A Dynamic Adjustment Method for Time-of-Use Electricity Prices at Charging Stations Based on Multi-Objective Optimization

Abstract

This invention discloses a method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization, comprising the following steps: acquiring electric vehicle charging demand forecast data, real-time distribution network load data, and market electricity price data; based on the charging demand forecast data, real-time load data, and market electricity price data, constructing a multi-objective optimization model with charging station operator revenue, distribution network load curve smoothness, and user charging costs as optimization objectives; this method can comprehensively consider the interests of multiple parties, such as charging station operator revenue, distribution network load curve smoothness, and electric vehicle user charging costs, and achieves dynamic adaptive adjustment of time-of-use electricity prices by establishing a multi-objective optimization model and combining it with a real-time data-driven mechanism, thereby effectively solving the technical problems that traditional static time-of-use electricity pricing schemes cannot adapt to dynamic changes in the power system and are difficult to balance the interests of multiple parties.

Description

Technical Field

[0001] This invention relates to the technical field of power systems, and specifically to a method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization. Background Technology

[0002] With the ongoing global energy structure transformation and the increasingly clear goal of carbon neutrality, electric vehicles, as the core carrier of clean transportation, are experiencing explosive market growth. Against this backdrop, the optimization of the planning, construction, and operation management of charging infrastructure, a crucial link in the promotion and application of electric vehicles, has become a key issue of common concern to both academia and industry. As the core hub connecting the power grid and electric vehicle users, the operational efficiency of charging stations not only directly affects the charging experience and costs for electric vehicle users, but also profoundly influences the spatiotemporal distribution characteristics of electricity load and the safe and stable operation of the power distribution network. Electricity costs constitute a significant proportion of the operating costs of charging stations; therefore, the rationality of the electricity pricing mechanism becomes a core variable affecting the economic benefits and load regulation capabilities of charging stations.

[0003] In the early stages of the electric vehicle industry's development, charging stations generally adopted a fixed-price model for operation and pricing, meaning a uniform charging service price was applied throughout the entire period. The significant advantage of this pricing method lies in its simplicity and low management costs; users could make charging decisions without needing to consider time-of-day differences. However, with the continuous increase in electric vehicle penetration and the sustained expansion of charging load, the inherent shortcomings of the fixed-price model have become increasingly apparent. Due to the lack of a price signal-guided mechanism for user charging behavior, electric vehicle charging load exhibits a highly concentrated characteristic over time, with many users tending to charge during the evening hours after get off work. This further widens the peak-valley difference in the distribution network, exacerbating the pressure on power system operation safety caused by load curve fluctuations. At the same time, the fixed-price model fails to reflect the time-varying characteristics of electricity supply and demand. During periods of low grid load, it often results in idle charging capacity and wasted resources, while during peak load periods, it may lead to safety hazards such as local transformer overload.

[0004] To effectively alleviate the aforementioned contradictions, time-of-use pricing (TOUPricing) has been introduced into the electric vehicle charging field as a demand-side management tool. The basic principle of TOUPricing is to divide the day into multiple time periods based on the peak-valley distribution characteristics of the power grid load, and set differentiated electricity prices for each time period. This incentivizes users to adjust their charging behavior, achieving the goal of peak shaving and valley filling. In practice, typical TOUPricing schemes usually include three levels: peak price, flat price, and off-peak price. Peak-hour prices are significantly higher than off-peak prices, with price differences reaching several times. Theoretically, TOUPricing does improve the spatiotemporal distribution of charging load to some extent. Multiple empirical studies have shown that a well-designed TOUPricing scheme can effectively reduce peak load by approximately 15% to 25%, and correspondingly improve charging utilization during off-peak hours.

[0005] However, with the rapid growth of renewable energy generation capacity and the deepening of electricity market reforms, the limitations of traditional time-of-use pricing schemes have become increasingly apparent. From a technical perspective, current mainstream time-of-use pricing schemes generally adopt a static, fixed time-of-use division and price setting method. The time-of-use division scheme is usually determined empirically based on historical load data, and the price level is set based on cost accounting or reference price differences from the previous cycle. This static adjustment mechanism has a fundamental inherent contradiction: it cannot adapt to the dynamic evolution of multiple factors such as fluctuations in renewable energy output, changes in load characteristics, and fluctuations in market electricity prices. Specifically, as the penetration rate of renewable energy sources such as photovoltaics and wind power in the grid continues to increase, their output exhibits significant randomness and intermittency. This leads to a more complex fluctuation pattern in the grid's net load curve (i.e., the load curve after renewable energy output). Traditional time-of-use pricing schemes based on fixed time-of-use divisions struggle to respond promptly to such dynamic changes. On the other hand, the behavior patterns of electric vehicle users are also constantly evolving, with significant differences in charging demand distribution across different regions and time periods, making it difficult for static pricing schemes to achieve precise load control.

[0006] From a system optimization perspective, existing time-of-use (TOU) pricing schemes also suffer from a one-sidedness in setting their objective functions. The operation of charging stations involves the coordination and balance of multiple stakeholders: operators pursue maximum revenue and improved equipment utilization, the grid focuses on smoothing the load curve and ensuring system operational safety, and users expect reduced charging costs and improved service convenience. However, traditional TOU pricing schemes often emphasize the achievement of a single objective, lacking a comprehensive consideration of the interests of all parties. For example, simply pursuing load balancing may lead to excessively low prices during off-peak hours, eroding operator profits, while an overemphasis on maximizing operator revenue may weaken the guiding effect of price signals on user behavior. This inherent conflict and constraint among multiple objectives constitutes the core technical challenge facing the design of current TOU pricing schemes.

[0007] In summary, how to construct a time-of-use pricing dynamic adjustment mechanism that can comprehensively consider the interests of multiple parties, adapt to the dynamic changes in the power system, and achieve optimized control of the load curve has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0008] To address the aforementioned deficiencies in existing technologies, this invention provides a method for dynamically adjusting time-of-use electricity prices for charging stations based on multi-objective optimization. This method comprehensively considers the interests of multiple parties, including the revenue of charging station operators, the smoothness of the power distribution network load curve, and the charging costs for electric vehicle users. By establishing a multi-objective optimization model and combining it with a real-time data-driven mechanism, it achieves dynamic adaptive adjustment of time-of-use electricity prices, thereby effectively solving the technical problems that traditional static time-of-use electricity pricing schemes cannot adapt to dynamic changes in the power system and are difficult to balance the interests of multiple parties.

[0009] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: A method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization includes the following steps: acquiring electric vehicle charging demand forecast data, real-time distribution network load data, and market electricity price data; constructing a multi-objective optimization model based on the charging demand forecast data, real-time load data, and market electricity price data, with charging station operator revenue, distribution network load curve smoothness, and user charging costs as optimization objectives; transforming the multi-objective optimization model into a single-objective optimization problem using a weighted summation method, and solving it using a non-dominated sorting genetic algorithm to obtain an initial electricity price adjustment scheme; verifying the feasibility of the initial electricity price adjustment scheme, if it passes, sending the electricity price adjustment scheme to the charging station operation management system for execution, if it fails, returning to the optimization model for iterative solution; monitoring the actual effect data after the execution of the electricity price adjustment scheme in real time, and adaptively correcting the optimization model parameters based on the deviation between the actual effect data and the expected target.

[0010] In the step of obtaining electric vehicle charging demand forecast data, historical charging order data is first collected through the charging station operation management system. The historical charging order data includes charging start time, charging end time, charging amount, charging power, and user type information. Then, a charging demand time series is constructed based on the historical charging order data, and a Long Short-Term Memory (LSTM) network model is used to make short-term predictions of charging demand, generating charging demand forecast data for multiple future time windows. The LSM network model consists of an input layer, a two-layer LSM unit, an attention mechanism layer, and a fully connected output layer. The input dimension of the input layer is the time series feature dimension. The number of hidden layer nodes of the two-layer LSM unit is set to 128. The attention mechanism layer is used to weight the hidden states at different time steps. The fully connected output layer outputs the charging demand forecast values ​​for the future set time period.

[0011] In the step of acquiring real-time load data of the distribution network, real-time load data is collected by smart meters and sensors deployed at key nodes of the distribution network, including main transformers, feeder outlets, and important load access points. The collected real-time load data includes active power, reactive power, voltage amplitude, and frequency deviation. The data acquisition sampling period is set to 15 minutes, and the data transmission adopts the IEC61850 communication protocol, with the data format conforming to the IEC61970 General Information Model standard. After the collected real-time load data undergoes outlier removal, missing value imputation, and smoothing filtering by the data preprocessing module, it is stored in a real-time database for use by optimization algorithms.

[0012] In the step of obtaining market electricity price data, day-ahead market clearing price data is obtained from the power trading center as the benchmark value of market electricity price, and ancillary service price data for real-time balancing market is obtained from the power grid dispatching system. The time resolution of the market electricity price data is matched with the time window for time-of-use price adjustment, and the market electricity price data includes the expected price value and price fluctuation range for each time period. In addition, the opportunity cost corresponding to the blocked capacity of new energy is calculated based on the renewable energy output forecast data, and it is included in the correction factor of the market electricity price data.

