Automobile electricity price and carbon emission charging scheduling system and method based on Actor-Critic

Abstract

The invention provides an automobile electricity price and carbon emission charging scheduling system and method based on Actor-Critic, and is applied to the technical field of data processing. The method comprises the following steps: processing power grid dynamic electricity price information and carbon emission signal information based on federated learning privacy protection parameter information to generate income evaluation parameter information; processing the user charging demand information based on historical charging and power grid load problem information, and generating charging strategy adjustment reference information; processing the charging strategy adjustment reference information based on federated learning privacy protection parameter information and different regional power grid information to generate a cross-regional collaborative optimization factor; processing the user charging demand information and regional information corresponding to the user charging demand information based on a cross-regional collaborative optimization factor, and generating charging demand classification information and priority information; and processing based on the scheduling model of the target Actor-Critic algorithm to generate charging scheduling distribution result information.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a charging scheduling system and method for automobile electricity price and carbon emission based on Actor-Critic. Background Art

[0002] As the world pays more attention to environmental protection and sustainable energy, the number of new energy vehicles, as an important alternative to traditional fuel vehicles, has shown explosive growth. However, the large-scale charging of new energy vehicles has brought tremendous pressure to the power grid. The traditional power grid did not fully consider the large-scale charging demand of electric vehicles when it was designed. As a result, during peak hours of electricity consumption, when a large number of new energy vehicles are charging at the same time, the power grid is prone to overload, which seriously affects the stability and reliability of the power grid and may even cause power outages. At the same time, energy costs and carbon emissions are becoming increasingly prominent. Charging costs are significantly affected by the dynamic electricity prices of the power grid, and the electricity prices vary greatly in different periods and regions. It is often difficult for users to reduce charging costs while taking into account electricity demand. In addition, the carbon emissions generated by the electricity consumed by charging new energy vehicles cannot be ignored. In the context of achieving carbon emission reduction goals, how to effectively reduce carbon emissions in charging scheduling has become an urgent problem to be solved.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore includes information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] The purpose of this application is to provide a charging scheduling system and method for automobile electricity prices and carbon emissions based on Actor-Critic, which at least overcomes the problems existing in the prior art to a certain extent. By comprehensively considering the dynamic electricity price information of the power grid, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information, the charging scheduling system and method are of great significance for optimizing the charging management of new energy vehicles, improving the operation efficiency of the power grid, and reducing energy costs and carbon emissions.

[0005] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present invention.

[0006] According to one aspect of the present application, a charging scheduling method for automobile electricity prices and carbon emissions based on Actor-Critic is provided, including: obtaining dynamic power grid electricity price information, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information; processing the dynamic power grid electricity price information and carbon emission signal information based on the federated learning privacy protection parameter information to generate benefit evaluation parameter information; processing the user charging demand information based on the historical charging and power grid load problem information to generate charging strategy adjustment reference information; processing the charging strategy adjustment reference information based on the federated learning privacy protection parameter information and power grid information in different regions to generate a cross-regional collaborative optimization factor; processing the user charging demand information and the regional information corresponding to the user charging demand information based on the cross-regional collaborative optimization factor to generate charging demand classification information and priority information, wherein the user charging demand information is used to characterize the user's vehicle type, charging time preference, battery power information, and power grid access point information; inputting the benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and power grid load problem information into a scheduling model based on a target Actor-Critic algorithm for processing to generate charging scheduling allocation result information.

[0007] Another aspect of the present application is a charging scheduling device for automobile electricity prices and carbon emissions based on Actor-Critic, characterized in that it includes: an acquisition module for acquiring dynamic power grid electricity price information, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information; a processing module for processing the dynamic power grid electricity price information and carbon emission signal information based on the federated learning privacy protection parameter information to generate benefit evaluation parameter information; processing the user charging demand information based on the historical charging and power grid load problem information to generate charging strategy adjustment reference information; based on the federated learning privacy protection parameter information The protection parameter information and the information of the power grid in different regions are used to process the reference information for adjusting the charging strategy, and generate a cross-regional collaborative optimization factor. Based on the cross-regional collaborative optimization factor, the user charging demand information and the regional information corresponding to the user charging demand information are processed to generate charging demand classification information and priority information, wherein the user charging demand information is used to characterize the user's vehicle type, charging time preference, battery power information and power grid access point information. The benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and power grid load problem information are input into the scheduling model based on the target Actor-Critic algorithm for processing to generate charging scheduling allocation result information.

[0008] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the above-mentioned Actor-Critic-based automobile electricity price and carbon emission charging scheduling method is implemented.

[0009] The present application provides a charging scheduling system and method for automobile electricity prices and carbon emissions based on Actor-Critic. The server obtains information such as dynamic power grid electricity prices, carbon emission signals, user charging needs, historical charging and power grid load problems, power grids in different regions, and privacy protection parameters of federated learning. Then, feature extraction and quantitative analysis are performed on the dynamic power grid electricity prices and carbon emission signals, and the benefit evaluation parameter information is generated in combination with the privacy protection parameters. At the same time, the historical charging and power grid load problem information and user charging demand information are processed to generate charging strategy adjustment reference information. Based on the federated learning privacy protection parameters and the power grid information in different regions, the charging strategy adjustment reference information is processed to obtain a cross-regional collaborative optimization factor for generating charging demand classification information and priority information. Finally, the benefit evaluation parameters, charging demand classification and priority information, and historical charging and power grid load problem information are input into the target model. After analysis and processing, strategy vector optimization adjustment, and deviation vector analytical conversion, the charging scheduling allocation result information is generated, covering the allocated power grid resources and target quantitative indicators, realizing charging scheduling optimization that comprehensively considers cost, efficiency, and power grid stability, and improving the charging management level of new energy vehicles.

[0010] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A flowchart of a method for charging scheduling of automobile electricity prices and carbon emissions based on Actor-Critic is shown in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a charging scheduling device for automobile electricity price and carbon emission based on Actor-Critic provided in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0012] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0013] Combine the following Figure 1 To describe the charging scheduling method of automobile electricity price and carbon emission based on Actor-Critic according to the exemplary embodiment of the present application. Figure 1 As shown, the method is applied to a server, comprising: S101, obtain dynamic power price information of the power grid, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information.

[0014] In one implementation, taking a regional power grid and new energy vehicle users in the region as an example, the regional power grid operator will adjust the electricity price in real time according to factors such as power supply and demand, power generation costs, etc. Through the power grid data monitoring system, it is obtained that the electricity price during the peak period (8:00-18:00) during the daytime on weekdays is 0.8 yuan / kWh, the electricity price during the low period (23:00-7:00) at night is 0.3 yuan / kWh; the electricity price during the peak period (10:00-16:00) on weekends is 0.7 yuan / kWh, etc. These electricity price data that change dynamically over time are the dynamic electricity price information of the power grid. The environmental protection department cooperates with the energy monitoring agency to monitor the carbon emissions in the power production process in the region. The data obtained show that the carbon emission intensity generated by thermal power generation is 0.8 kg / kWh, the carbon emission intensity generated by hydropower generation is almost 0, and the carbon emission intensity generated by wind power generation is 0.05 kg / kWh. At the same time, according to regional environmental protection policies and carbon emission targets, information such as the total amount of carbon emissions allowed in different time periods and the real-time carbon emission remaining amount can also be obtained, which constitutes the carbon emission signal information.

