Method and system for motivating participation of electric vehicle user in demand response based on Internet of Vehicles

Abstract

The invention provides an excitation system and method for participation of an electric vehicle user in demand response based on the Internet of Vehicles. The system comprises a user response cost evaluation module, a user grouping module, an incentive mechanism design module, an incentive mechanism optimization module and a trust guarantee module, and the user response cost evaluation module calculates response cost of each user based on Internet of Vehicles data and user behavior data. The user grouping module divides each user into different groups based on the Internet of Vehicles data and the user behavior data, the incentive mechanism design module designs a multi-level incentive mechanism based on the response cost of the user and the grouping result, and the incentive mechanism optimization module optimizes the incentive parameters of each level of incentive mechanism based on a reinforcement learning method. And the trust guarantee module records the user response data and the distribution result of the incentive mechanism based on the block chain technology. The system can accurately quantify dominant and implicit costs, realizes differential excitation, and can also realize dynamic adjustment of an excitation strategy.

Description

Technical Field

[0001] The present application belongs to the field of smart grid technology, and specifically relates to an incentive system and method for electric vehicle users to participate in demand response based on the Internet of Vehicles. Background Art

[0002] In recent years, with the rapid adoption of electric vehicles, their fleet size has continued to expand, presenting new challenges and opportunities for the power system. On the one hand, the uncertainty and volatility of electric vehicle charging loads exacerbate the peak-shaving pressure on the power grid. On the other hand, as controllable loads, electric vehicles have the potential to participate in demand response, providing flexibility resources for the power system. Demand response is an important demand-side management method that uses price signals or incentive mechanisms to guide users to adjust their electricity consumption behavior in response to changes in power system demand. The development of connected vehicle technology provides technical support for real-time monitoring, analysis, and control of electric vehicle charging behavior, creating conditions for the development of refined demand response incentive mechanisms.

[0003] Currently, the following main charging and discharging management solutions exist: First, an electric vehicle demand response mechanism based on time-of-use electricity pricing. Time-of-use electricity prices are designed based on grid load conditions, and price differences are used to guide electric vehicles to charge during off-peak hours, thereby reducing system peak load. This solution uses a unified time-of-use electricity price, so all users face the same price signal, making the system simple and easy to implement. Second, a direct load control solution based on incentive subsidies. Electric vehicle users sign demand response contracts, allowing the system to directly control charging power during specific periods in exchange for a fixed subsidy. This solution categorizes users into different categories and sets differentiated subsidy amounts based on the degree of control allowed. Third, a dynamic incentive mechanism based on game theory treats demand response participants as game players. By designing incentive functions, the system reaches a Nash equilibrium, achieving a balance between system benefits and user utility. This solution considers the mutual influence between users and can, to a certain extent, avoid the "herd effect." Fourth, a crowdsourcing-based incentive mechanism has been studied. This introduces a bidding mechanism that allows electric vehicle users to proactively declare their response capabilities and price expectations. The system then selects appropriate participants based on cost-effectiveness. This solution achieves efficient resource allocation through market-based mechanisms and can reveal the true response costs.

[0004] Although existing technologies have made some progress in promoting the participation of electric vehicles in demand response, they still have the following obvious shortcomings: First, existing incentive mechanisms generally lack an accurate assessment of user response costs, especially the lack of quantification methods for implicit costs such as the loss of convenience caused by charging delays and additional battery cycle losses, resulting in a lack of reasonable basis for incentive design. Second, most existing solutions adopt static or semi-static incentive strategies, which are difficult to adapt to the dynamic changes in electric vehicle user behavior patterns and grid demand, and the incentive efficiency is low. Third, existing technologies ignore the cultivation of users' enthusiasm for long-term participation, lack a comprehensive evaluation of users' response contributions and a long-term incentive mechanism, resulting in insufficient stability in user participation. Fourth, existing incentive designs fail to fully utilize the real-time data resources provided by the Internet of Vehicles, lack in-depth exploration of key information such as users' charging habits, driving patterns, and response willingness, and lack incentive precision. Summary of the Invention

[0005] The technical problem solved by this application is: how to provide an intelligent charging and discharging optimization system with high incentive accuracy, dynamic changes according to demand and long-term incentive.