[0013] In the step of constructing the multi-objective optimization model, the first optimization objective is to maximize the revenue of the charging station operator, and its objective function expression is: ; in, Indicates the first electric vehicles in the first Charging power during the period Indicates the first Electricity price for charging services during specific time periods. Indicates the first The electricity purchase price for electricity purchased from the grid during a specific time period. Indicates the length of the time period. This represents the total number of time periods within the optimization period. This indicates the total number of charging vehicles during the optimization period.

[0014] The second optimization objective is to maximize the smoothness of the distribution network load curve, and its objective function expression is as follows: ; in, Indicates the first Total load value of the distribution network during the time period This represents the average load value of the distribution network. The total number of time periods is represented; the smoothness of the load curve is quantified by the variance of the load curve, and the smaller the variance value, the smoother the load curve.

[0015] The third optimization objective is to minimize the user's charging cost, and its objective function expression is: ; The meanings of each variable are the same as those in the first objective function; user charging cost is defined as the total charging fee paid by the user during the optimization period.

[0016] In the step of transforming the multi-objective optimization model into a single-objective optimization problem using the weighted summation method, the weight coefficients of each objective are first set according to the weight preferences of the three stakeholders: the charging station operator, the power distribution network dispatching department, and the user. These weight coefficients are denoted as follows: And satisfy Then construct the weighted objective function: ; in, These are the maximum values ​​of the three objective functions during their respective single-objective optimization, used to normalize each objective function; the operator revenue objective is optimized in the positive direction, while the load smoothness objective and user cost objective are optimized in the negative direction.

[0017] In the step of obtaining the initial electricity price adjustment scheme based on the non-dominated sorting genetic algorithm, the algorithm uses real number encoding to encode the electricity price decision variables, with each decision variable corresponding to the electricity price adjustment coefficient for a time period; the initial population is generated randomly within the feasible domain of electricity prices, and the population size is set to 100; the fitness function is the weighted objective function F; the selection operation adopts a binary tournament selection strategy, the crossover operation adopts a simulated binary crossover operator with a crossover probability set to 0.9, and the mutation operation adopts a polynomial mutation operator with a mutation probability set to 0.1; the maximum number of iterations of the algorithm is set to 500 times. When the change in fitness value of the best individual for three consecutive generations is less than the set threshold aZvZC8Ev43fmzz0V.answerText0^{-6}$, the algorithm converges and outputs the current optimal solution as the initial electricity price adjustment scheme.

[0018] In the step of verifying the feasibility of the initial electricity price adjustment plan, the verification includes verification of electricity price constraints, load constraints, and equipment safety constraints. The electricity price constraint requires that the electricity price in each time period is not lower than the lower limit and not higher than the upper limit. The lower limit is set at 1.05 times the purchase price, and the upper limit is set at 3 times the purchase price. The load constraint requires that the total load of the distribution network after the charging load is superimposed does not exceed the rated capacity of the transformer and is not lower than the minimum technical output. The equipment safety constraint requires that the charging power does not exceed the rated power of the charging pile and the charging current does not exceed the cable current carrying capacity. If any constraint condition is not met, the plan is marked as infeasible and returned to the optimization model for iterative solution.

[0019] In the step of sending the electricity price adjustment plan to the charging station operation management system for execution, the optimized electricity price plan is first used to generate an electricity price adjustment instruction set in chronological order. The instruction set includes the start time, end time, and corresponding electricity value for each time period. Then, the electricity price adjustment instruction set is sent to the charging station operation management system through the data interface. The charging station operation management system updates the charging service price for each time period according to the received electricity price adjustment instructions and pushes the price change information to electric vehicle users through the user information publishing system.

[0020] In the step of real-time monitoring of the actual effect data after the implementation of the electricity price adjustment plan, actual charging data is collected by energy management terminals deployed on the charging station side. The actual charging data includes the actual charging amount, actual charging power, actual charging cost and user response behavior data for each time period. At the same time, actual load data is collected by the distribution network monitoring system, and the variance of the actual load curve is calculated. The collected actual effect data is compared with the target value expected by the optimization model, and the deviation rate is calculated.

[0021] In the step of adaptively correcting the optimization model parameters based on the deviation between the actual effect data and the expected target, a rolling optimization strategy under the model prediction control framework is adopted. After each optimization cycle, the optimization model parameters for the next cycle are adjusted based on the actual effect data of the previous cycle. The parameter correction adopts the gradient descent method, and the correction direction is to reduce the deviation between the actual effect and the expected target. The learning rate is set to 0.01. At the same time, after several optimization cycles, the long short-term memory network prediction model is incrementally trained using accumulated historical data to update the network weights and improve the accuracy of charging demand prediction.

[0022] Furthermore, in a preferred embodiment of the present invention, the multi-objective optimization model further includes a fourth optimization objective, namely, maximizing the utilization rate of charging station equipment, the objective function of which is expressed as: ; in, Indicates the first Actual charging capacity at the charging station during the specified time period. Indicates the first The rated charging capacity of the charging station during certain hours This indicates the total number of time periods; equipment utilization rate is defined as the ratio of actual charging capacity to rated rechargeable capacity.

[0023] Furthermore, in another preferred embodiment of the present invention, the non-dominated sorting genetic algorithm introduces an elite retention strategy during the solution process, directly copying the top 10 individuals in each generation of the population to the next generation of the population to prevent the optimal solution from being lost during the evolution process; at the same time, a feasibility repair operation is added after the crossover operation to repair the infeasible individuals generated by the crossover so that they meet the constraints.

[0024] Furthermore, in another preferred embodiment of the present invention, the timing of the issuance of the electricity price adjustment plan adopts a rolling optimization strategy, performing an optimization calculation every 15 minutes. Each optimization calculation generates an electricity price adjustment plan for the next 4 hours. After each optimization calculation is completed, the latest electricity price adjustment plan is issued and executed, while canceling the instructions generated by the previous optimization calculation that have not yet been executed.

[0025] Furthermore, in another preferred embodiment of the present invention, when a sudden change in the distribution network load or a large fluctuation in renewable energy output is detected, an emergency optimization mode is triggered. In this mode, the time window of the optimization cycle is shortened to 5 minutes, and the weighting coefficient of the load smoothness target is increased to above 0.5 to quickly respond to changes in system status.

[0026] Furthermore, in another preferred embodiment of the present invention, the collection of user response behavior data is achieved through a communication module deployed on the charging pile side. The communication module records the user's response behavior to electricity price adjustments, including whether the user's charging behavior changes after receiving a price change notification and the specific form of the change. Based on the collected user response behavior data, a user price sensitivity model is constructed to predict the degree of user response to different electricity price adjustment schemes.

[0027] Furthermore, in another preferred embodiment of the present invention, the input features of the long short-term memory network model further include date type features, weather features, and temperature features; date type features include weekday, weekend, and holiday labels; weather features include sunny, cloudy, overcast, and rainy / snowy weather labels; temperature features include maximum temperature, minimum temperature, and average temperature values; these features are input into the network model together with historical charging demand data after being processed by one-hot encoding or normalization.

[0028] Furthermore, in another preferred embodiment of the present invention, the weight coefficients of each objective function in the multi-objective optimization model are determined by fuzzy hierarchical analysis. First, a judgment matrix is ​​constructed by expert scoring. Then, the eigenvectors of the judgment matrix are calculated as the subjective weights of each objective. Finally, the objective weights are combined and weighted to obtain the final weight coefficients.

[0029] Furthermore, in another preferred embodiment of the present invention, the feasibility verification module further includes a charging demand satisfaction verification, requiring that the user charging demand satisfaction rate caused by the optimized electricity price scheme is not lower than a set threshold, whereby the user charging demand satisfaction rate is defined as the ratio of users who can complete charging within the expected time.

[0030] Furthermore, in another preferred embodiment of the present invention, when there is a conflict between multiple optimization objectives, making it impossible to find an optimal solution that simultaneously satisfies all objectives, the Pareto optimal solution set is used for description, the Pareto front is obtained by solving the non-dominated sorting genetic algorithm, and the optimal solution set on the Pareto front is presented to the operation decision-makers, who then select the final electricity price adjustment scheme based on actual operation needs.

[0031] Furthermore, in another preferred embodiment of the present invention, the system further includes an electricity price pre-release module, which releases electricity price information for future periods to users a certain time before the electricity price adjustment plan is officially implemented, so that users can plan their charging behavior in advance; the pre-release time is determined according to the average user response time, and the pre-release duration is adjustable from 30 minutes to 2 hours.

[0032] Furthermore, in another preferred embodiment of the present invention, the outlier removal method used in the data preprocessing module is the 3σ principle based on box plots, the missing value imputation method is the linear interpolation method based on historical data of the same type, the smoothing filter adopts the Savitzky-Golay filter, the filter window length is set to 5, and the polynomial order is set to 2.

