feature distances is output as the traffic flow data of the first candidate sample.

6. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 4, characterized in that, The calculation of the first feature distance between each second candidate sample traffic flow data and the first candidate sample traffic flow data in the teacher sample traffic flow data sequence includes: Obtain the first feature weighted value and the first feature dispersion value of the first candidate sample traffic flow data, and obtain the second feature weighted value and the second feature dispersion value of each second candidate sample traffic flow data, and determine the covariance value between the first candidate sample traffic flow data and each second candidate sample traffic flow data; For each second candidate sample traffic flow data, the similarity of the traffic flow scale between the second candidate sample traffic flow data and the first candidate sample traffic flow data is calculated based on the first feature weighting value and the second feature weighting value; The similarity of traffic flow changes between the second candidate sample traffic flow data and the first candidate sample traffic flow data is calculated based on the first feature dispersion value and the second feature dispersion value. The similarity of traffic flow distribution between the second candidate sample traffic flow data and the first candidate sample traffic flow data is calculated based on the first feature dispersion value, the second feature dispersion value, and the covariance value. The first feature distance between the second candidate sample traffic flow data and the first candidate sample traffic flow data is calculated based on the similarity of traffic flow scale, the similarity of traffic flow change, and the similarity of traffic flow distribution.

7. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 2, characterized in that, The determination of global network error parameters based on the first loss function value between the first charging demand prediction data and the third charging demand prediction data, the second loss function value between the second charging demand prediction data and the third charging demand prediction data, and the third loss function value between the third charging demand prediction data and the third sample charging demand data includes: Obtain the first weight value corresponding to the first loss function value, obtain the second weight value corresponding to the second loss function value, and obtain the third weight value corresponding to the third loss function value; The first error parameter of the student charging pile scheduling network is determined based on the first weight value, the first loss function value, the second weight value, and the second loss function value, and the second error parameter of the student charging pile scheduling network is determined based on the third weight value and the third loss function value. The global network error parameters are determined based on the first error parameter and the second error parameter.

8. The intelligent charging pile scheduling method based on traffic flow prediction according to claim 7, characterized in that, The steps for determining the first loss function value and the second loss function value include: The system obtains the first dynamic flow characteristics of the first charging pile scheduling entity and the second dynamic flow characteristics of the non-charging pile scheduling entity in the student sample traffic flow data, and obtains the third dynamic flow characteristics of the second charging pile scheduling entity in the student sample traffic flow data. Determine the third feature distance between the third charging demand prediction data and the first charging demand prediction data, and determine the fourth feature distance between the third charging demand prediction data and the second charging demand prediction data; Based on the first charging demand prediction data, obtain the number of first entities corresponding to the first charging pile scheduling entity and the number of second entities corresponding to the non-charging pile scheduling entity in the student sample traffic flow data; based on the second charging demand prediction data, obtain the number of third entities corresponding to the second charging pile scheduling entity in the student sample traffic flow data. Based on the third feature distance, the first dynamic flow feature, and the first number of subjects, calculate the fourth loss function value between the first charging demand prediction data and the third charging demand prediction data for the first charging pile scheduling subject; and based on the third feature distance, the third dynamic flow feature, and the second number of subjects, calculate the fifth loss function value between the first charging demand prediction data and the third charging demand prediction data for the non-charging pile scheduling subject. The first loss function value is calculated based on the fourth loss function value and the fifth loss function value, and the second loss function value is calculated based on the fourth feature distance, the second dynamic flow feature, and the third number of subjects.

9. A smart charging pile scheduling system based on traffic flow prediction, characterized in that, The intelligent charging pile scheduling system based on traffic flow prediction includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the intelligent charging pile scheduling method based on traffic flow prediction as described in any one of claims 1-8.

Images