[0006] The present application provides an incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles, the incentive system comprising:

[0007] A user response cost evaluation module, which calculates the response cost of each user based on Internet of Vehicles data and user behavior data;

[0008] A user grouping module, which divides each user into different groups based on Internet of Vehicles data and user behavior data;

[0009] An incentive mechanism design module, which designs a multi-level incentive mechanism based on the user's response cost and grouping results;

[0010] An incentive mechanism optimization module, which optimizes incentive parameters of incentive mechanisms at all levels based on a reinforcement learning method;

[0011] A trust assurance module records user response data and the allocation results of the incentive mechanism based on blockchain technology.

[0012] Optionally, the user response cost evaluation module includes:

[0013] a travel convenience loss assessment unit, wherein the travel convenience loss assessment unit assesses the travel convenience loss based on the user behavior data;

[0014] A battery health loss assessment unit, wherein the battery health loss assessment unit assesses the battery health loss based on the Internet of Vehicles data;

[0015] a charging time value evaluation unit, wherein the charging time value evaluation unit evaluates the charging time value loss based on the user behavior data;

[0016] A weighted integration unit is configured to calculate a user's response cost based on the travel convenience loss, the battery health loss, and the charging time value loss.

[0017] Optionally, the user grouping module includes:

[0018] A feature space construction unit, wherein the feature space construction unit constructs static features, behavioral features, and response features based on the Internet of Vehicles data and the user behavior data;

[0019] A feature extraction unit, wherein the feature extraction unit extracts a feature vector from the static feature, the behavioral feature, and the response feature;

[0020] A clustering unit performs clustering processing based on the extracted feature vectors and determines the group category of each user according to the clustering processing result.

[0021] Optionally, the user grouping module further includes a dynamic feature unit, and the dynamic feature unit is used to dynamically update the user's behavioral features.

[0022] Optionally, the multi-level incentive mechanism includes a real-time monetary incentive mechanism based on cost compensation, a long-term incentive mechanism based on points, and a non-material incentive mechanism based on social reputation.

[0023] Optionally, the real-time monetary incentive mechanism based on cost compensation includes:

[0024] A marginal cost pricing unit, wherein the marginal cost pricing unit calculates the marginal cost of user participation based on the response cost;

[0025] A bid price response curve unit, which establishes price response functions for different user groups based on the user grouping results and predicts user participation rates under different incentive levels;

[0026] A price discrimination optimization unit is configured to design differentiated incentive prices for different user groups based on user grouping results.

[0027] Optionally, the incentive mechanism optimization module optimizes incentive parameters of incentive mechanisms at each level based on a reinforcement learning method, including:

[0028] The optimization problem of the incentive parameters of each incentive mechanism is transformed into a Markov decision process. The state space includes multi-dimensional information such as the current grid load status, the distribution of available electric vehicle resources, historical response status, and market prices. The action space is the combination of incentive parameters for each user group. The reward function is a weighted combination of demand response effect and incentive cost.

[0029] The deep Q-network algorithm is used to learn the optimal combination of excitation parameters.

[0030] Optionally, the trust assurance module includes:

[0031] A differential privacy unit is used to desensitize private data in Internet of Vehicles data and user behavior data.

[0032] Optionally, the trust assurance module includes:

[0033] A trust evaluation unit is configured to calculate a reputation value based on historical response completion rate, response punctuality, response volume accuracy, and abnormal behavior frequency in the user response data.

[0034] The present application also provides an incentive method for an incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles, the incentive method comprising:

[0035] Calculate the response cost of each user based on Internet of Vehicles data and user behavior data;

[0036] Divide users into different groups based on Internet of Vehicles data and user behavior data;

[0037] Design a multi-level incentive mechanism based on the user's response cost and grouping results;

[0038] Optimize the incentive parameters of each level of incentive mechanism based on reinforcement learning method;

[0039] Based on blockchain technology, user response data and incentive mechanism allocation results are recorded.

[0040] The present application provides a system and method for encouraging electric vehicle users to participate in demand response based on the Internet of Vehicles, which has the following technical effects:

[0041] The user response cost assessment module accurately quantifies explicit and implicit costs, providing a basis for reasonable incentive design. By grouping users for differentiated incentives, efficient resource allocation is achieved, and a multi-level incentive system for long-term participation is designed to increase users' enthusiasm for continuous participation. An adaptive incentive optimization algorithm based on reinforcement learning is used to achieve dynamic adjustment of incentive strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of an incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to one or more embodiments.