[0033] Furthermore, in another preferred embodiment of the present invention, the system further includes a historical database for storing historical charging data, historical load data, historical electricity price data, and historical optimization result data; the historical database adopts a hybrid storage architecture combining relational database and time-series database, with the relational database storing user information and device information, and the time-series database storing time-series operational data.

[0034] Furthermore, in another preferred embodiment of the present invention, the charging demand prediction model also integrates an online learning mechanism, which can update the model parameters in real time based on newly collected actual charging data. The online learning method adopts an incremental learning method based on a sliding window, with the window length set to the amount of data for one week.

[0035] Furthermore, in another preferred embodiment of the present invention, the non-dominated sorting genetic algorithm adopts adaptive crossover probability and mutation probability during the evolution process. The crossover probability and mutation probability are dynamically adjusted according to the population diversity index. When the population diversity is lower than a set threshold, the mutation probability is appropriately increased to increase the population diversity. When the population diversity is higher than the set threshold, the crossover probability is appropriately increased to accelerate the convergence speed.

[0036] Furthermore, in another preferred embodiment of the present invention, a penalty factor is introduced into the objective function of the optimization model to handle constraint violations. The penalty factor is set using a dynamic penalty factor strategy, with the initial value of the penalty factor set to 10, and it increases exponentially to a set upper limit as the number of iterations increases.

[0037] Furthermore, in another preferred embodiment of the present invention, the system further includes a visualization module for displaying in real time the current electricity pricing scheme, the achievement of optimization targets, load curve comparisons, and user response status; the visualization module supports access from both web and mobile devices, and the display interface adopts a responsive design.

[0038] Furthermore, in another preferred embodiment of the present invention, when a charging station participates in the power ancillary services market, the objective function of the optimization model needs to be supplemented with an ancillary service revenue term, which includes peak shaving revenue, frequency regulation revenue, and reserve capacity revenue. The ancillary service revenue is determined by calculating the difference between the call cost obtained from participating in ancillary services and the charging revenue forgone.

[0039] Furthermore, in another preferred embodiment of the present invention, the acquisition of market electricity price data further includes acquiring new energy output forecast data from a renewable energy power forecasting system, calculating the new energy consumption space based on the new energy output forecast results, and incorporating this space as a constraint into the optimization model. When there is a surplus of new energy output, the electricity price during off-peak hours is appropriately reduced to incentivize charging loads to consume new energy.

[0040] Furthermore, in another preferred embodiment of the present invention, the time-of-use electricity price dynamic adjustment system for charging stations includes a data acquisition layer, a data processing layer, an optimization decision-making layer, and an execution control layer; the data acquisition layer is responsible for collecting electric vehicle charging demand data, distribution network load data, market electricity price data, and user behavior data; the data processing layer is responsible for preprocessing, feature extraction, and demand forecasting of the collected raw data; the optimization decision-making layer is responsible for constructing a multi-objective optimization model, solving the optimization problem, and generating an electricity price adjustment scheme; the execution control layer is responsible for distributing the electricity price adjustment scheme to the charging station operation management system and monitoring the execution effect.

[0041] Furthermore, in another preferred embodiment of the present invention, the data acquisition layer and the data processing layer communicate using a message queue. The message queue uses the Kafka distributed message middleware, and the message topics include charging demand data topics, load data topics, electricity price data topics, and behavior data topics. The data producers and consumers of each topic are decoupled using a publish-subscribe model.

[0042] Furthermore, in another preferred embodiment of the present invention, the optimization algorithm of the optimization decision layer is deployed in a containerized manner. Each optimization task runs in an independent Docker container. The resource configuration of the container is dynamically adjusted according to the scale of the optimization problem. The computing resources include the number of CPU cores and the size of memory. The containers achieve task scheduling and result aggregation through a distributed coordination service.

[0043] Furthermore, in another preferred embodiment of the present invention, the interface between the execution control layer and the charging station operation management system adopts the OpenADR2.0 standard protocol to realize bidirectional interaction of electricity price signals. The charging station operation management system adjusts the charging service price according to the received electricity price signal, and at the same time feeds back the user's charging response behavior to the optimization decision layer to achieve closed-loop control.

[0044] Furthermore, in another preferred embodiment of the present invention, when a charging station manages multiple distributed charging stations simultaneously, the optimization model is solved using a distributed optimization architecture. Each charging station is equipped with a local optimizer. The local optimizer performs preliminary optimization calculations based on local data and local optimization objectives, and then uploads the local optimization results to the global optimizer for coordination. The global optimizer generates the globally optimal electricity price adjustment scheme after integrating the local optimization results.

[0045] Furthermore, in another preferred embodiment of the present invention, the distributed optimization architecture uses the alternating direction multiplier method to achieve coordinated optimization among charging stations. The local optimizer of each charging station minimizes the local cost function under local constraints. Global constraints are coordinated by introducing Lagrange multipliers. During the iteration process, each local optimizer and the global optimizer exchange Lagrange multiplier information and local gradient information through an encrypted communication link.

[0046] Furthermore, in another preferred embodiment of the present invention, the system further includes a security module for encrypting and protecting the transmission and storage of electricity price data. The data transmission adopts the TLS1.3 protocol, and the data storage adopts the AES-256 encryption algorithm. At the same time, an anomaly detection mechanism is set up to automatically trigger a security alarm and start an emergency response process when an anomaly in the electricity price data or an attack on the system is detected.

[0047] Furthermore, in another preferred embodiment of the present invention, the solution results of the multi-objective optimization model are output in a structured data format. The output content includes the start time, end time, electricity price adjustment coefficient, absolute value of electricity price, and predicted value of the achievement of optimization objectives for each time period. The data format adopts JSON format and contains complete metadata information.

[0048] Furthermore, in another preferred embodiment of the present invention, when the power grid dispatching department issues a load control instruction, the optimization decision layer automatically corrects the optimization model after receiving the load control instruction, adds load control constraints to the model, including the maximum load value and the duration of load reduction, and the optimization model re-solves the electricity price adjustment scheme under the premise of satisfying the load control constraints.

[0049] Furthermore, in another preferred embodiment of the present invention, the user charging behavior response model is constructed using a user-based collaborative filtering algorithm. This algorithm calculates the behavioral similarity between users based on historical charging data, and then predicts the response probability of the target user to the new electricity price based on the historical response behavior of similar users. The response probability is used as a parameter input into a multi-objective optimization model to improve the accuracy of the optimization results.

[0050] Furthermore, in another preferred embodiment of the present invention, the hyperparameters of the long short-term memory network model are automatically tuned using a Bayesian optimization method. The hyperparameters include the number of hidden layer nodes, learning rate, dropout ratio, and batch size. The objective function of the Bayesian optimization is the root mean square error of the model on the validation set. The hyperparameter space is modeled and iteratively updated using a Gaussian process surrogate model.

[0051] Furthermore, in another preferred embodiment of the present invention, the system further includes a data quality monitoring module, which performs real-time quality detection on the raw data collected by the data acquisition layer. The data quality indicators include data integrity, data consistency, data timeliness, and data accuracy. When the data quality indicators are lower than a set threshold, a data repair or data source switching process is automatically triggered.

[0052] Furthermore, in another preferred embodiment of the present invention, the algorithm for solving the optimization model can also employ a deep neural network method based on reinforcement learning. This method constructs a deep Q-network as a policy network. The network input is the current system state, including load data, electricity price data, and charging demand data. The network output is the electricity price adjustment action for each time period. By continuously interacting with the environment to update the network weights, the long-term cumulative benefits are maximized.

[0053] Furthermore, in another preferred embodiment of the present invention, when using a reinforcement learning-based optimization method, the design of the reward function comprehensively considers three dimensions: operator revenue, load smoothness, and user cost. The expression for the reward function is: ; in, Rewards related to operator revenue, Rewards related to load smoothness, Rewards related to user costs, The weighting coefficients for each reward item are used; the reinforcement learning training process employs an experience replay mechanism and target network technology to improve training stability and convergence speed.

[0054] Furthermore, in another preferred embodiment of the present invention, the system also integrates blockchain technology for trusted storage of electricity price data. After the electricity price adjustment scheme is generated, it is encapsulated into data blocks and timestamped and digitally signed. The blocks are linked by hash values ​​to form an immutable blockchain ledger, providing reliable technical support for electricity price dispute resolution and audit traceability.

[0055] Furthermore, in another preferred embodiment of the present invention, when there are multiple competing charging station operators in the area where the charging station is located, the optimization model introduces competitive factors to construct a game model. The electricity price decision of each operator is regarded as the strategy choice of the game participants. Each participant seeks to maximize its own benefits under rational conditions, and the equilibrium solution of the game is used as the final electricity price adjustment scheme.

[0056] Furthermore, in another preferred embodiment of the present invention, the game model is modeled using the Cournot competition model, where each charging station competes for output with the goal of maximizing its own profit. In the present invention, output competition is transformed into electricity price competition. The optimal electricity price strategy of each charging station is obtained by solving the Cournot equilibrium. The solution of the Cournot equilibrium adopts an iterative algorithm based on gradient ascent.