[0015] New energy vehicle users upload charging requirements to the server through mobile phone charging apps or vehicle systems. For example, user A's vehicle type is a pure electric sedan with a battery capacity of 60 kWh and a current battery charge of 30%. He plans to start charging after get off work (18:30) and hopes to be fully charged within 2 hours. The usual charging location for this vehicle is a private charging pile at home in the city; user B's vehicle is an electric SUV with a battery capacity of 80 kWh and a remaining charge of 20%. It is expected to be charged before the weekend self-driving tour (9:00 am on Saturday) and requires charging to 80%. The charging location is a public charging pile in a highway service area. The user's vehicle type, charging time preference, battery charge information, and grid access point information are summarized as user charging demand information. Relevant information is extracted from the historical database of grid operation. In the past month, due to the centralized charging of a large number of new energy vehicles in the area on weekday evenings (19:00-22:00), the grid in some areas was overloaded and the transformer had an overheating alarm; at the same time, during certain periods, due to irregular user charging habits, the charging power fluctuated greatly, affecting the stability of the grid. Information such as power fluctuations, grid overload, and abnormal charging time that occurred during these historical charging processes constitutes historical charging and grid load problem information.

[0016] In addition, the region is composed of multiple sub-grids, such as the city center grid, suburban grid, and industrial park grid. The power supply of the city center grid is mainly a mixture of thermal power and hydropower, with a large load and stable electricity demand; the suburban grid has a high proportion of wind power and a relatively small load, but there will be power shortages during the peak of residential electricity consumption at night; the industrial park grid mainly guarantees electricity for industrial production and has high requirements for power stability. There are differences in power supply composition, load characteristics, power supply capacity, etc. in different regional power grids. This information is the information of different regional power grids.

[0017] In the case of multi-institutional cooperation, federated learning technology is used to protect the data privacy of all parties. For example, power grid operators, new energy vehicle manufacturers and energy research institutions jointly participate in data cooperation. The privacy protection parameters of federated learning set by each party include the type of encryption algorithm (such as the use of homomorphic encryption algorithm to ensure that data is calculated in an encrypted state), the scope of data sharing (only statistical information that has been aggregated and anonymized is shared, and sensitive data of specific users or power grid nodes is not involved), and the authority management of participants (specifying that power grid operators can only view data related to power grid loads, and car manufacturers can only obtain aggregated data related to vehicle charging behavior, etc.) and other information. These parameters ensure that the privacy and security of all parties are protected during data sharing and collaborative processing.

[0018] S102, processing the dynamic electricity price information of the power grid and the carbon emission signal information based on the federated learning privacy protection parameter information to generate benefit evaluation parameter information.

[0019] In one implementation, feature extraction is performed on the dynamic electricity price information of the power grid to generate electricity price fluctuation amplitude features, electricity price peak and valley period features, and electricity price regional difference features. In a certain week, the dynamic electricity price monitoring data of the power grid in the region showed that the highest electricity price during the peak period during the day on Monday was 0.9 yuan / kWh, while the lowest electricity price during the valley period at night was 0.25 yuan / kWh. By calculation, it is found that the fluctuation amplitude of the electricity price on this day is (0.9-0.25) ÷ 0.25 = 2.6, that is, the fluctuation amplitude of the electricity price within a day is large. Within a week, the fluctuation amplitude of the electricity price on each day is different. Through statistical analysis of the data for one week, the characteristic value of the average electricity price fluctuation amplitude of the week is obtained to reflect the degree of fluctuation of the dynamic electricity price of the power grid in the region.

[0020] After long-term monitoring, it is found that the peak hours of electricity prices in this area on weekdays are usually 8-10 am and 7-9 pm. During this period, the demand for electricity is large and the electricity price is high; the valley period is 11 pm-6 am the next day. During this period, the demand for electricity is small and the electricity price is low. The peak hours on weekends are concentrated from 10 am to 4 pm, and the valley period is from 10 pm to 8 am the next day. These fixed peak and valley periods form the peak and valley period characteristics of electricity prices in this area, which are of great reference value for new energy vehicle users to reasonably arrange charging time. The area is divided into three main areas: A, B, and C. Area A is the downtown commercial area with high electricity supply costs. Its peak electricity price is generally 0.1-0.2 yuan / kWh higher than other areas; Area B is a residential area with relatively stable electricity prices; Area C is an industrial park. Due to the large industrial electricity consumption and special electricity policies, its peak and valley electricity prices are more different, and the valley electricity price is even 0.15 yuan / kWh lower than that of Area B. This difference in electricity prices between different regions constitutes the regional difference characteristics of electricity prices.

[0021] The carbon emission signal information is processed by feature extraction to generate the characteristics of carbon emission intensity changes, the characteristics of the time period when the carbon emission peak occurs, and the characteristics of carbon emission trends in different regions. The main sources of electricity in this region are thermal power, hydropower and wind power. In the past year, with the improvement of environmental protection technology, the carbon emission intensity of thermal power has dropped from 0.85 kg / kWh to 0.8 kg / kWh, showing a gradually decreasing trend; the carbon emission intensity of hydropower has remained at an extremely low level, almost negligible; the carbon emission intensity of wind power is 0.05 kg / kWh, which is relatively stable. By monitoring the changes in the carbon emission intensity of different power sources, the overall carbon emission intensity change characteristics of the region are obtained, reflecting the changes in carbon emission intensity over time in the power production process.

[0022] Through the analysis of regional carbon emission data, it is found that during the winter heating period each year, due to the increase in thermal power generation, the peak of carbon emissions usually occurs from December to February of the following year. In a day, the peak of carbon emissions often occurs during the peak electricity consumption period, such as 7 pm to 9 pm on weekdays. These fixed time periods are the characteristics of the time periods when the carbon emission peak occurs, which helps to understand the time distribution pattern of carbon emissions.

[0023] In region A, due to its intensive commercial activities and high electricity consumption, it mainly relies on thermal power supply, and its carbon emissions show a relatively high and stable trend; in region B, residents mainly use electricity, and the proportion of hydropower and wind power is gradually increasing, and its carbon emissions are slowly declining; although region C has a large industrial electricity consumption, its carbon emissions are also gradually decreasing with the green transformation of industry and the increase in the use of clean energy. These differences in carbon emission trends in different regions form the characteristics of carbon emission trends in different regions.

[0024] The characteristics of electricity price fluctuation amplitude, peak and valley period characteristics, and regional difference characteristics of electricity prices are quantitatively analyzed and processed to generate quantitative characteristic values ​​of electricity prices. The fluctuation amplitude of electricity prices for each day of the week is weighted averaged, and different weights are given to take into account the different electricity consumption patterns on weekdays and weekends. Finally, the quantitative characteristic value of electricity price fluctuation amplitude for the week is 2.3. For the peak and valley period characteristics of electricity prices, according to the proportion of electricity consumption and the difference in electricity prices during the peak and valley periods, a comprehensive peak and valley electricity price quantitative characteristic value is calculated to be 0.5 (comprehensively considering factors such as the duration of the peak and valley periods, the difference in electricity prices, and the proportion of electricity consumption in each period). For the regional difference characteristics of electricity prices, by comparing the degree of difference in electricity prices in different regions, the quantitative characteristic value of regional differences is obtained to be 0.18 (reflecting the relative size of the difference in electricity prices in each region). These quantitative characteristic values ​​are integrated to form a complete set of quantitative characteristic values ​​of electricity prices.