[0043] Figure 2 The present invention is a flowchart of the main steps of a method for motivating electric vehicle users to participate in demand response based on the Internet of Vehicles according to one or more embodiments. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0045] Before describing the various embodiments of the present application in detail, the technical concept of the present application will be briefly described first: the current incentive mechanism has problems such as low incentive accuracy, difficulty in dynamic changes, and weak long-term incentives. To this end, the present application provides an incentive system and method for electric vehicle users to participate in demand response based on the Internet of Vehicles. First, the user response cost evaluation module is used to accurately quantify explicit and implicit costs, providing a basis for reasonable incentive design. By grouping users for differentiated incentives, efficient resource allocation is achieved, a multi-level incentive system for long-term participation is designed, and the enthusiasm of users for continuous participation is improved. Finally, an adaptive incentive optimization algorithm based on reinforcement learning is designed to achieve dynamic adjustment of incentive strategies. The specific principles of the incentive system and method for electric vehicle users to participate in demand response based on the Internet of Vehicles of the present application will be described below in conjunction with more embodiments.

[0046] Specifically, if Figure 1 As shown, the incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles (IoV) in this first embodiment includes a user response cost assessment module 10, a user grouping module 20, an incentive mechanism design module 30, an incentive mechanism optimization module 40, and a trust assurance module 50. The user response cost assessment module 10 calculates the response cost of each user based on IoV data and user behavior data. The user grouping module 20 divides each user into different groups based on IoV data and user behavior data. The incentive mechanism design module 30 designs a multi-level incentive mechanism based on the user response cost and grouping results. The incentive mechanism optimization module 40 optimizes the incentive parameters of each level of the incentive mechanism based on the reinforcement learning method. The record supervision module 50 records user response data and the allocation results of the incentive mechanism based on blockchain technology.

[0047] Exemplarily, Internet of Vehicles data, user behavior data, and power grid operation data are collected through the Internet of Vehicles system. Internet of Vehicles data includes data such as charging status, battery health, location information, driving records, vehicle static information, and inferred / aggregated information. The charging status includes whether it is charging, real-time charging power, and the current battery power percentage. Battery health is an indicator that reflects the degree of battery aging or capacity decay. Location information includes the vehicle's real-time geographic location (such as latitude and longitude) and parking location information (used to determine whether it is at home, work, or a public charging station). Driving records include the start and end time of the trip, the start and end locations, mileage, average speed, etc. Vehicle static information includes vehicle model information, rated battery capacity, charging power limit, and vehicle age. Inferred / aggregated information includes the distribution of available electric vehicle resources, which is aggregated based on the real-time location and charging status information of a group of vehicles and is used to evaluate regional response potential.

[0048] Exemplarily, user behavior data includes charging preferences, travel habits, charging behavior characteristics, demand response interaction history, and user feedback. Charging preferences include the user's set target charging SOC, preferred charging start / end time period, preferred charging location type (home, work, specific public charging station, etc.), charging speed requirements or preferences, travel habits including average daily mileage, common travel routes or destinations, travel patterns (e.g., regular commuting time / routes, random travel), and future travel needs (time, destination, required power) predicted based on historical data. Charging behavior characteristics include historical charging frequency, historical charging time distribution, historical charging location preferences, and charging regularity index (quantifying the predictability of charging behavior). Demand response interaction history includes historical records of participation in demand response events, historical response rate (proportion of accepting invitations), response delay time (time from receiving an invitation to confirming or starting a response), response completion quality, response duration, historical response completion rate, response punctuality, response volume accuracy, frequency of abnormal behavior, and user reputation value. User feedback includes evaluations, scores, or text suggestions submitted by users through applications or other channels regarding their demand response experience.

[0049] For example, grid operation data includes load conditions, electricity price information, and demand response event data. Grid load conditions include the current real-time load of a region or system, historical load data, and future short-term / medium-term load forecasts. Electricity price information includes the basic electricity price structure, real-time electricity price signals, electricity market prices, and future electricity price forecasts. Demand response event information includes the unique identifier of the demand response event, the cause or type of the event trigger, the target area or user range of the event, the notification time of the event, the planned start time, the planned end time, the specific response requirements, and the urgency or priority of the event.