[0057] Furthermore, in another preferred embodiment of the present invention, the dynamic electricity price adjustment method can also be extended to other demand-side management scenarios, including industrial and commercial user load management, distributed energy storage scheduling, and air conditioning load regulation. In these application scenarios, only the objective function and constraints of the optimization model need to be adjusted accordingly, while keeping the core framework of multi-objective optimization and dynamic adjustment unchanged.

[0058] Furthermore, in another preferred embodiment of the present invention, when the electric vehicle adopts the battery swapping mode, the objective function of the optimization model needs to add battery inventory balance constraints and battery charging and discharging efficiency constraints. The battery inventory balance constraints require that the battery inventory quantity of each battery swapping station be maintained within a set range, and the battery charging and discharging efficiency constraints take into account the energy loss during the battery charging and discharging process.

[0059] Furthermore, in another preferred embodiment of the present invention, the system also supports bidirectional interaction with the power grid dispatching system. When an emergency occurs on the power grid side and the load needs to be adjusted quickly, the power grid dispatching system sends a load control request to the charging station. After receiving the request, the charging station prioritizes responding to the control request and adjusts the electricity price to achieve a rapid load response. The economic losses incurred during the response process are compensated by the power grid side according to the pre-signed ancillary service agreement.

[0060] Furthermore, in another preferred embodiment of the present invention, the solution results of the multi-objective optimization model can also serve as a reference for investment planning decisions for charging stations. By analyzing the equipment utilization and revenue under different electricity price schemes, data support can be provided for the capacity configuration and site layout of charging stations.

[0061] Furthermore, in another preferred embodiment of the present invention, the dynamic electricity price adjustment system supports data docking with the upper-level energy management platform. The energy management platform can issue aggregated optimization targets to the charging station, and the charging station adjusts the parameters of the local optimization model according to the received aggregated targets to achieve multi-energy complementarity and source-grid-load-storage coordinated optimization.

[0062] Furthermore, in another preferred embodiment of the present invention, when the charging station is equipped with distributed photovoltaic and energy storage facilities, the optimization model needs to add photovoltaic output prediction and energy storage charging and discharging strategy optimization. The photovoltaic output prediction adopts a long short-term memory network method similar to charging demand prediction, and the energy storage charging and discharging strategy optimization adopts a dynamic programming algorithm to achieve the economic optimal operation of the energy storage equipment.

[0063] Furthermore, in another preferred embodiment of the present invention, the objective function of the energy storage charging and discharging strategy optimization includes maximizing the energy storage charging and discharging revenue, which includes electricity price arbitrage revenue and ancillary service revenue. The constraints include energy storage capacity constraints, charging and discharging power constraints, and charging and discharging cycle number constraints. The number of energy storage charging and discharging cycles within the optimization period does not exceed a set upper limit to extend the life of the energy storage device.

[0064] Furthermore, in another preferred embodiment of the present invention, the photovoltaic power output prediction model and the charging demand prediction model adopt a dual-tower architecture with a shared feature extraction layer. The two models share the parameters of the bottom feature extraction network, and only output the photovoltaic power output prediction value and the charging demand prediction value independently at the top output layer. This parameter sharing method can improve the prediction accuracy while reducing the total number of model parameters.

[0065] Furthermore, in another preferred embodiment of the present invention, when the charging station is equipped with both photovoltaic and energy storage, the joint optimization of electricity price optimization and energy storage scheduling adopts a two-layer optimization framework. The upper layer optimization determines the time-of-use electricity price scheme, and the lower layer optimization optimizes the energy storage charging and discharging strategy according to the electricity price scheme determined by the upper layer. The upper and lower layers are solved iteratively until the objectives of both layers converge to the optimal.

[0066] Compared with the prior art, the present invention has the following beneficial effects: First, by establishing a multi-objective optimization model that comprehensively considers the interests of charging station operators, the smoothness of the distribution network load curve, and users' charging costs, the problem of one-sided decision-making in traditional static time-of-use pricing schemes is effectively solved, and the overall balance and coordinated optimization of the interests of all parties is achieved.

[0067] Second, by using a non-dominated sorting genetic algorithm combined with a weighted summation method for multi-objective optimization, a set of optimal solutions can be obtained on the Pareto front, providing operational decision-makers with a flexible choice space to select the most suitable electricity price adjustment scheme according to actual operational needs.

[0068] Third, by introducing a real-time data-driven mechanism and a rolling optimization strategy, the electricity price adjustment scheme can be adaptively adjusted according to the dynamic changes in the power system status, effectively overcoming the shortcomings of the traditional static time-of-use pricing scheme that cannot respond to fluctuations in renewable energy output and changes in load characteristics.

[0069] Fourth, by applying long short-term memory networks to electric vehicle charging demand forecasting, the accuracy of short-term charging demand forecasting was improved, providing a reliable data foundation for the construction of optimization models, thereby significantly improving the accuracy and effectiveness of electricity price adjustment schemes.

[0070] Fifth, by introducing equipment utilization rate targets and incorporating them into a multi-objective optimization model, the utilization efficiency of charging station equipment is improved while ensuring operator revenue and load control effectiveness, thus achieving a balance between economic and social benefits.

[0071] Sixth, by constructing a user price sensitivity model and using it as a parameter input to optimize the model, the accuracy of predicting user response behavior is improved, enabling the electricity price adjustment scheme to better guide user charging behavior and achieve the expected peak shaving and valley filling effect.

[0072] Seventh, by integrating blockchain technology to achieve credible storage of electricity price data, it provides technical support for electricity price dispute resolution and audit traceability, and enhances the credibility and transparency of the system.

[0073] Eighth, by supporting a distributed optimization architecture, the system achieves collaborative optimization of electricity pricing schemes for multiple charging stations, ensuring the local interests of each station while achieving global optimization, thus improving the scalability and applicability of the system. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0075] Figure 1 This is a flowchart of the method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to the present invention. Detailed Implementation

[0076] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0077] Please see Figure 1 As shown in the figure, this embodiment discloses a method for dynamically adjusting the time-of-use electricity price of charging stations based on multi-objective optimization. The technical solution of the present invention will be described in detail below with reference to specific embodiments.

[0078] The system architecture of this invention adopts a four-layer design, from bottom to top: a data acquisition layer, a data processing layer, an optimization decision layer, and an execution control layer. The data acquisition layer is responsible for acquiring various types of raw data from external systems. The data processing layer preprocesses, extracts features, and forecasts demand from the raw data. The optimization decision layer constructs a multi-objective optimization model based on the processed data and solves it to generate an electricity price adjustment scheme. The execution control layer is responsible for distributing the electricity price adjustment scheme to the charging station operation and management system and monitoring its implementation. Each layer seamlessly connects with the others through standardized data interfaces and communication protocols, forming a complete closed loop of data and control flows.

[0079] In the specific implementation of the data acquisition layer, the acquisition of electric vehicle charging demand forecasting data first involves collecting historical charging order data through the charging station operation management system. Historical charging order data includes five core fields: charging start time, charging end time, charging amount, charging power, and user type information. The charging start and end times are recorded in a standard timestamp format accurate to the second; the charging amount is recorded in kilowatt-hours; the charging power is recorded in kilowatts; and the user type information is divided into four categories: private users, taxi users, ride-hailing users, and logistics vehicle users. The data collection time span must include at least six months of complete charging records to ensure the sufficiency and representativeness of the training data.

[0080] After acquiring historical charging order data, a charging demand time series is constructed based on this data. Specifically, using hours as the basic time granularity, the historical charging order data is aggregated and statistically analyzed in chronological order to obtain the total charging volume, total number of charging sessions, and average charging power within each hourly segment, forming a three-dimensional time series feature vector. The first dimension of the three-dimensional time series feature vector is the total charging volume sequence, the second dimension is the total number of charging sessions sequence, and the third dimension is the average charging power sequence. The constructed time series data is divided into a training set, a validation set, and a test set according to chronological order, with the training set accounting for 70%, the validation set accounting for 15%, and the test set accounting for 15%.

[0081] Charging demand prediction employs a Long Short-Term Memory (LSTM) network model for short-term forecasting. The LTM network model consists of four components: an input layer, two LTM units, an attention mechanism layer, and a fully connected output layer. The input layer's input dimension is the time-series feature dimension, specifically three dimensions, corresponding to three features: total charging capacity, total number of charging cycles, and average charging power. The number of hidden nodes in the two LTM units is set to 128. Each LTM unit includes three gating mechanisms: a forget gate, an input gate, and an output gate. The activation function for the forget gate, input gate, and output gate is the sigmoid function. Cell state updates use the hyperbolic tangent function as the activation function. Residual connections are used between the two LTM units to mitigate the gradient vanishing problem during deep network training.

[0082] An attention mechanism layer is positioned between the two-layer Long Short-Term Memory (LSTM) unit and the fully connected output layer to weight the hidden states at different time steps. The attention mechanism is implemented using additive attention, specifically comprising two weight matrices and one attention vector. Both weight matrices are 128x128, and the attention vector is 128. The attention score is calculated as score(h_t, s) = v^Ttanh(W_hh_t + W_ss), where h_t represents the hidden state of the LSM unit at time step t, s represents the current state of the decoder, W_h and W_s are the corresponding weight matrices, and v is the attention vector. After calculating the attention score for each time step, a Softmax function is used for normalization to obtain the attention weights. These attention weights are then weighted and summed with the hidden states at the corresponding time steps to obtain the context vector.