[0025] The characteristics of carbon emission intensity changes, the characteristics of the time period when the carbon emission peak occurs, and the characteristics of carbon emission trends in different regions are quantitatively analyzed and processed to generate quantitative characteristic values ​​of carbon emissions. In terms of the characteristics of carbon emission intensity changes, the average decline rate of thermal power carbon emission intensity in the past year is calculated, and the quantitative characteristic value of carbon emission intensity changes is -0.05 (the negative sign indicates a decline). For the characteristics of the time period when the carbon emission peak occurs, a quantitative characteristic value reflecting the importance of carbon emissions during the peak period is calculated to be 0.6 based on the proportion of carbon emissions during the winter heating period and the peak electricity consumption period of the day. The quantification of carbon emission trend characteristics in different regions is obtained by taking a weighted average of the carbon emission change rates in each region, and the overall regional carbon emission trend quantitative characteristic value is -0.03. These values ​​constitute a set of carbon emission quantitative characteristic values.

[0026] Based on the privacy protection parameter information of federated learning, the quantitative characteristic values ​​of electricity prices and carbon emissions are processed to generate comprehensive benefit evaluation characteristics and corresponding weight calculation result information, and the comprehensive benefit evaluation characteristics and corresponding weight calculation result information are processed to generate benefit evaluation parameter information, wherein the benefit evaluation parameter information is used to characterize the degree of influence of dynamic power grid electricity prices and carbon emission signals on the charging benefits of new energy vehicles. Under the framework of federated learning, power grid operators, new energy vehicle manufacturers and energy research institutions jointly participate in data cooperation. According to the set privacy protection parameters of federated learning, such as using homomorphic encryption algorithms to ensure that data is calculated in an encrypted state, only statistical information that has been aggregated and anonymized is shared, etc. Each participant uses its own data advantages to conduct collaborative analysis of the quantitative characteristic values ​​of electricity prices and carbon emissions. The power grid operator provides power grid cost and power supply stability data, the automobile manufacturer provides vehicle energy consumption and user charging habit data, and the energy research institution provides assessment data on the impact of carbon emissions on the environment. Through the fusion analysis of these data, a comprehensive benefit evaluation feature is generated. For example, it is found that the fluctuation range of electricity prices is closely related to the charging cost of users, and it is used as an important comprehensive benefit evaluation feature; the change of carbon emission intensity has a great impact on the environment and energy policies, and is also included in the comprehensive benefit evaluation feature. At the same time, the corresponding weights are calculated according to the degree of influence of each feature on the charging income of new energy vehicles. For example, the weight of the fluctuation range of electricity prices is 0.4, the weight of the change of carbon emission intensity is 0.3, and so on, and the weight calculation result information is obtained.

[0027] The comprehensive benefit evaluation characteristics and the corresponding weight calculation result information are further processed. Multiply each comprehensive benefit evaluation characteristic by its corresponding weight, and then add them up to obtain a comprehensive benefit evaluation parameter value. Assume that the benefit evaluation parameter value obtained by calculation is 0.7 (this value comprehensively reflects the impact of the dynamic electricity price of the power grid and the carbon emission signal on the charging benefits of new energy vehicles). This benefit evaluation parameter information can help new energy vehicle users understand the charging benefits under different electricity prices and carbon emissions, and also provide an important basis for the charging dispatch system to formulate a reasonable charging strategy. If the benefit evaluation parameter value is high, it means that the current electricity price and carbon emissions are more favorable to the charging benefits; otherwise, the charging strategy needs to be adjusted to increase the benefits.

[0028] S103, processing the user charging demand information based on historical charging and grid load problem information to generate charging strategy adjustment reference information.

[0029] In one implementation, historical charging and grid load problem information is extracted and classified to generate historical charging power fluctuation information, grid load overload information, and charging time abnormality information. By analyzing the historical operation data of the regional power grid, historical charging and grid load problem information in the past month is extracted. For example, it is found that the charging power fluctuates greatly in some periods. During the period from 20:00 to 21:00 on a certain weekday evening, due to the centralized charging of a large number of new energy vehicles, the charging power in a certain area rose rapidly from 500 kilowatts to 1200 kilowatts within this hour, and then dropped to 800 kilowatts within half an hour, forming an obvious power fluctuation, which is the historical charging power fluctuation information. In the past month, the regional power grid has experienced grid load overload on many working days in the evening (19:00-22:00). For example, at 20:30 on a certain working day, the grid load in some areas exceeded 20% of its rated load, resulting in excessive temperature of some transformers and an overload alarm signal. These situations constitute grid load overload information. At the same time, abnormal charging time was also found. Under normal circumstances, it takes 1-1.5 hours to fully charge a certain model of pure electric car using fast charging equipment. However, during certain periods of time, due to unstable grid voltage or charging pile failure, the charging time of some vehicles is extended to 3-4 hours. This is abnormal charging time information.

[0030] Based on the user's vehicle type, charging time preference, battery power information, and grid access point information contained in the user's charging demand information, the historical charging power fluctuation information, grid load overload information, and charging time abnormality information are processed to generate charging problem type judgment information and charging problem severity assessment information. It is known that user A's vehicle type is a pure electric sedan with a battery capacity of 60 kWh and a current battery power of 30%. It is planned to start charging after get off work (18:30) and hopes to be fully charged within 2 hours. The charging location is a private charging pile at home in the urban area. When referring to the historical charging power fluctuation information, it is found that the charging power in the area fluctuates greatly during the period (18:30-20:30) when user A plans to charge. Combined with user A's charging demand, it is judged that this power fluctuation may cause the charging time to be extended, affecting the user's usage plan, and it is determined as a charging problem of the "charging time affected" type.

[0031] Regarding the severity assessment of this problem, considering that user A expects to fully charge in 2 hours, and power fluctuations may extend the charging time by 1-2 hours, which seriously affects the user's convenience, the severity of this problem is assessed as "relatively serious". Let's look at user B. The vehicle is an electric SUV with a battery capacity of 80 kWh and a remaining power of 20%. It is expected to be charged before the weekend self-driving tour (9:00 am on Saturday), and the power is required to be charged to 80%. The charging location is selected at the public charging pile in the highway service area. Referring to the grid load overload information, it is found that the grid in the area where the highway service area is located often has load overload on weekend mornings (9:00-11:00). Combined with user B's charging needs, it is judged that this overload situation may cause the charging speed to slow down or even fail to charge normally, and it is judged as a charging problem of the "charging interruption risk" type. Since user B urgently needs to charge before the self-driving tour, the charging interruption will seriously affect the travel plan, so the severity of this problem is assessed as "serious".

[0032] Based on the charging problem type judgment information and charging problem severity assessment information, the target data in the user's charging demand information is marked and screened to generate the charging demand abnormal data screening results. According to the above charging problem type judgment information and severity assessment information, the target data in the user's charging demand information is marked and screened. For user A, his charging time preference (18:30-20:30), battery power information (current power 30%, expected to be fully charged) and grid access point information (private charging pile at home in the urban area) are marked as abnormal data, because these data are related to charging problems caused by charging power fluctuations. For user B, his charging time preference (9:00-11:00 am on Saturday), battery power information (remaining power 20%, expected to be charged to 80%) and grid access point information (public charging pile in high-speed service area) are marked as abnormal data, which are closely related to charging problems that may be caused by grid overload. These marked abnormal data are summarized to form the charging demand abnormal data screening results.