[0050] In one or more embodiments, the user response cost evaluation module 10 breaks through the limitation of insufficient understanding of the real cost of users in traditional incentive mechanisms and establishes a comprehensive and accurate cost quantification framework. Specifically, the user response cost evaluation module 10 includes a travel convenience loss evaluation unit 101, a battery health loss evaluation unit 102, a charging time value evaluation unit 103 and a weighted integration unit 104. The travel convenience loss evaluation unit 101 evaluates the travel convenience loss based on the user behavior data, the battery health loss evaluation unit 102 evaluates the battery health loss based on the Internet of Vehicles data, the charging time value evaluation unit 103 evaluates the charging time value loss based on the user behavior data, and the weighted integration unit 104 calculates the user's response cost based on the travel convenience loss, battery health loss and charging time value loss.

[0051] Exemplarily, the travel convenience loss assessment unit 101 establishes a user travel pattern and charging habit analysis model, predicts the user's future travel needs through the Markov chain method, and quantifies the convenience loss that charging adjustments may cause to future travel. In other words, the risk of insufficient power for future travel caused by the response behavior is calculated, and combined with the urgency of travel time and the importance of the trip, it is converted into economic loss. For example, the state can be defined as a combination of time period, location area and battery status. The time period can be divided into each hour, the location area can be defined as "home", "workplace", and "common public charging points", and the battery status can be divided into "high power" (>80%), "medium power" (30%-80%), and "low power" (<30%). The model estimates the transition probability P(st+1|st) between states based on historical Internet of Vehicles data (time, location, SOC). Loss quantification: The model predicts the probability P(Insufficient SOC for Trip) of being in a "low power" state and located in a "location area" where travel is required (such as "home" and in the commuting time period) within the next N hours. The convenience loss can be estimated as:

[0052] C convenience = P(Insufficient SOC for Trip) × (Alternative Travel Cost + λ × Estimated Delay Time)

[0053] In the formula, the alternative travel cost (such as taxi fare) and the time value coefficient λ can be inferred based on user profiles or historical behavior.

[0054] The battery health loss assessment unit 102 analyzes the accelerated effect of the additional charge and discharge cycles caused by participating in demand response on the battery capacity decay based on the lithium-ion battery cycle life model. The model takes into account key factors such as charge and discharge depth, rate, temperature, etc., accurately calculates the battery life loss that may be caused by each response behavior, and converts it into economic cost. Among them, the model type can adopt a semi-empirical model that combines the number of cycles and calendar life. For example, the cycle life loss is based on the equivalent full cycle number (which can be calculated from the charge and discharge depth DoD of the response process through the rain flow counting method) and the battery characteristic curve (describing the single cycle loss under different DoD, charge and discharge rate C-rate, and temperature T); the calendar life loss is based on the Arrhenius equation, taking into account the influence of storage SOC and temperature. The integration method is to calculate the equivalent cycle loss Q caused by the additional charge and discharge process caused by demand response according to its specific DoD, C-rate, duration and estimated battery temperature T. loss_cycle The economic cost translates into:

[0055]

[0056] Among them, Q EOL It is the total equivalent cycle capacity throughput when the battery reaches the end-of-life condition (such as capacity decay of 20%).

[0057] The charging time value evaluation unit 103 uses the user's historical charging behavior data to establish an individualized time value function to quantify the subjective value loss of charging delays in different time periods. The function evaluates the inconvenience caused to the user by charging time adjustment by analyzing the user's charging period preference and time sensitivity. The function can use a binary Logit selection model. Based on the user's historical charging records, each charging decision is regarded as a selection event. The input features of the model include the charging start time (hour of the day), day of the week, current SOC, expected next travel time, electricity price at the time, whether it is a peak period, etc. The model outputs the probability P (ChargeNow) that the user chooses "Charge Now" under this condition. The time value function V (t) can be positively correlated with this probability, for example: V (t) = log (P (ChargeNow | features at time t))

[0058] Furthermore, if the user prefers preferred Charging, but due to the response demand is required at t actual Charging, the time value loss can be estimated as:

[0059] C time =max(0, V(t preferred )-V(t actual ))×μ

[0060] Here, μ is the coefficient that converts the difference in log probability into monetary value, which can be set through a small-scale user survey or reference to relevant economic research.