[0083] The fully connected output layer consists of one fully connected layer and one output layer. The fully connected layer has 64 neurons, and the ReLU activation function is used. The output layer has the same number of neurons as the prediction time window, and the identity function is used. The overall output of the model is the predicted charging demand for a given future time period. The prediction time window is set to four hours, meaning it predicts the charging demand for each of the next four hours. During training, the Adam optimizer is used for parameter updates, with a learning rate of 0.001, a batch size of 32, and 200 training epochs. The early stopping strategy involves stopping training if the validation set loss does not decrease for 20 consecutive epochs.

[0084] In a specific embodiment, the input features of the Long Short-Term Memory (LSTM) network model also include date type features, weather features, and temperature features. Date type features are represented using one-hot encoding, containing three category labels: weekday, weekend, and holiday, represented by a three-dimensional binary vector. Weather features are also represented using one-hot encoding, containing four category labels: sunny, cloudy, overcast, and rainy / snowy, represented by a four-dimensional binary vector. Temperature features are normalized, mapping temperature values ​​to the range of 0.1 to 0.9, specifically calculated using the formula T_normalized = (T - T_min) / (T_max - T_min) × 0.8 + 0.1, where T is the actual temperature value, and T_min and T_max are the historical minimum and maximum temperatures, respectively. These features, after one-hot encoding or normalization, are input into the network model along with historical charging demand data. The total dimension of the input feature vector is thirteen dimensions: three-dimensional time series features, three-dimensional date type features, four-dimensional weather features, and three-dimensional temperature features.

[0085] Real-time load data for the distribution network is acquired through smart meters and sensors deployed at key nodes. The selection of key nodes follows the principle of covering the main transformer, each feeder outlet, and important load connection points. The main transformer node is located at the substation outlet and is used to monitor the active power, reactive power, voltage amplitude, and frequency deviation on both the primary and secondary sides of the transformer. Feeder outlet nodes are located at the beginning of each feeder and are used to monitor the load status of each feeder. Important load connection points include large industrial and commercial users, distributed photovoltaic (PV) connection points, and energy storage power station connection points, and are used to monitor the real-time operating status of these important loads.

[0086] The data acquisition sampling period is set to fifteen minutes, meaning real-time load data for each key node is collected every fifteen minutes. Data transmission uses the IEC 61850 communication protocol, which supports both sampled value transmission and general object-oriented substation event transmission mechanisms. The data format conforms to the IEC 61970 General Information Model standard, organized using a standard data object model, containing information at three levels: class name, attribute, and attribute value. The collected real-time load data undergoes outlier removal, missing value imputation, and smoothing filtering by the data preprocessing module before being stored in the real-time database for use by optimization algorithms. The real-time database uses the time-series database InfluxDB, and the data write frequency is consistent with the sampling period, occurring every fifteen minutes.

[0087] The outlier removal method used in the data preprocessing module is based on the three-standard-deviation principle of box plots. Specifically, for each monitoring indicator, the mean and standard deviation of its historical data are calculated, and data points exceeding the range of the mean plus or minus three standard deviations are identified as outliers and removed. The missing value imputation method is based on linear interpolation of similar historical data. When data at a certain time point is missing, linear interpolation is performed using data from two time points before and after that point to fill in the missing value. The smoothing filter uses a Savitzky-Golay filter with a filter window length of five and a polynomial order of two. This filter can effectively filter out high-frequency noise while maintaining the signal trend.

[0088] Market electricity price data is obtained from the day-ahead market clearing price data of the power trading center as the benchmark value for market electricity prices. This data is acquired through integration with the power trading center's trading system. The data includes expected prices and price fluctuation ranges for each time period, with the time resolution set to the hourly level to match the time window for time-of-use price adjustments. Simultaneously, ancillary service price data for the real-time balancing market is obtained from the grid dispatching system. This data includes three categories: peak-shaving ancillary service prices, frequency regulation ancillary service prices, and reserve capacity ancillary service prices. Furthermore, the opportunity cost corresponding to the hindered renewable energy capacity is calculated based on renewable energy output forecast data. This calculation is based on the difference between the forecasted renewable energy output and the actual absorbable capacity. The opportunity cost is calculated by multiplying the hindered capacity by the day-ahead market clearing price for the current time period, and this opportunity cost is incorporated as a correction factor into the market electricity price data.

[0089] The multi-objective optimization model is constructed with the following optimization objectives: charging station operator revenue, distribution network load curve smoothness, and user charging cost. The first optimization objective is to maximize the charging station operator revenue, and its objective function expression is: ; in, This represents the charging power of the i-th electric vehicle in time period t, in kilowatts. This represents the electricity price for charging services during time period t, expressed in yuan per kilowatt-hour. The objective function represents the electricity purchase price from the grid during time period t, expressed in yuan per kilowatt-hour. Δt indicates the time period length is set to one hour, T indicates the total number of time periods within the optimization period is set to ninety-six (i.e., the next four hours), and N indicates the total number of charging vehicles within the optimization period, output by the charging demand forecasting model. The economic meaning of this objective function is that the net revenue obtained by the charging station from providing charging services to electric vehicle users equals sales revenue minus the electricity purchase cost.

[0090] The second optimization objective is to maximize the smoothness of the distribution network load curve, and its objective function expression is as follows: ; in, This represents the total load value of the distribution network in time period t, in kilowatts. The average load value of the distribution network is represented by , and T represents the total number of time periods. The objective function is quantified using the variance of the load curve; a smaller variance indicates a smoother load curve. The load curve smoothness index reflects the effect of charging load on the peak-to-valley difference of the distribution network load curve. The optimization objective is to make the charging load complementary to the original load of the distribution network, that is, to increase the charging load during peak load periods to reduce the peak-to-valley difference, and to reduce the charging load during off-peak periods to fill the valley.

[0091] The third optimization objective is to minimize the user's charging cost, and its objective function expression is: ; The meanings of the variables are the same as those in the first objective function. User charging cost is defined as the total charging fee paid by the user during the optimization period, and the optimization objective is to minimize this cost to improve user satisfaction.

[0092] The weighted summation method transforms a multi-objective optimization model into a single-objective optimization problem. The specific implementation process is as follows: First, based on the weight preferences of the three stakeholders—charging station operators, power grid dispatching departments, and users—the weight coefficients for each objective are set, denoted as [equation missing]. And satisfy In one specific embodiment, the operator weight is set to 0.4, the load smoothness weight to 0.35, and the user cost weight to 0.25. The weight coefficients can also be determined using fuzzy hierarchical analysis. Specifically, a judgment matrix is ​​first constructed through expert scoring, and the elements of the judgment matrix are assigned values ​​using the Saaty nine-scale method. Then, the eigenvectors of the judgment matrix are calculated as the subjective weights of each objective. Finally, the objective weights are combined and weighted to obtain the final weight coefficients.

[0093] The formula for constructing the weighted objective function is: ; in, These are the maximum values ​​of the three objective functions during their respective single-objective optimization, used to normalize each objective function. The operator revenue objective is optimized in the positive direction, meaning the larger the objective function value, the better; the load smoothness objective and user cost objective are optimized in the negative direction, meaning the smaller the objective function value, the better. Therefore, these two objective functions are negatively marked in the weighted objective function.

[0094] The specific implementation of the non-dominated sorting genetic algorithm to obtain the initial electricity price adjustment scheme uses real-number encoding to encode the electricity price decision variables. Each decision variable corresponds to an electricity price adjustment coefficient for a specific time period, with the coefficient ranging from 0.95 to 2.95. This corresponds to a lower limit of 95% of the purchased electricity price and an upper limit of 295% of the purchased electricity price. The initial population is generated randomly within the feasible electricity price domain, with a population size of 100. The fitness function is the weighted objective function F.

[0095] The selection operation employs a binary tournament selection strategy. Each time, two individuals are randomly selected from the population, their fitness values ​​are compared, and the individual with the higher fitness value is selected for the next generation. This process is repeated until a number of individuals equal to the population size are selected. The crossover operation uses a simulated binary crossover operator with a crossover probability of 0.9 and a distribution exponent of 20. The mutation operation uses a polynomial mutation operator with a mutation probability of 0.1 and a distribution exponent of 20.

[0096] The algorithm's maximum number of iterations is set to 500, and the convergence criterion is that the change in the fitness value of the best individual across three consecutive generations is less than a set threshold of 10 to the power of -6. When the convergence criterion is met, the algorithm converges and outputs the current optimal solution as the initial electricity price adjustment scheme.