[0033] The abnormal charging demand data screening results are integrated and quantified to generate charging strategy adjustment reference information, where the charging strategy adjustment reference information is used to characterize the correlation between user charging demand and historical charging and grid load problems. For user A's abnormal data, considering the impact of power fluctuations on charging time, it is calculated that under the current power fluctuations, the expected charging time will be extended by 1.5 hours. Based on this, it is recommended that user A start charging half an hour in advance, or charge in other periods with smaller power fluctuations, such as the low-peak period at night (23:00-7:00), to ensure that it can be fully charged within the expected time. For user B's abnormal data, due to the high risk of overload of the grid in the high-speed service area, the probability of normal charging during this period is calculated to be 60%. Therefore, it is recommended that user B check the real-time usage of the charging pile in advance. If the grid load in the service area is found to be too high, you can choose charging piles in other nearby service areas, or charge at home in advance the night before departure to reduce the risk of charging interruption.

[0034] These charging strategy recommendations for different users are integrated to generate reference information for charging strategy adjustment. This information clearly shows the relationship between user charging needs and historical charging and grid load issues, providing an important basis for the subsequent formulation of reasonable charging scheduling strategies. For example, for users like user A who charge during power fluctuation periods, the scheduling system can guide them to charge at a more appropriate time to balance the grid load; for users like user B who face the risk of grid overload, the scheduling system can provide more reliable charging location recommendations to ensure the user's charging needs.

[0035] S104: Based on the federated learning privacy protection parameter information and the power grid information of different regions, the charging strategy adjustment reference information is processed to generate a cross-regional collaborative optimization factor.

[0036] In one implementation, the charging strategy adjustment reference information is subjected to target element extraction and data classification processing to generate charging power adjustment information, charging time adjustment information, and charging area adjustment information. The charging strategy adjustment reference information is analyzed to extract target elements and classify them. Assume that when analyzing the association between user charging demand and historical charging and grid load problems, it is found that when user A charges at 19:00-20:00 in the evening, the grid load overload risk in the area is high, and the charging power fluctuations affect the charging efficiency. To solve this problem, the charging power adjustment information is generated to reduce the charging power by 30% during this period to reduce the grid load pressure; the charging time adjustment information is to recommend user A to postpone the charging time to 22:00-23:00 to avoid the peak of electricity consumption; the charging area adjustment information is for some users who have high charging costs or inconvenience in charging in specific areas. For example, user B often charges at public charging piles in the downtown business district (A area), which is expensive. It is recommended that he use the suburban grid (relatively low electricity prices) to charge when going to the suburbs to do business to save costs.

[0037] Based on the privacy protection parameter information of federated learning and the information of power grids in different regions, the charging power adjustment information, charging time adjustment information, and charging area adjustment information are comprehensively analyzed to generate adjustment strategy importance assessment information and adjustment strategy feasibility assessment information. Based on the privacy protection parameter information of federated learning and the information of power grids in different regions (such as the stability of power supply, electricity price differences, load conditions, etc. in different regions), the above adjustment information is comprehensively analyzed. In terms of importance, adjusting the charging power and time is crucial to alleviating the problem of overload of the power grid, because it is directly related to the stable operation of the power grid and the normal electricity consumption of other users. The importance assessment score is set to 8 points (out of 10 points); and adjusting the charging area is of great significance to reducing the charging cost of users, especially for users who often travel across regions. The importance assessment score is set to 7 points. From the perspective of feasibility, adjusting the charging power can be technically achieved through smart charging piles, and has little impact on users, so the feasibility assessment score is set to 9 points; adjusting the charging time requires users to change their charging habits, but considering that the electricity price is lower during the night off-peak period, it is attractive to users, so the feasibility assessment score is set to 7 points; adjusting the charging area is highly feasible for users with cross-regional travel needs, but it is less feasible for users with a fixed range of activities, so the feasibility assessment score is set to 6 points after comprehensive consideration. In this way, the importance assessment information and feasibility assessment information of the adjustment strategy are generated.

[0038] According to the adjustment strategy importance assessment information and the adjustment strategy feasibility assessment information, the key data in the charging strategy adjustment reference information is screened and associated to generate the charging strategy target data screening results. For the charging power adjustment information, the user data that charges during the peak load period of the power grid and has large power fluctuations is screened, such as the relevant charging data of user A; for the charging time adjustment information, the users who charge during peak hours and have the possibility of adjusting the charging time are associated, such as user A and some users with flexible charging time; for the charging area adjustment information, the user data that often travels across regions and charges in high-cost areas is selected, such as the data of user B. These screened data are integrated to form the charging strategy target data screening results.

[0039] The results of the charging strategy target data screening are integrated and processed in combination with the privacy protection parameter information of federated learning (such as data encryption method, data sharing scope, etc.) and the weight information in the power grid information of different regions (such as the importance weight of power grids in different regions, the weight of the power grid in the city center is set to 0.4 due to the large load, the weight of the suburban power grid is set to 0.3, and the weight of the power grid in the industrial park is set to 0.3). For example, the value of a cross-regional collaborative optimization factor is calculated to be 0.65. This value shows that under the current circumstances, the direction of cross-regional collaborative optimization is to encourage users to reasonably allocate charging behaviors in different regions, giving priority to charging in areas with lower load and lower cost; in terms of degree, it means that there is a 65% possibility of achieving more efficient charging scheduling through cross-regional collaborative optimization. Through this cross-regional collaborative optimization factor, the system can guide new energy vehicle users in different regions to reasonably adjust their charging strategies, achieve cross-regional optimal allocation of power resources, balance the load of power grids in different regions, reduce user charging costs, and improve the overall charging efficiency and grid operation stability.

[0040] S105 , processing the user charging demand information and the regional information corresponding to the user charging demand information based on the cross-region collaborative optimization factor to generate charging demand classification information and priority information.

[0041] In one implementation, data extraction and processing is performed on the user's charging demand information to generate user vehicle type information, charging time preference information, battery power information, and grid access point information. Assume that there are three new energy vehicle users in the area. User A's vehicle is a pure electric sedan with a battery capacity of 60 kWh and a current power of 30%. He is used to starting charging at 18:30 after get off work, and the usual charging location is a private charging pile at home in the city; User B's vehicle is an electric SUV with a battery capacity of 80 kWh and a remaining power of 20%. It is planned to charge before the weekend self-driving tour (9:00 am on Saturday), and the charging location is a public charging pile in a highway service area; User C's vehicle is an electric microcar with a battery capacity of 35 kWh and a current power of 40%. It is usually charged around 22:00 in the evening, and the charging location is a charging pile in the company's parking lot. By extracting and processing the charging demand information of these users, we can obtain the user's vehicle type information (User A: pure electric sedan; User B: electric SUV; User C: electric microcar), charging time preference information (User A: 18:30; User B: Saturday morning 9:00; User C: around 22:00), battery power information (User A: 30%; User B: 20%; User C: 40%) and grid access point information (User A: private charging station at home in the city; User B: public charging station in the highway service area; User C: charging station in the company parking lot).

[0042] Based on the regional information corresponding to the user's charging demand information, the user's vehicle type information, charging time preference information, battery power information, and grid access point information are standardized and integrated, and various types of information are converted into a unified processable format to generate a standardized charging demand information set. The area is divided into different sub-grids such as the city center regional power grid, suburban power grid, and industrial park power grid. The regional information corresponding to user A shows that the urban power grid where it is located has stable voltage and high power supply reliability; the high-speed service area corresponding to user B is in the connecting area of ​​the suburban power grid and the city center regional power grid, and there may be a shortage of power supply during peak power consumption; the company where user C is located is located in the industrial park, and the regional power grid mainly guarantees industrial electricity consumption, and there are certain policy preferences for civilian charging at night.