[0061] The weighted integration unit 104 combines the above-mentioned costs into a unified response cost evaluation model through a weighted integration method, which is expressed as:

[0062] C total (u i ,a)=w1·C conv (u i ,a)+w2·C bat (u i ,a)+w3·C time (u i , a)

[0063] Among them, C total (u i , a) represents user u i The total cost of executing response behavior a, C conv 、C bat and C time Representing the loss of travel convenience, battery health, and charging time value, respectively, w1, w2, and w3 are the corresponding weight coefficients. To ensure the accuracy and adaptability of the model, the system continuously optimizes the weight parameters based on feedback from users' actual response behavior, achieving adaptive adjustments by minimizing the difference between predicted costs and users' actual participation in decision-making. Through the comprehensive weighting of these three levels of evaluation, the module can calculate a personalized response cost for each user in each demand response event, providing a scientific basis for subsequent incentive pricing. The innovation of this module lies in its comprehensive cost quantification, including implicit costs, which enables the incentive mechanism to be established on a more reasonable economic basis.

[0064] In one or more embodiments, the clustering-based user grouping module 20 provides decision support for differentiated incentive strategies by refining user classification. This module breaks through the limitation of traditional methods that regard all users as homogeneous groups, and realizes accurate user profiling and classification management. Exemplarily, the user grouping module 20 includes a feature space construction unit 201, a feature extraction unit 202, and a clustering unit 203. The feature space construction unit 201 constructs static features, behavioral features, and response features based on Internet of Vehicles data and user behavior data; the feature extraction unit 202 extracts feature vectors from static features, behavioral features, and response features; the clustering unit 203 performs clustering processing based on the extracted feature vectors, and determines the group category of each user according to the clustering processing results.

[0065] Specifically, static features include fixed attributes such as vehicle model information, battery capacity, charging power limit, and vehicle age. Behavioral features are extracted by analyzing historical user behavior data and include average daily mileage, charging frequency, charging time distribution, charging location preference, and charging regularity index. Response features are extracted based on a user's historical performance in demand response, including historical response rate, response delay, response completion quality, and response duration.

[0066] Furthermore, when feature extraction unit 202 performs feature extraction, the module uses feature engineering techniques to process the original features, including standardization, dimensionality reduction, and feature selection. To address the issue of inter-feature correlation, principal component analysis (PCA) is applied to extract the main feature vectors. To address the issue of uneven feature importance, a feature importance assessment method is introduced to assign different weights to features.

[0067] Finally, clustering unit 203 uses the processed feature data in combination with an improved K-means++ clustering algorithm and a hierarchical clustering method to perform user grouping. During the clustering process, the silhouette coefficient and gap statistics methods are used to determine the optimal number of clusters. In addition, a Gaussian mixture model (GMM) is introduced as a supplement to improve the stability and interpretability of the clustering results. In other embodiments, a fuzzy clustering method is used to allow users to belong to multiple groups simultaneously and to assign different incentive strategies based on membership, thereby enhancing the flexibility of grouping.

[0068] Furthermore, the user clustering module 20 also includes a dynamic feature unit 204, which is used to dynamically update user behavior features. To address the time-varying nature of user behavior, the dynamic feature unit 204 incorporates a dynamic feature update mechanism. This mechanism regularly updates user behavior features using a sliding window approach, with the window width adaptively adjusted based on feature stability. Clustering results are adjusted using incremental learning techniques, reflecting changes in user behavior while maintaining clustering stability.

[0069] Based on the clustering results, the system divides users into different groups. Typical groups include:

[0070] (1) Highly flexible: Charging time and location are highly flexible, suitable for deep participation in demand response.

[0071] (2) Strong planning: Charging behavior is regular but predictable, suitable for advance appointment response.

[0072] (3) Economically sensitive: highly sensitive to economic incentives and suitable for price-guided responses.

[0073] (4) Convenience-first type: Focus on convenience and suitable for low-intrusion response methods.

[0074] (5) Low participation tendency: low willingness to participate, requiring higher incentives or simplified participation processes.

[0075] This refined grouping provides the basis for the subsequent design of differentiated incentive strategies, enabling the system to customize the most suitable incentive plans for different user groups.

[0076] In one or more embodiments, the multi-tiered incentive mechanism includes real-time monetary incentives based on cost compensation, long-term incentives based on points, and non-material incentives based on social reputation. This three-tiered, complementary incentive system meets users' diverse motivational needs and balances short-term engagement motivation with long-term engagement habit formation.