[0097] In a preferred embodiment, the non-dominated sorting genetic algorithm incorporates an elite preservation strategy during the solution process, directly replicating the top ten best individuals from each generation into the next generation to prevent the optimal solution from being lost during evolution. Simultaneously, a feasibility repair operation is added after the crossover operation to correct infeasible individuals resulting from crossover. Specifically, the repair method involves adjusting the values ​​of decision variables that exceed the feasible region to the nearest feasible region boundary value. The algorithm employs an adaptive adjustment strategy for the crossover and mutation probabilities, dynamically adjusting them based on the population diversity index. The population diversity index is calculated using the standard deviation of the Euclidean distance between decision variables among individuals in the population. When the population diversity is below a set threshold of 0.1, the mutation probability is increased to 0.2; when the population diversity is above a set threshold of 0.5, the crossover probability is increased to 0.95.

[0098] The feasibility verification module performs a feasibility verification on the initial electricity price adjustment plan, including three aspects: electricity price constraint verification, load constraint verification, and equipment safety constraint verification. The electricity price constraint requires that the electricity price at any time period not be lower than the lower limit and not exceed the upper limit. The lower limit is set at 105% of the purchased electricity price, and the upper limit is set at 300% of the purchased electricity price. The load constraint requires that the total load of the distribution network after the charging load is superimposed not to exceed the transformer's rated capacity and not to be lower than the minimum technical output. The transformer's rated capacity is determined based on the nameplate parameters of the substation's main transformer, and the minimum technical output is set at 20% of the rated capacity. The equipment safety constraint requires that the charging power not exceed the rated power of the charging pile and the charging current not exceed the cable's current carrying capacity. The rated power of the charging pile is determined based on the charging pile equipment model, and the cable's current carrying capacity is determined based on the cable specifications and laying method. If any constraint condition is not met, the plan is marked as infeasible and returned to the optimization model for iterative solution.

[0099] The feasibility verification module also includes a charging demand satisfaction verification, requiring that the optimized electricity pricing scheme results in a user charging demand satisfaction rate that is not lower than a set threshold. The user charging demand satisfaction rate is defined as the percentage of users who can complete charging within the expected time, and the set threshold is 85%. When the charging demand satisfaction rate is lower than this threshold, the scheme is determined to be infeasible and returned to the optimization model for iterative solution.

[0100] After the initial electricity price adjustment plan passes feasibility verification, it is issued to the charging station operation and management system for execution. The issuance process begins by generating an electricity price adjustment instruction set from the optimized electricity price plan in chronological order. This instruction set includes the start and end times of each time period, along with the corresponding electricity value. The instruction set is encapsulated in JSON format. The JSON object contains a `timeSlots` array field, with each array element containing three attributes: `startTime`, `endTime`, and `price`.

[0101] Then, the electricity price adjustment instruction set is sent to the charging station operation and management system via its data interface. The charging station operation and management system updates the charging service prices for each time period based on the received instructions and pushes the price change information to electric vehicle users through the user information publishing system. The user information publishing system supports three methods: SMS push, APP push, and WeChat push, allowing users to choose their preferred method. The price change information includes the effective time of the price adjustment, the adjusted unit price, and an explanation of the reason for the price adjustment.

[0102] Real-time monitoring of the actual effects of the electricity price adjustment plan is achieved through energy management terminals deployed at charging stations. The energy management terminals collect actual charging data including actual charging volume, actual charging power, actual charging cost, and user response behavior data for each time period. The accuracy of actual charging volume collection is required to be no less than 99%, and the collection period for actual charging power is five minutes. The actual charging cost is calculated by multiplying the actual charging volume by the actual electricity price. User response behavior data is collected through communication modules deployed at the charging piles. These modules record user responses to the electricity price adjustment, including whether and how charging behavior changes after receiving price change notifications. Changes include four categories: earlier charging time, later charging time, changed charging power, and canceled charging. Simultaneously, actual load data is collected through the distribution network monitoring system, and the variance of the actual load curve is calculated. The collected actual effect data is compared with the target value expected by the optimization model to calculate the deviation rate.

[0103] The optimization model parameters are adaptively adjusted based on the deviation between actual performance data and expected targets, employing a rolling optimization strategy within the model predictive control framework. After each optimization cycle, the optimization model parameters for the next cycle are adjusted based on the actual performance data from the previous cycle. Parameter adjustment uses gradient descent, with the adjustment direction aimed at reducing the deviation between actual performance and expected targets, and the learning rate is set to 0.01. Simultaneously, after several optimization cycles, the long short-term memory network prediction model is incrementally trained using accumulated historical data to update network weights and improve the accuracy of charging demand prediction. Incremental training employs a sliding window-based online learning method, with a window length set to one week (168 hours) of data. Each incremental training iteration uses data from the most recent window for parameter updates.

[0104] In one specific embodiment, the timing of the electricity price adjustment plan is optimized using a rolling optimization strategy. An optimization calculation is performed every fifteen minutes, generating a price adjustment plan for the next four hours. After each optimization calculation, the latest price adjustment plan is issued and executed, while any unexecuted instructions generated in the previous optimization calculation are cancelled. This rolling optimization strategy ensures that the electricity price adjustment plan remains consistent with the current system state, preventing plan failure due to changes in system state.

[0105] When a sudden change in distribution network load or a significant fluctuation in renewable energy output is detected, an emergency optimization mode is triggered. In this mode, the optimization cycle time window is shortened to five minutes, and the weighting coefficient of the load smoothness target is increased to above 0.5. The criteria for determining a sudden load change are a change in the total distribution network load exceeding 20% ​​of its rated capacity within ten minutes, and a significant fluctuation in renewable energy output are a change in the predicted photovoltaic or wind power output exceeding 30% of its rated capacity within ten minutes. In emergency optimization mode, other parameters of the optimization algorithm remain unchanged; the rapid response to changes in system state is achieved by shortening the optimization cycle and increasing the load smoothness weighting.

[0106] The system's overall architecture employs standardized technical solutions at each layer. Data communication between the data acquisition layer and the data processing layer utilizes a message queue, specifically the Kafka distributed message middleware. Message topics include charging demand data, load data, electricity price data, and behavioral data. Data producers and consumers for each topic are decoupled using a publish-subscribe model. The Kafka cluster is deployed with three nodes, each configured with an eight-core CPU, 16GB of memory, and 500GB of SSD storage.

[0107] The optimization algorithm for the decision-making layer is deployed using a containerized approach, with each optimization task running in an independent Docker container. The container's resource configuration is dynamically adjusted based on the scale of the optimization problem, including the number of CPU cores and memory size. An optimization task with a population size of 100 is configured with two CPU cores and four GB of memory, while an optimization task with a population size of 200 is configured with four CPU cores and eight GB of memory. Task scheduling and result aggregation between containers are achieved through a distributed coordination service implemented using a Zookeeper cluster.

[0108] The interface between the execution control layer and the charging station operation management system adopts the OpenADR2.0 standard protocol to achieve bidirectional interaction of electricity price signals. The OpenADR2.0 protocol defines the event types and response mechanisms for electricity price signals. The charging station operation management system adjusts the charging service price based on the received electricity price signals and simultaneously feeds back user charging response behavior to the optimization decision layer to achieve closed-loop control. Interface security is ensured using the OAuth2.0 authentication mechanism.

[0109] The system also includes a visualization module for real-time display of current electricity pricing schemes, optimization target achievement status, load curve comparisons, and user response status. The visualization module supports both web and mobile access. The web version is developed using the Vue.js framework, and the mobile version uses the Flutter framework. The display interface employs a responsive design to adapt to different screen sizes. The historical database stores historical charging data, historical load data, historical electricity price data, and historical optimization results data. The historical database uses a hybrid storage architecture combining relational and time-series databases. The relational database uses PostgreSQL to store user and device information, while the time-series database uses InfluxDB to store time-series operational data.

[0110] The security module encrypts and protects the transmission and storage of electricity price data. Data transmission uses the TLS 1.3 protocol, and data storage uses the AES-256 encryption algorithm. An anomaly detection mechanism is also implemented, automatically triggering security alarms and initiating emergency response procedures when abnormal electricity price data or system attacks are detected. Anomaly detection combines rule-based and machine learning-based methods. Rule-based methods detect obvious abnormal data patterns, such as electricity prices exceeding reasonable ranges, while machine learning methods detect covert abnormal behaviors, such as unusual access patterns.

[0111] The technical effects of the present invention will be described in detail below through specific embodiments.

[0112] Example 1: A city's charging station cluster uses the time-of-use pricing dynamic adjustment method described in this invention for electricity price management. This city's charging station cluster comprises ten charging stations with a total of 500 charging spaces and a total installed capacity of 50 megawatts. The historical data collection period is six months, including complete charging records, load data, and electricity price data from the past six months.

[0113] Data collected from the data acquisition layer shows that the average daily charging volume of this charging station cluster is approximately 8,000 kWh, with peak charging periods concentrated between 10:00 AM and 12:00 PM and between 6:00 PM and 10:00 PM, and off-peak charging periods concentrated between midnight and 6:00 AM. Distribution network load data shows that the average daily load of the distribution network in this area is approximately 100 MW, with peak load periods largely overlapping with peak charging periods, resulting in a peak-to-valley difference of approximately 40%.