[0043] Based on this regional information, the user's vehicle type, charging time preference, battery power, and grid access point information are standardized and integrated. For example, the user's vehicle type information is uniformly coded, with pure electric sedans coded as 01, electric SUVs coded as 02, and electric microcars coded as 03; the charging time preference information is uniformly converted into a value in hours (18:30 is converted to 18.5, 9:00 is converted to 9, and 22 is taken around 22:00); the battery power information is uniformly converted into a value expressed in percentage; the grid access point information is classified and coded according to the regional grid, such as the private charging pile at home in the city is coded as A01, the public charging pile at the highway service area is coded as B02, and the charging pile at the company parking lot is coded as C03. After integration, a standardized charging demand information set is generated. Taking user A as an example, its standardized information is [01,18.5,30,A01].

[0044] Based on the cross-regional collaborative optimization factor, the importance of the standardized charging demand information set is evaluated and screened. According to the optimization direction and degree reflected by the cross-regional collaborative optimization factor, the importance weight of each information in the classification and priority judgment of charging demand is generated. Assuming that the value of the cross-regional collaborative optimization factor obtained by the previous calculation is 0.7, it means that cross-regional collaborative optimization has a higher positive impact on charging scheduling under the current circumstances, and is more inclined to guide users to reasonably allocate charging behaviors in different regions, giving priority to charging in areas with lower loads and lower costs. Based on this, the importance of the standardized charging demand information set is evaluated and screened. For charging time preference information, since cross-regional collaborative optimization emphasizes avoiding peak hours, the importance weight of the time preference for charging during peak hours (such as 18:00-20:00) is reduced; while the importance weight of the time preference for charging during valley hours (such as 22:00-6:00) is increased. For example, user A originally charged at 18:30, and his time preference weight may be set to 0.3, which is adjusted to 0.2 after evaluation; user C charges at 22:00, and his time preference weight is increased from 0.3 to 0.4.

[0045] For battery power information, the importance weight of battery power information of users with low power and urgent need to charge in priority judgment is increased. For example, if user B has 20% remaining power and needs to charge before self-driving travel, the weight of his battery power is increased from 0.2 to 0.3; user C's current power is 40%, and the weight is adjusted from 0.2 to 0.15. For grid access point information, the weight of areas with high load (such as some charging piles in the downtown area grid) is reduced, and the weight of areas with low load and low cost (such as some charging piles in the suburbs) is increased. For example, the weight of user A's private charging pile in the city home is reduced from 0.3 to 0.25, and if there is a charging pile in the suburbs, its weight may be increased from 0.2 to 0.3. In this way, the importance weight of each information in the classification and priority judgment of charging needs is generated.

[0046] The importance weights of each piece of information in the classification and priority judgment of charging demand are processed to generate charging demand classification information and priority information. According to the weighted calculation of the comprehensive score, user B has a low battery and the charging time is during the peak period on weekends. Although the weight of the public charging piles in the highway service area has been adjusted due to regional factors, the comprehensive score shows that his charging demand is more urgent, so he can be classified as "urgent charging demand" and the priority is set to high; although user A's current battery is not very low, the charging time is during the peak electricity consumption period in the evening, and the comprehensive score makes it classified as "general charging demand" and the priority is set to medium; user C has a relatively high battery and the charging time is during the low period at night, so he is classified as "delayable charging demand" and the priority is set to low. In this way, the classification and priority of charging needs of different users are clarified, which provides an important basis for subsequent charging scheduling.

[0047] S106, inputting the benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and grid load problem information into a scheduling model based on the target Actor-Critic algorithm for processing to generate charging scheduling allocation result information.

[0048] In one implementation, the dynamic power price information of the power grid, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, privacy protection parameter information of federated learning, and corresponding charging scheduling allocation result samples are obtained for training; similar information and samples are obtained for verification; and the initial scheduling model of the preset Actor-Critic algorithm is obtained. The dynamic power price information of the power grid in the past year is collected from the power grid operation system in the region, including power price data at different time periods on weekdays and weekends, such as the average power price of 0.8 yuan / kWh during the peak period (8:00-18:00) during the daytime on weekdays, and the average power price of 0.3 yuan / kWh during the low period (23:00-7:00) at night. Carbon emission signal information is obtained from the environmental protection and energy monitoring department, such as the carbon emission intensity of thermal power generation in the region is 0.8 kg / kWh, the carbon emission intensity data of hydropower and wind power, and the total carbon emission limit and real-time surplus at different time periods.

[0049] Through the mobile phone charging APP and vehicle system of new energy vehicle users, information is collected, such as user A's vehicle type is a pure electric sedan, the battery capacity is 60 kWh, the current power is 30%, he is used to starting charging at 18:30 after get off work, and expects to be fully charged in 2 hours, and the usual charging location is the private charging pile at home in the city; user B's vehicle is an electric SUV, the battery capacity is 80 kWh, the remaining power is 20%, and it is planned to charge to 80% at the public charging pile in the highway service area at 9:00 am on weekends. Obtain historical charging and grid load problem information from the power grid history database, such as in the past month, during the weekday evening from 19:00 to 22:00, some areas caused overload of the grid due to centralized charging, and the transformer overheated alarm; charging power fluctuated greatly during certain periods, etc. At the same time, obtain information on power grids in different regions, such as the differences in power supply composition, load characteristics, power supply capacity, etc. between the power grid in the city center, the suburban power grid, and the industrial park power grid. In multi-institutional cooperation, set the privacy protection parameters of federated learning, such as using homomorphic encryption algorithms for data encryption calculations, only sharing aggregated and anonymized statistical information, power grid operators can only view data related to power grid load, and car manufacturers can only obtain summary data on vehicle charging behavior, etc. Collect the charging scheduling and allocation results of this information in different time periods as samples, such as user A in a certain time period according to the system's recommended charging strategy, actual charging power, charging time, and cost data. Obtain similar information and samples for verification in the same way for subsequent model verification.

[0050] Select a preset Actor-Critic algorithm initial scheduling model, which is a two-layer neural network structure. Both the Actor network and the Critic network contain a hidden layer. The number of nodes in the input layer of the Actor network is determined according to the number of input data features. Assuming that 10 features are considered (such as electricity price fluctuations, carbon emission intensity, vehicle type, etc.), the number of input layer nodes is 10; the number of hidden layer nodes is set to 30, which is determined through experience and experiments and is used to extract features and perform nonlinear transformations on input data; the number of output layer nodes is 3, corresponding to the charging power adjustment, charging time adjustment, and charging area selection (assuming it is divided into three areas). The number of input layer nodes of the Critic network is also 10, the number of hidden layer nodes is also 30, and the number of output layer nodes is 1, which is used to evaluate the state value.

[0051] Data cleaning, feature extraction and normalization are performed on the training grid dynamic electricity price information, carbon emission signal information, user charging demand information, historical charging and grid load problem information, grid information in different regions, and federated learning privacy protection parameter information to generate pre-processed training feature data, where the pre-processed training feature data includes electricity price fluctuation amplitude characteristics, electricity price peak and valley period characteristics, electricity price regional difference characteristics, carbon emission intensity change characteristics, carbon emission peak period characteristics, carbon emission trend characteristics in different regions, user vehicle type information, charging time preference information, battery power information, and grid access point information. Check whether there are abnormal values ​​for the training grid dynamic electricity price information. If there is an obvious error in the electricity price at a certain moment (such as exceeding the normal range by several times), remove or correct it; process incomplete or erroneous data in the user charging demand information, such as missing battery power information or data with incorrect charging time format. Extract the characteristics of electricity price fluctuation amplitude from the dynamic electricity price information of the power grid, such as calculating the difference between the electricity price during the peak and valley hours of each day in a week, and then finding the average value; extract the characteristics of the peak and valley periods of electricity prices, and determine the peak and valley periods on weekdays and weekends; extract the characteristics of regional differences in electricity prices, and compare the differences in electricity prices of power grids in different regions. For carbon emission signal information, extract the characteristics of carbon emission intensity changes, and calculate the rate of change of thermal power carbon emission intensity in the past year; extract the characteristics of the time period when the carbon emission peak occurs, and determine the carbon emission peak during the winter heating period and the peak electricity consumption period of the day; extract the characteristics of carbon emission trends in different regions, and analyze the carbon emission change trends in different regions (such as city centers, suburbs, and industrial parks). Extract user vehicle type information, charging time preference information, battery power information, and grid access point information from user charging demand information.