[0077] For example, a real-time monetary incentive mechanism based on cost compensation directly targets the economic cost of user participation in demand response and designs dynamically differentiated price incentives. This real-time monetary incentive mechanism based on cost compensation includes a marginal cost pricing unit, a bid price response curve unit, and a price discrimination optimization unit. The marginal cost pricing unit calculates the marginal cost of user participation based on the response cost, using the following formula:

[0078] M(u i , e) = α·C total (u i , a e )·(1+β·U(e))

[0079] Among them, M(u i , e) represents user u i The monetary incentive for participating in the response to event e, C total is the total cost of user response, U(e) is the system urgency function, and α and β are adjustment coefficients.

[0080] Furthermore, the bid price response curve unit establishes price response functions for different user groups based on the user grouping results, predicts user participation rates at different incentive levels, and optimizes incentive resource allocation. The price discrimination optimization unit also designs differentiated incentive prices for different user groups based on the user grouping results, minimizing total incentive costs while meeting system response requirements.

[0081] For example, a long-term incentive mechanism based on points can cultivate users' long-term participation habits through a points accumulation and redemption system. The specific implementation includes:

[0082] (1) Multi-factor point calculation: The points earned by a user each time they successfully participate in demand response are calculated based on the response difficulty, response quality, and system contribution. Response difficulty considers the notification lead time and response duration; response quality considers response completion and response timeliness; and system contribution considers load reduction and the value of the grid during the response period.

[0083] (2) Diversified points redemption: Points can be used to redeem charging coupons, parking discounts, value-added services and physical rewards, etc., forming a complete incentive closed loop. The redemption ratio is dynamically adjusted according to the product type and market value.

[0084] (3) Points level system: set point level thresholds so that users of different levels can enjoy differentiated services and priority, which encourages users to continuously accumulate points.

[0085] For example, the non-material incentive mechanism based on social reputation satisfies users’ needs for social recognition and self-realization through the following means:

[0086] (1) Environmental protection contribution list: Based on the cumulative contribution of users participating in demand response, the carbon emission reduction and energy saving are calculated to form a regional or national ranking list.

[0087] (2) Grid-friendly user certification: A multi-level grid-friendly user certification system is established, and certified users are given special identification and priority in the use of charging facilities and green travel activities.

[0088] These three layers of incentive mechanisms complement each other, forming a three-dimensional incentive framework. The system dynamically adjusts the weight of each layer based on user segmentation and individual preferences to maximize incentive effectiveness. In terms of multi-tiered incentive mechanism design, team incentives can also be introduced to encourage users to form response teams and participate together. Team members share incentives and monitor each other, improving overall response effectiveness. A prepayment mechanism can also be introduced, where users receive incentives in advance but are required to achieve the agreed response target to retain them. Otherwise, they must return part or all of the money, thus strengthening the binding force through pre-emptive incentives.

[0089] In one or more embodiments, the incentive mechanism optimization module 40 adaptively optimizes the incentive parameters of each level of incentive mechanism based on the reinforcement learning method to balance the system cost and response effect. The module formalizes the incentive optimization problem as a sequential decision-making problem and uses the reinforcement learning algorithm to continuously optimize the incentive strategy in practice.

[0090] First, the incentive mechanism optimization module 40 models the incentive mechanism optimization problem as a Markov decision process (MDP). The state space includes multi-dimensional information such as grid load conditions, available electric vehicle resource distribution, historical response conditions, and market prices. For example, the state vector S t The specific design can be:

[0091] [L t+1 ,...,L t+N ,N c1 ,...,N cK ,R c1 ,...,R cK ,P t , P t+1 ,...,H t , D t ]

[0092] Where L is the normalized forecast load sequence, N ck is the number of available vehicles for the kth user group, R ck is the recent average response success rate of the kth user group, P is the normalized market electricity price sequence, H t is the hour of the day, D t is the day of the week, and the entire state vector is flattened and normalized.

[0093] The action space is the combination of incentive parameters for each user group. If DQN is used, the actions need to be discretized. For each user group k, M real-time monetary incentive levels (such as 1.0 times, 1.2 times, 1.5 times, etc. of the base cost) and N points reward levels (such as 1.0 times, 1.2 times, 1.5 times, etc. of the base points) can be defined. The total action space size is (M×N) K If K is large, resulting in an overly large action space, consider: (1) optimizing the parameters independently for each group (decomposition); (2) using a DQN variant that can handle large action spaces (such as Wolpertinger Dueling Q-networks); (3) if the parameters are continuous, explicitly use an Actor-Critic algorithm such as DDPG (Deep Deterministic Policy Gradient) or SAC (Soft Actor-Critic), in which the Actor network directly outputs continuous incentive parameter values.