[0114] A long short-term memory (LSTM) network model was used for charging demand prediction. The model's input features included thirteen dimensions: three-dimensional time series features, three-dimensional date type features, four-dimensional weather features, and three-dimensional temperature features. The model was trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 32. After 200 training epochs, the root mean square error on the test set was 125 kWh, and the mean absolute percentage error was 7.8%. This prediction accuracy meets the requirements of optimization decision-making for the accuracy of the input data.

[0115] The weight coefficients of the multi-objective optimization model were determined using the fuzzy hierarchical analysis method. By inviting five industry experts to score the model, a judgment matrix was constructed and the eigenvector was calculated. The weights for operator revenue were determined to be 0.42, load smoothness to be 0.35, and user cost to be 0.23.

[0116] A non-dominated sorting genetic algorithm was used for optimization, with a population size of 100, a crossover probability of 0.9, a mutation probability of 0.1, and a maximum number of iterations of 500. The optimization computation took approximately 32 seconds on the server, meeting real-time requirements.

[0117] After the optimized electricity price adjustment plan was implemented, and following a month of operational testing, the actual results showed that: the average monthly revenue of charging station operators increased by 23.5% compared to before implementation; the peak-valley difference rate of the distribution network load curve decreased from 40% to 26%; and the average monthly charging cost for users decreased by 8.2%. All three optimization objectives achieved significant improvements.

[0118] Comparative Example 1: A comparative test was conducted using the traditional static time-of-use pricing scheme. The traditional static time-of-use pricing scheme divides a day into three periods: peak, flat, and valley. The peak period is from 10:00 AM to 4:00 PM, the flat period is from 4:00 PM to 10:00 PM, and the valley period is from 10:00 PM to 10:00 AM the next day. The electricity prices are 1.2 yuan, 0.8 yuan, and 0.4 yuan per kilowatt-hour, respectively.

[0119] Data from the same period using the traditional static time-of-use pricing scheme shows that: the average monthly revenue of charging station operators is 18.3% lower than that of the scheme proposed in this invention; the peak-valley difference rate of the distribution network load curve only decreased from 40% to 35%; and the average monthly charging cost for users is 5.7% higher than that of the scheme proposed in this invention. The traditional static scheme cannot adaptively adjust to changes in real-time load and charging demand, and is inferior to the scheme proposed in this invention in all three indicators: operator revenue, load smoothness, and user cost.

[0120] Example 2: Based on Example 1, a fourth optimization objective is introduced: maximizing the utilization rate of charging station equipment. The objective function expression for the equipment utilization rate objective is: ; in, This represents the actual charging amount at the charging station during time period t. Let t represent the rated rechargeable capacity of the charging station in time period t, and T represent the total number of time periods. Equipment utilization rate is defined as the ratio of actual charging capacity to rated rechargeable capacity.

[0121] After incorporating equipment utilization into the multi-objective optimization model, the weight allocation of the four-objective optimization model was adjusted as follows: operator revenue weight 0.35, load smoothness weight 0.30, user cost weight 0.20, and equipment utilization weight 0.15. Using a weighted summation method, the four-objective optimization model was transformed into a single-objective optimization problem. The resulting electricity price adjustment scheme achieved a significant improvement in equipment utilization.

[0122] Actual operational data shows that the utilization rate of charging station equipment increased from 45% to 62%, an increase of 37.8%. Meanwhile, the operator's revenue, load smoothing, and user costs remained near the levels of Example 1, achieving a balance between economic and social benefits.

[0123] Comparative Example 2: A comparison was made between an optimization scheme without an equipment utilization target and a four-target optimization scheme with an equipment utilization target. The results show that, in terms of operator revenue, the two schemes are essentially equal, with a difference of no more than 2%; in terms of load smoothness, the four-target scheme is slightly better than the three-target scheme, with the peak-to-valley difference rate decreasing from 26% to 24%; in terms of user costs, the difference between the two schemes is no more than 3%; and in terms of equipment utilization, the four-target scheme is significantly better than the three-target scheme, with a utilization rate increase of 37.8%. These results validate the necessity and effectiveness of introducing an equipment utilization target.

[0124] Example 3: Building upon Example 1, this example further considers the scenario of charging stations participating in the electricity ancillary services market. Ancillary service revenue is added to the objective function of the optimization model. Ancillary service revenue includes three components: peak shaving revenue, frequency regulation revenue, and reserve capacity revenue. Peak shaving revenue is determined by calculating the difference between the call cost obtained from participating in peak shaving ancillary services and the charging revenue forgone. Frequency regulation revenue is determined by calculating the clearing price of participating in frequency regulation ancillary services and the actual call capacity. Reserve capacity revenue is determined by calculating the opportunity cost of reserving reserve capacity.

[0125] Actual operational data shows that after participating in the electricity ancillary services market, charging stations generate an average monthly ancillary service revenue of 28,000 yuan, accounting for approximately 12% of total revenue. Participation in the ancillary services market requires adding corresponding constraints to the optimization model, including the maximum available power for ancillary services, response time requirements, and duration requirements. Under the condition of satisfying the ancillary service constraints, the optimized electricity price adjustment scheme can still achieve the three basic optimization objectives, while also generating additional ancillary service revenue.

[0126] Example 4: Based on Example 1, consider a scenario where a charging station is equipped with distributed photovoltaic (PV) and energy storage facilities. The PV capacity is 500 kW, and the energy storage capacity is 500 kWh. A two-layer optimization framework is used for joint optimization of electricity price and energy storage scheduling. The upper-layer optimization determines the time-of-use electricity price scheme, and the lower-layer optimization optimizes the energy storage charging and discharging strategy based on the electricity price scheme determined by the upper layer.

[0127] The photovoltaic (PV) power output prediction employs a long short-term memory (LSTM) network approach similar to that used for charging demand prediction. The model structure is identical to the charging demand prediction model, and the output is the predicted PV power output for the next four hours. Both the PV power output prediction model and the charging demand prediction model use a dual-tower architecture with a shared feature extraction layer. The parameters of the bottom-level feature extraction network are shared, while the top-level output layer independently outputs the predicted PV power output and charging demand values. This parameter-sharing approach reduces the total number of model parameters by approximately 30% while improving prediction accuracy by approximately 5%.

[0128] The energy storage charging and discharging strategy optimization adopts a dynamic programming algorithm. The objective function includes maximizing the revenue from energy storage charging and discharging, which includes electricity price arbitrage revenue and ancillary service revenue. Constraints include an energy storage capacity constraint of 500 kWh, a charging and discharging power constraint of 250 kW, a charging and discharging cycle constraint of no more than once per day, and an energy storage charging and discharging cycle number not exceeding a set upper limit within the optimization period to extend the lifespan of the energy storage equipment.

[0129] Actual operational data shows that after equipping the charging station with photovoltaic and energy storage facilities, the self-sufficiency rate of the charging station increased from 20% to 45%, and the average monthly income from energy storage participating in electricity price arbitrage reached 12,000 yuan. The overall income of the charging station further increased by 15% compared with Example 1.

[0130] Example 5: Considering a scenario of collaborative optimization across multiple distributed charging stations. The ten charging stations in this city's charging station cluster employ a distributed optimization architecture for collaborative optimization. Each charging station has a local optimizer. The local optimizer performs preliminary optimization calculations based on local data and local optimization objectives, and then uploads the local optimization results to the global optimizer for coordination. The global optimizer generates a globally optimal electricity price adjustment plan after integrating the results of all local optimizations.

[0131] The distributed optimization architecture employs the alternating direction multiplier method to achieve coordinated optimization among charging stations. Each charging station's local optimizer minimizes its local cost function under local constraints, while global constraints are addressed through the introduction of Lagrange multipliers. During iteration, each local optimizer and the global optimizer exchange Lagrange multiplier information and local gradient information via an encrypted communication link. The communication link is encrypted using the TLS 1.3 protocol to ensure data transmission security.

[0132] Actual operational data shows that the distributed optimization architecture can effectively coordinate electricity pricing decisions across charging stations, achieving global optimization while protecting the local interests of each station. Compared to the centralized optimization scheme, the distributed optimization scheme improves computational efficiency by approximately 40%, while the difference in the global target value of the optimization results does not exceed 3%.

[0133] A specific embodiment of the present invention also includes the construction of a user response behavior model. This model employs a user-based collaborative filtering algorithm. This algorithm calculates the behavioral similarity between users based on historical charging data, and then predicts the response probability of target users to the new electricity price based on the historical response behavior of similar users. Specifically, users are first clustered according to their charging behavior patterns, grouping users with similar charging behaviors into one category. Then, the response probability distribution of each category of users to different electricity price levels is calculated. The response probability is used as a parameter input to a multi-objective optimization model. Considering the probability distribution of user responses during the optimization process allows the electricity price adjustment scheme to better guide user charging behavior and achieve the expected peak shaving and valley filling effects.