[0052] Normalize numerical features such as electricity price fluctuations and carbon emission intensity change rates, and map them to the [0,1] interval to make different features comparable. For example, the maximum and minimum values ​​of electricity price fluctuations are normalized. Assuming that the maximum value of electricity price fluctuations in a certain area is 0.6 yuan / kWh, the minimum value is 0.1 yuan / kWh, and the fluctuation at a certain moment is 0.3 yuan / kWh, then the normalized value is (0.3-0.1) / (0.6-0.1)=0.4. Encode the user's vehicle type information, such as pure electric sedans are coded as 01, electric SUVs are coded as 02, etc.; the charging time preference information is converted into a numerical value in hours and then normalized; the battery power information is directly expressed as a percentage, without the need for normalization; the grid access point information is encoded according to the region to which it belongs, and then normalized accordingly. Finally, the preprocessed training feature data is generated, including the characteristics of electricity price fluctuation amplitude, electricity price peak and valley time periods, electricity price regional difference characteristics, carbon emission intensity change characteristics, carbon emission peak time period characteristics, carbon emission trend characteristics in different regions, user vehicle type information, charging time preference information, battery power information, and grid access point information.

[0053] The preprocessed training feature data is processed to generate the scheduling model prediction parameters, where the scheduling model prediction parameter vector is used to characterize the prediction information of the charging scheduling allocation results, the key factors affecting the charging scheduling, and the optimization strategy. The Actor-Critic algorithm model is trained by the back propagation algorithm using the preprocessed training feature data. During the training process, the model continuously adjusts the connection weights between neurons to minimize the error between the predicted charging scheduling allocation results and the actual results. For example, in one training, the model predicts a charging scheduling strategy (charging power, time, and area) based on the current input features (such as a user's vehicle type, charging time, current electricity price, and carbon emissions), compares the predicted results with the actual charging scheduling allocation result samples, and calculates the error. Based on the error, the back propagation algorithm adjusts the weights of the neurons in each layer of the model to make the model more accurate in subsequent predictions. After multiple iterations of training, the model learns the characteristic patterns in the data and generates a scheduling model prediction parameter vector. This vector contains prediction information on the charging scheduling allocation results (such as expected charging power, charging time, and charging area), as well as prediction information on key factors affecting charging scheduling (such as the impact of electricity price fluctuations on charging costs, the impact of grid loads in different regions on charging feasibility, etc.) and optimization strategies (such as recommendations to increase charging power during periods of low electricity prices).

[0054] Based on the scheduling model prediction parameter vector, the preset Actor-Critic algorithm's initial scheduling model is optimized and trained to generate a trained Actor-Critic algorithm scheduling model. The weight information in the prediction parameter vector is applied to the connection weights of the initial model to further adjust the model's parameters. During the training process, the learning rate is set to 0.01, which determines the step size of the model's parameter update at each iteration; the number of iterations is set to 500, and the model is continuously fitted to the training data through multiple iterations. For example, in each iteration, the model calculates the output results based on the new weights, compares them with the actual results, and adjusts the weights again to make the model's prediction results on the training data more accurate. After such optimization training, a trained Actor-Critic algorithm scheduling model is generated. At this point, the model can more accurately predict the charging scheduling strategy for various types of input information, and can better consider the impact of various factors on charging scheduling.

[0055] Based on the grid dynamic electricity price information, carbon emission signal information, user charging demand information, historical charging and grid load problem information, grid information in different regions, federated learning privacy protection parameter information, and corresponding charging scheduling allocation result samples for verification, the trained Actor-Critic algorithm scheduling model is simulated for charging scheduling processing to generate verification results. The verification samples are input into the trained model, and the model outputs the charging scheduling prediction results for these users, such as predicting the charging power, charging time, and charging area of ​​user A in the current verification time period. The prediction results are compared with the actual charging scheduling allocation results, and relevant indicators are calculated, such as the mean square error between the predicted charging power and the actual charging power, and the error between the predicted charging time and the actual charging time, to generate verification results. These indicators are used to evaluate the performance of the model on the verification samples to determine whether the model can accurately predict the charging scheduling situation.

[0056] Based on the verification results, the trained Actor-Critic algorithm scheduling model is evaluated and adjusted to generate the scheduling model of the target Actor-Critic algorithm. If the verification results show that the model has a large error or low accuracy, for example, the mean square error exceeds the set threshold, it means that the model has overfitting or underfitting problems. If it is found that the model is underfitting on the verification sample (large error and low accuracy), the number of hidden layer neurons can be increased and retrained; if overfitting is found (good performance on the training sample, but poor performance on the verification sample), the learning rate can be appropriately reduced or the number of iterations can be reduced. After multiple evaluations and adjustments, the model achieves better performance on the verification sample, and finally generates the scheduling model of the target Actor-Critic algorithm. This target model can more accurately evaluate and make decisions on the charging scheduling of new energy vehicles, and provide strong support for the subsequent input of benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and power grid load problem information into the model for processing, and generate accurate charging scheduling allocation result information.

[0057] In another embodiment, the scheduling model based on the target Actor-Critic algorithm analyzes and processes the benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and grid load problem information to generate a charging resource demand associated evaluation value. Through the method shown in the previous article, the charging demand classification information and priority information of users A, B, and C, as well as the benefit evaluation parameter information and historical charging and grid load problem information are obtained. User A is a pure electric car with a current power of 30%. It plans to charge at a private charging pile at home at 18:30, which is a general charging demand; user B is an electric SUV with a remaining power of 20%. It charges at a public charging pile in a highway service area at 9:00 am on Saturday, which is an emergency charging demand; user C is an electric micro car with a power of 40%. It charges at a charging pile in the company parking lot at 22:00 in the evening, which is a delayable charging demand. The benefit evaluation parameter information shows that the current period of electricity price fluctuations are moderate, and changes in carbon emission intensity have little impact on costs. The historical charging and grid load problem information shows that the grid load in some areas is high from 18:00 to 20:00.

[0058] This information is input into the scheduling model of the target Actor-Critic algorithm. The model will comprehensively consider these factors and analyze the charging resource requirements of each user. For user A, the model takes into account that his charging time is during a period of high grid load and his power is not extremely low, and calculates his charging resource demand correlation evaluation value as 0.5 (the value range is 0-1, the higher the value, the more urgent the resource demand); for user B, because he is in an urgent charging demand state and charges during the peak hours on weekends, the model gives a charging resource demand correlation evaluation value of 0.8; user C has a relatively high power and charges during the low load period, and the charging resource demand correlation evaluation value is 0.3. These evaluation values ​​reflect the degree of demand for charging resources for each user under the current circumstances, taking into account factors such as user demand, grid conditions, electricity prices, and carbon emissions.