[0094] The reward function is designed as a weighted combination of demand response effect and incentive cost, expressed as:

[0095]

[0096] Among them, Q(s,a) is the response amount obtained by taking action a in state s, Q(s,a) is the corresponding incentive cost, C(s,a) is the target response amount, and C budget is the upper budget limit, λ1 and λ2 are weight coefficients.

[0097] Based on the MDP framework, the module uses the Deep Q-Network (DQN) algorithm to learn the optimal incentive strategy. The DQN network consists of an input layer, multiple hidden layers, and an output layer. It takes in state information and outputs value estimates for different actions. To improve learning stability, the incentive mechanism optimization module 40 uses the following techniques:

[0098] (1) Dual network structure: Maintain two neural networks with the same structure, the evaluation network and the target network. The evaluation network is used for action selection and parameter update, and the target network is used to calculate the target Q value to reduce the over-estimation problem.

[0099] (2) Prioritized experience replay: The sampling probability is determined based on the temporal difference error of the sample, and samples with high error are learned first to improve learning efficiency.

[0100] (3) Dueling DQN structure: The Q value is decomposed into a state value function and an action advantage function, respectively estimating the intrinsic value of the state and the relative advantage of each action, thereby improving learning accuracy.

[0101] To cope with the uncertainty and complexity of the actual environment, the module also introduces the following enhanced technologies:

[0102] (1) Thompson sampling: Maintains the probability distribution of each action value based on the Bayesian method, balances exploration and utilization, and adapts to environmental changes.

[0103] (2) Multi-scenario learning: Build a multi-scenario reinforcement learning framework based on scenario trees to train the model under different system loads, user responses, and market price scenarios to improve the robustness of the strategy.

[0104] (3) Adversarial sample training: Generate training samples under adverse conditions to enhance the performance of the model in extreme cases.

[0105] In addition, the incentive mechanism optimization module 40 establishes a user response behavior prediction model as an environmental simulator for reinforcement learning. The prediction model is based on historical data and uses a recurrent neural network (RNN) structure to capture time series characteristics to predict the response probability and response amount of various types of users under different incentive levels. The prediction model is regularly updated with newly collected data to maintain the accuracy of the environmental simulation.

[0106] Through the above-mentioned technical combination, the system can realize the continuous optimization of incentive parameters and adaptively adjust the incentive strategy according to grid demand, user response and market conditions, so as to achieve the goal of "getting the maximum response effect with the minimum incentive cost".

[0107] The above-mentioned optimized excitation parameters include the following categories:

[0108] (1) Monetary incentive parameters: basic incentive price, user group differentiation coefficient, system urgency influencing factor, time period differentiation price coefficient (price multiplier in different time periods), response volume step price parameter (rate of change of incentive price with increase in response volume).

[0109] (2) Points incentive parameters: basic points assignment, response difficulty coefficient, response quality weight coefficient (weight of each quality indicator), historical behavior adjustment coefficient, points decay rate (points depreciation rate over time), and points redemption ratio (points requirements for different redemption products).

[0110] (3) Social reputation incentive parameters: weight coefficients in the environmental contribution scoring algorithm, medal awarding thresholds, social reputation attenuation function parameters, and various achievement unlocking conditions.

[0111] (4) Comprehensive optimization parameters: weight distribution of the three-tier incentive mechanism, weight coefficient in the reward function, and incentive strategy preference coefficient for different user groups.

[0112] In actual operation, the reinforcement learning incentive optimization module continuously monitors the system status and user feedback, and continuously adjusts the values ​​of these incentive parameters so that the incentive mechanism can maintain the best effect in different situations. For example, the weight of monetary incentives may be increased during peak load periods of the power grid, and when the user response rate is already high, the parameters may be adjusted to prioritize cost control. Through this dynamic optimization, the system can adapt to changes in power grid demand, evolution of user behavior, and fluctuations in market conditions, and maintain the long-term effectiveness and economy of the incentive mechanism. In other implementations, it is possible to consider adopting a multi-agent reinforcement learning framework, treating each user group as an independent agent, and achieving overall optimization through collaborative learning. Meta-learning technology can also be introduced to enable the system to quickly adapt to new user groups or new market environments, accelerating the learning process.