[0134] In constructing the user price sensitivity model, the collected user response behavior data includes whether users' charging behavior changes after receiving price change notifications, and the specific form of such changes. The specific forms of changes are categorized into four types: earlier charging time, later charging time, change in charging power, and cancellation of charging. Based on the collected user response behavior data, a user price sensitivity model is constructed, and the model output is the probability distribution of user responses under different electricity price levels. This model can predict the degree of user response to different electricity price adjustment schemes, providing a basis for optimizing the model's parameter settings.

[0135] The specific implementation of this invention also includes the integrated application of blockchain technology. After generation, the electricity price adjustment scheme is encapsulated into data blocks and timestamped and digitally signed. The data block structure consists of a block header and a block body. The block header contains the hash value, timestamp, version number, and Merkle root of the previous block, while the block body contains the electricity price adjustment scheme data. The blocks are linked by hash values ​​to form an immutable blockchain ledger, providing reliable technical support for electricity price dispute resolution and audit traceability. The blockchain network is deployed in a consortium blockchain format, with members including charging station operators, power grid companies, and user representatives.

[0136] The specific embodiments of the present invention also include alternative schemes for reinforcement learning optimization methods. When using a reinforcement learning-based optimization method, a deep Q-network is constructed as the policy network. The network input is the current system state, including load data, electricity price data, and charging demand data. The network output is the electricity price adjustment action for each time period. The deep Q-network adopts a convolutional neural network structure, containing three convolutional layers and two fully connected layers. The number of convolutional kernels in the convolutional layers are 32, 64, and 64, respectively; the kernel sizes are 8x8, 4x2, and 3x1, respectively; and the stride is 4, 2, and 1, respectively. The number of neurons in the fully connected layers is 512, and the action space size is [not specified].

[0137] The reinforcement learning training process employs an experience replay mechanism and a target network technique to improve training stability and convergence speed. The experience replay pool has a capacity of 100,000 experience samples, and the batch size is set to 32. The target network is updated every 1,000 steps. The reward function is designed by comprehensively considering three dimensions: operator revenue, load smoothness, and user cost. The reward function expression is: ; in, For operator revenue-related incentives, the calculation method is the ratio of actual revenue to target revenue. The reward item related to load smoothness is calculated as the reciprocal of the peak-to-valley difference rate. For user cost-related rewards, the calculation method is the ratio of target cost to actual cost. α, β, and γ are the weighting coefficients for each reward item, set to 0.4, 0.35, and 0.25, respectively.

[0138] In practical engineering applications, the system deployment of this invention must meet the following technical requirements. The hardware devices for the data acquisition layer include smart meters, sensors, and data acquisition units. The accuracy level of the smart meters must be no less than 0.5%, the measurement accuracy of the sensors no less than 1%, and the sampling frequency of the data acquisition unit must be no less than once per second. The server configuration for the data processing layer must be no less than an eight-core CPU, 16GB of RAM, and 500GB of SSD storage. The computing server configuration for the optimization decision layer must be no less than a sixteen-core CPU, 32GB of RAM, and one GPU accelerator card. The execution control layer needs to maintain a stable network connection with the charging station operation management system, the power grid dispatching system, and user terminals, with a network latency not exceeding 100 milliseconds.

[0139] Through the detailed description of the above specific embodiments, the technical solution of this invention can achieve the following technical effects: By establishing a multi-objective optimization model that comprehensively considers the interests of charging station operators, the smoothness of the distribution network load curve, and users' charging costs, a balanced and coordinated optimization of the interests of all parties is achieved; by using a non-dominated sorting genetic algorithm combined with a weighted summation method for multi-objective optimization, a set of optimal solutions can be obtained on the Pareto front, providing operational decision-makers with a flexible selection space; by introducing a real-time data-driven mechanism and a rolling optimization strategy, the electricity price adjustment scheme can be adaptively adjusted according to the dynamic changes in the power system state; by applying a long short-term memory network to electric vehicle charging demand forecasting, the accuracy of short-term charging demand forecasting is improved, providing a reliable data foundation for the construction of the optimization model.

[0140] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization, characterized in that, include: Acquire electric vehicle charging demand forecast data, real-time power grid load data, and market electricity price data; Based on the aforementioned charging demand forecast data, real-time load data, and market electricity price data, a multi-objective optimization model is constructed with charging station operator revenue, distribution network load curve smoothness, and user charging cost as optimization objectives. The multi-objective optimization model is transformed into a single-objective optimization problem using a weighted summation method, and the initial electricity price adjustment scheme is obtained by solving the problem using a non-dominated sorting genetic algorithm. The feasibility of the initial electricity price adjustment plan is verified. If it passes, the electricity price adjustment plan is sent to the charging station operation and management system for execution. If it fails, the optimization model is returned for iterative solution. The actual effect data after the implementation of the electricity price adjustment plan is monitored in real time, and the optimization model parameters are adaptively corrected based on the deviation between the actual effect data and the expected target.

2. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of obtaining electric vehicle charging demand forecast data, historical charging order data is first collected through the charging station operation management system. The historical charging order data includes charging start time, charging end time, charging amount, charging power and user type information. Then, a charging demand time series is constructed based on the historical charging order data, and a long short-term memory network model is used to make short-term predictions of charging demand, generating charging demand forecast data for multiple future time windows. The structure of the Long Short-Term Memory (LSTM) network model includes an input layer, a two-layer LSM unit, an attention mechanism layer, and a fully connected output layer. The number of hidden layer nodes in the two-layer LSM unit is set to 128, and the attention mechanism layer is used to perform weighted processing on the hidden states at different time steps.

3. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 2, characterized in that, The input features of the Long Short-Term Memory Network model also include date type features, weather features, and temperature features; the date type features use one-hot encoding to represent weekday, weekend, and holiday labels; the weather features use one-hot encoding to represent sunny, cloudy, overcast, and rainy / snowy weather labels; and the temperature features are normalized.

4. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of acquiring real-time load data of the distribution network, real-time load data is collected by smart meters and sensors deployed at key nodes of the distribution network, including main transformers, feeder outlets, and important load access points; the collected real-time load data includes active power, reactive power, voltage amplitude, and frequency deviation; the data acquisition sampling period is set to fifteen minutes, and the data transmission adopts the IEC61850 communication protocol.

5. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of constructing the multi-objective optimization model, the first optimization objective is to maximize the revenue of the charging station operator, and its objective function expression is: ； The second optimization objective is to maximize the smoothness of the distribution network load curve, and its objective function expression is as follows: ； The third optimization objective is to minimize the user's charging cost, and its objective function expression is: 。 6. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, The multi-objective optimization model also includes a fourth optimization objective, namely, maximizing the utilization rate of charging station equipment, the objective function of which is: 。 7. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of transforming the multi-objective optimization model into a single-objective optimization problem using the weighted summation method, the weight coefficients of each objective are set according to the weight preferences of the three stakeholders: the charging station operator, the power distribution network dispatching department, and the user. These weight coefficients are denoted as follows: And satisfy ; The weighted objective function is constructed as follows: 。 8. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of obtaining the initial electricity price adjustment scheme based on the non-dominated sorting genetic algorithm, the algorithm uses real number encoding to encode the electricity price decision variables. The initial population is generated randomly within the feasible region of electricity prices, and the population size is set to one hundred. The selection operation adopts a binary tournament selection strategy, the crossover operation adopts a simulated binary crossover operator with a crossover probability of 0.9, and the mutation operation adopts a polynomial mutation operator with a mutation probability of 0.

1. The maximum number of iterations of the algorithm is set to five hundred. When the fitness value of the best individual in three consecutive generations changes less than a set threshold, the algorithm converges and outputs the current optimal solution as the initial electricity price adjustment scheme.

9. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of verifying the feasibility of the initial electricity price adjustment plan, the verification includes electricity price constraint verification, load constraint verification, and equipment safety constraint verification. The electricity price constraint requires that the electricity price in each time period is not lower than the lower limit and not higher than the upper limit. The lower limit is set at 105% of the purchased electricity price, and the upper limit is set at 300% of the purchased electricity price. The load constraint requires that the total load of the distribution network after the charging load is superimposed does not exceed the rated capacity of the transformer and is not lower than the minimum technical output. The equipment safety constraint requires that the charging power does not exceed the rated power of the charging pile and the charging current does not exceed the cable current carrying capacity.

10. The method for dynamic adjustment of time-of-use electricity prices for charging stations based on multi-objective optimization according to claim 1, characterized in that, In the step of real-time monitoring of the actual effect data after the implementation of the electricity price adjustment plan, actual charging data is collected by energy management terminals deployed on the charging station side. The actual charging data includes the actual charging amount, actual charging power, actual charging cost and user response behavior data for each time period. Simultaneously, actual load data is collected through the power distribution network monitoring system, and the variance of the actual load curve is calculated; the collected actual effect data is compared with the target value expected by the optimization model, and the deviation rate is calculated. In the step of adaptively correcting the optimization model parameters based on the deviation between the actual effect data and the expected target, a rolling optimization strategy under the model predictive control framework is adopted. After each optimization cycle, the optimization model parameters for the next cycle are adjusted based on the actual effect data of the previous cycle. The parameter correction adopts the gradient descent method, and the learning rate is set to 0.01.

Images