[0059] Based on the charging resource demand correlation evaluation value, the charging strategy decision vector within the target Actor-Critic algorithm scheduling model is optimized and adjusted to generate a charging scheduling deviation vector. Assuming that the original charging strategy decision vector of the model is [charging power adjustment, charging time adjustment, charging area selection], for user A, the original decision vector may be [maintain normal power, charge on time, current area]. Since its charging resource demand correlation evaluation value is 0.5, the model determines that the current strategy may need to be adjusted to balance the grid load and user demand. After optimization and adjustment, the generated charging scheduling deviation vector is [reduce charging power by 10%, postpone charging by 0.5 hours, maintain the current area].

[0060] For user B, due to the urgent demand, the model will increase the allocation of its charging resources. The original decision vector may be [normal power, charging as planned, current area], and the adjusted charging scheduling deviation vector is [increase charging power by 20%, prioritize charging, maintain the current area]. For user C, considering that its demand is not urgent and it has an advantage during the low-load period of the power grid, the original decision vector is [normal power, charging on time, current area], and the adjusted deviation vector is [maintain normal power, charging can be appropriately delayed by 0.5-1 hour, maintain the current area]. These deviation vectors reflect the direction and degree of adjustment of the model to the original charging strategy based on the associated evaluation value of the user's charging resource demand.

[0061] The charging scheduling deviation vector is parsed and converted to generate the charging scheduling allocation result information, wherein the charging scheduling allocation result information is used to characterize the grid resource information allocated to the charging of new energy vehicles and the target quantitative indicators of charging resource allocation. Taking user A as an example, the charging scheduling deviation vector is parsed, and the adjusted charging strategy is to start charging at 19:00, reduce the charging power by 10%, and charge at the private charging pile (current area) at home. At the same time, combined with the grid resource information, it is determined that the grid resources allocated to user A are part of the power resources in the household power consumption area during this period, and due to the power reduction, the charging time may be extended, but it will not affect the overall grid load balance. From the perspective of the target quantitative indicators of charging resource allocation, the charging cost of user A may be slightly reduced due to power adjustment and time delay, carbon emissions will also be reduced accordingly, and it will not cause too much pressure on the grid, ensuring the normal power consumption of other users.

[0062] For user B, after analyzing the deviation vector, it is determined that its charging strategy is to immediately charge at the public charging pile in the highway service area with a 20% increase in power. The grid resources allocated to user B are sufficient power in the service area under priority protection, ensuring that it can complete charging as soon as possible to meet the urgent needs of self-driving tours. In terms of target quantitative indicators, although the charging cost may increase due to the increase in power, it meets the urgent needs of users, and through reasonable scheduling, the impact on the overall power grid is within an acceptable range. The charging scheduling allocation result of user C is to start charging between 22:30-23:00, maintain normal power, and charge at the charging pile in the company's parking lot. The allocated grid resources come from the surplus power during the low-peak period of the industrial park at night, realizing the rational use of power resources. From the perspective of target quantitative indicators, the charging cost of user C is low, which plays a role in peak-shaving and valley-filling for the grid load, and also meets its demand for delayed charging. In this way, the generated charging scheduling and allocation result information clarifies the grid resource information allocated to each new energy vehicle user for charging, as well as the target quantitative indicators for charging resource allocation from multiple aspects such as cost, efficiency, and grid load balance, providing a basis for achieving efficient and reasonable charging scheduling.

[0063] The server obtains information such as dynamic power grid electricity prices, carbon emission signals, user charging needs, historical charging and power grid load problems, power grids in different regions, and federated learning privacy protection parameters. Next, feature extraction and quantitative analysis are performed on the dynamic power grid electricity prices and carbon emission signals, and the benefit evaluation parameter information is generated in combination with the privacy protection parameters. At the same time, historical charging and power grid load problem information and user charging demand information are processed to generate charging strategy adjustment reference information. Based on the federated learning privacy protection parameters and power grid information in different regions, the charging strategy adjustment reference information is processed to obtain the cross-regional collaborative optimization factor, which is used to generate charging demand classification information and priority information.

[0064] After that, by obtaining training and verification data, the preset Actor-Critic algorithm initial scheduling model is cleaned, feature extracted, normalized, optimized, trained and verified to obtain the target scheduling model. Finally, the benefit evaluation parameters, charging demand classification and priority information, and historical charging and grid load problem information are input into the target model. After analysis and processing, strategy vector optimization adjustment and deviation vector analytical conversion, the charging scheduling allocation result information is generated, covering the allocated grid resources and target quantitative indicators, realizing charging scheduling optimization that comprehensively considers cost, efficiency and grid stability, and improving the charging management level of new energy vehicles.

[0065] In one embodiment, if Figure 2 As shown, the present application also provides a charging scheduling device for automobile electricity price and carbon emission based on Actor-Critic, including: Acquisition module 201, used to obtain dynamic power price information of the power grid, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information; The processing module 202 is used to process the dynamic electricity price information and carbon emission signal information of the power grid based on the federated learning privacy protection parameter information to generate the benefit evaluation parameter information; process the user charging demand information based on the historical charging and power grid load problem information to generate the charging strategy adjustment reference information; process the charging strategy adjustment reference information based on the federated learning privacy protection parameter information and the power grid information of different regions to generate the cross-regional collaborative optimization factor; process the user charging demand information and the regional information corresponding to the user charging demand information based on the cross-regional collaborative optimization factor to generate the charging demand classification information and priority information, wherein the user charging demand information is used to characterize the user's vehicle type, charging time preference, battery power information and power grid access point information; input the benefit evaluation parameter information, charging demand classification information and priority information and the historical charging and power grid load problem information into the scheduling model based on the target Actor-Critic algorithm for processing to generate the charging scheduling allocation result information.

[0066] Each embodiment in this application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the charging scheduling method, electronic device, electronic device, and readable storage medium embodiment for evaluating the Actor-Critic-based automobile electricity price and carbon emissions, since they are basically similar to the above-mentioned embodiment of the charging scheduling method for automobile electricity price and carbon emissions based on Actor-Critic, the description is relatively simple, and the relevant parts can be referred to the partial description of the above-mentioned embodiment of the charging scheduling method for automobile electricity price and carbon emissions based on Actor-Critic.

Claims

1. A charging scheduling method for automobile electricity price and carbon emission based on Actor-Critic, characterized in that: include: Obtain dynamic power price information, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information; Based on the privacy protection parameter information of federated learning, the dynamic electricity price information and carbon emission signal information of the power grid are processed to generate the benefit evaluation parameter information; Process user charging demand information based on historical charging and grid load problem information to generate charging strategy adjustment reference information; Based on the privacy protection parameter information of federated learning and the information of power grids in different regions, the reference information for adjusting charging strategies is processed to generate cross-regional collaborative optimization factors. Based on the cross-regional collaborative optimization factor, the user charging demand information and the regional information corresponding to the user charging demand information are processed to generate charging demand classification information and priority information, wherein the user charging demand information is used to characterize the user's vehicle type, charging time preference, battery power information and grid access point information; The benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and grid load problem information are input into the scheduling model based on the target Actor-Critic algorithm for processing to generate charging scheduling allocation result information.

2. The method according to claim 1, characterized in that Based on the privacy protection parameter information of federated learning, the dynamic electricity price information and carbon emission signal information of the power grid are processed to generate the benefit evaluation parameter information, including: Perform feature extraction and processing on the dynamic electricity price information of the power grid to generate electricity price fluctuation amplitude characteristics, electricity price peak and valley time characteristics, and electricity price regional difference characteristics; Perform feature extraction and processing on carbon emission signal information to generate carbon emission intensity change characteristics, carbon emission peak period characteristics, and carbon emission trend characteristics in different regions; Quantitative analysis is performed on the characteristics of electricity price fluctuation amplitude, peak and valley period characteristics, and regional differences in electricity prices to generate quantitative characteristic values ​​of electricity prices; Quantitatively analyze and process the characteristics of carbon emission intensity changes, carbon emission peak period characteristics, and carbon emission trend characteristics in different regions to generate quantitative characteristic values ​​of carbon emissions; Based on the privacy protection parameter information of federated learning, the quantitative characteristic values ​​of electricity prices and carbon emissions are processed to generate comprehensive benefit evaluation characteristics and corresponding weight calculation result information; The comprehensive benefit evaluation characteristics and the corresponding weight calculation result information are processed to generate benefit evaluation parameter information, wherein the benefit evaluation parameter information is used to characterize the impact of the dynamic electricity price of the power grid and the carbon emission signal on the charging income of new energy vehicles.

3. The method according to claim 1, characterized in that Based on historical charging and grid load problem information, user charging demand information is processed to generate charging strategy adjustment reference information, including: Extract and classify historical charging and grid load problem information to generate historical charging power fluctuation information, grid load overload information, and abnormal charging time information; Based on the user's vehicle type, charging time preference, battery power information, and grid access point information contained in the user's charging demand information, the historical charging power fluctuation information, grid load overload information, and charging time abnormality information are processed to generate charging problem type judgment information and charging problem severity assessment information; Mark and filter target data in the user's charging demand information based on the charging problem type judgment information and the charging problem severity assessment information to generate a charging demand abnormal data filtering result; The abnormal charging demand data screening results are integrated and quantified to generate charging strategy adjustment reference information, where the charging strategy adjustment reference information is used to characterize the correlation between user charging demand and historical charging and grid load problems.

4. The method according to claim 1, characterized in that Based on the privacy protection parameter information of federated learning and the information of power grids in different regions, the reference information for adjusting the charging strategy is processed to generate cross-regional collaborative optimization factors, including: Extract target elements and classify data for charging strategy adjustment reference information to generate charging power adjustment information, charging time adjustment information, and charging area adjustment information; Based on the privacy protection parameter information of federated learning and the information of power grids in different regions, the charging power adjustment information, charging time adjustment information, and charging area adjustment information are comprehensively analyzed to generate adjustment strategy importance assessment information and adjustment strategy feasibility assessment information; Based on the adjustment strategy importance assessment information and the adjustment strategy feasibility assessment information, the key data in the charging strategy adjustment reference information is screened and associated to generate a charging strategy target data screening result; The charging strategy target data screening results are integrated and processed in combination with the privacy protection parameter information of federated learning and the weight information in the power grid information of different regions to generate a cross-regional collaborative optimization factor, where the cross-regional collaborative optimization factor is used to characterize the direction and degree of cross-regional collaborative optimization of new energy vehicle charging strategies in different regions.

5. The method according to claim 1, characterized in that Based on the cross-regional collaborative optimization factor, the user charging demand information and the regional information corresponding to the user charging demand information are processed to generate charging demand classification information and priority information, including: Extract and process user charging demand information to generate user vehicle type information, charging time preference information, battery power information, and grid access point information; Based on the regional information corresponding to the user's charging demand information, the user's vehicle type information, charging time preference information, battery power information, and grid access point information are standardized and integrated, and various types of information are converted into a unified processable format to generate a standardized charging demand information set; Based on the cross-regional collaborative optimization factors, the importance of the standardized charging demand information set is evaluated and screened, and the importance weight of each information in the charging demand classification and priority judgment is generated according to the optimization direction and degree reflected by the cross-regional collaborative optimization factors; The importance weight of each information in charging demand classification and priority judgment is processed to generate charging demand classification information and priority information.

6. The method according to claim 1, characterized in that Get the scheduling model of the target Actor-Critic algorithm, including: Obtain dynamic grid electricity price information, carbon emission signal information, user charging demand information, historical charging and grid load problem information, grid information in different regions, federated learning privacy protection parameter information, and corresponding charging scheduling allocation result samples for training; obtain similar information and samples for verification; obtain the initial scheduling model of the preset Actor-Critic algorithm; Perform data cleaning, feature extraction and normalization on the dynamic power grid electricity price information, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information used for training, and generate pre-processed training feature data, wherein the pre-processed training feature data includes electricity price fluctuation amplitude characteristics, electricity price peak and valley time characteristics, electricity price regional difference characteristics, carbon emission intensity change characteristics, carbon emission peak time characteristics, carbon emission trend characteristics in different regions, user vehicle type information, charging time preference information, battery power information, and power grid access point information; Processing the preprocessed training feature data to generate scheduling model prediction parameters, wherein the scheduling model prediction parameter vector is used to characterize the prediction information of the charging scheduling allocation results, the key factors affecting the charging scheduling, and the optimization strategy; Based on the scheduling model prediction parameter vector, the preset Actor-Critic algorithm's initial scheduling model is optimized and trained to generate a trained Actor-Critic algorithm scheduling model; Based on the dynamic electricity price information of the power grid, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, federated learning privacy protection parameter information and corresponding charging scheduling allocation result samples used for verification, the trained Actor-Critic algorithm scheduling model is simulated for charging scheduling processing to generate verification results; Based on the verification results, the trained Actor-Critic algorithm scheduling model is evaluated and adjusted to generate a scheduling model for the target Actor-Critic algorithm.

7. The method according to claim 6, characterized in that The benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and grid load problem information are input into the scheduling model based on the target Actor-Critic algorithm for processing to generate charging scheduling allocation result information, including: The dispatch model based on the target Actor-Critic algorithm analyzes and processes the benefit evaluation parameter information, charging demand classification information and priority information, and historical charging and grid load problem information to generate the charging resource demand correlation evaluation value; Based on the charging resource demand correlation evaluation value, the charging strategy decision vector within the target Actor-Critic algorithm scheduling model is optimized and adjusted to generate a charging scheduling deviation vector; The charging scheduling deviation vector is parsed and converted to generate charging scheduling allocation result information, wherein the charging scheduling allocation result information is used to characterize the grid resource information allocated to the charging of new energy vehicles and the target quantitative index of charging resource allocation.

8. A charging scheduling device for automobile electricity price and carbon emission based on Actor-Critic, characterized in that: The device comprises: The acquisition module is used to obtain dynamic power price information of the power grid, carbon emission signal information, user charging demand information, historical charging and power grid load problem information, power grid information in different regions, and federated learning privacy protection parameter information; A processing module is used to process the dynamic electricity price information and carbon emission signal information of the power grid based on the federated learning privacy protection parameter information to generate the benefit evaluation parameter information; process the user charging demand information based on the historical charging and power grid load problem information to generate the charging strategy adjustment reference information; process the charging strategy adjustment reference information based on the federated learning privacy protection parameter information and the power grid information of different regions to generate the cross-regional collaborative optimization factor; process the user charging demand information and the regional information corresponding to the user charging demand information based on the cross-regional collaborative optimization factor to generate the charging demand classification information and priority information, wherein the user charging demand information is used to characterize the user's vehicle type, charging time preference, battery power information and power grid access point information; input the benefit evaluation parameter information, charging demand classification information and priority information and the historical charging and power grid load problem information into the scheduling model based on the target Actor-Critic algorithm for processing to generate the charging scheduling allocation result information.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; The first processor is configured to execute the Actor-Critic based automobile electricity price and carbon emission charging scheduling method as described in any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the method for charging scheduling of automobile electricity price and carbon emission based on Actor-Critic is implemented.

Images