[0113] In one or more embodiments, the trust assurance module provides a transparent and fair technical foundation for the entire incentive mechanism based on blockchain, solves the trust problem of traditional centralized systems, and ensures the sovereignty of user data and the fairness of incentive execution. The trust assurance module adopts a consortium chain architecture and is jointly maintained by multiple parties such as power grid operators, charging facility operators, electric vehicle manufacturers and user representatives. Compared with public chains, consortium chains have higher transaction processing capabilities and lower energy consumption, and have better decentralization and credibility than private chains. They are suitable for the data scale and trust requirements of demand response scenarios. The key information recorded on the blockchain includes: demand response events (event ID, start time, end time, response requirements, etc.), user participation behavior (user ID, response time, response volume, etc.) and incentive distribution (incentive amount, points reward, medal awarding, etc.). Through distributed ledger technology, once this information is recorded, it cannot be tampered with, ensuring the credibility and traceability of the data.

[0114] Smart contracts are the core component of this module, automatically executing demand response rules. The main smart contract functions include:

[0115] (1) Response event management: Automatically publish demand response events, manage event lifecycles, and verify event legitimacy.

[0116] (2) Response verification: Verify whether the user response behavior meets the requirements according to preset rules, including response time verification, response volume verification and behavior consistency verification.

[0117] (3) Incentive calculation and distribution: Based on the response verification results, the user's deserved incentives are automatically calculated according to the preset formula, and the incentive distribution process is executed.

[0118] (4) Dispute Resolution: Handle user objections to response verification or incentive calculation and provide a transparent arbitration mechanism.

[0119] Furthermore, to protect user privacy, the trust assurance module 50 also includes a differential privacy unit, which is used to desensitize the privacy data in the Internet of Vehicles data and user behavior data. Before the user data is uploaded to the chain, the differential privacy unit adds random noise that meets the differential privacy requirements to sensitive information (such as specific charging location, driving trajectory, etc.), ensuring that the value of data analysis is retained while preventing individual information leakage. Specifically, for the query function f and the data set D, the returned result is:

[0120]

[0121] Where Δf is the sensitivity of the query function, ∈ is the privacy budget parameter, and Noise is random noise that satisfies a specific distribution (such as Laplace distribution or Gaussian distribution). Depending on the data sensitivity and application requirements, ∈ can be set between 0.5 and 5. For example, ∈ used for internal statistical analysis can be slightly larger, while ∈ used for external reporting needs to be smaller. The allocation of ∈ can be managed based on query type and frequency. For queries that count the number of users in a certain area, Δf = 1. For calculating the average charging time, Δf is the maximum possible charging time for a single user divided by the total number of users.

[0122] Furthermore, the trust assurance module 50 also includes a trust evaluation unit, which calculates a reputation value based on the historical response completion rate, response punctuality, response volume accuracy, and abnormal behavior frequency in the user's response data. The user's reputation value affects their priority in obtaining high-value response tasks and their eligibility for preferential incentives, forming a positive incentive cycle. The calculation formula is as follows:

[0123]

[0124] Where η is the update rate (such as 0.05), ω i is the weight of each indicator (such as ω1 = 0.5, ω2 = 0.2, ω3 = 0.2, ω4 = 0.1). Each indicator is calculated based on the most recent T response events. AnomalyPenalty is a penalty item calculated based on the number and severity of abnormal behaviors. All score items are normalized to the interval [0, 1].

[0125] By combining blockchain technology with mechanisms such as differential privacy and reputation assessment, this module achieves a balance between protecting user privacy and ensuring system transparency and credibility, providing a safe and reliable technical environment for users to participate in demand response.

[0126] In other implementations, zero-knowledge proof technology can be considered to achieve a higher level of privacy protection and verify response behavior without leaking any user data. Decentralized identity (DID) technology can also be introduced, allowing users to participate in demand response by self-managing their identity credentials while maintaining data sovereignty.

[0127] The above five core functional modules work together under a unified system architecture to form a data-driven closed-loop optimization system. The information flow between modules is as follows:

[0128] (1) The user response cost evaluation module provides user cost feature data for the clustering-based user segmentation module and provides a cost compensation benchmark for the multi-level incentive mechanism design module.

[0129] (2) The clustering-based user segmentation module provides user group division for the multi-level incentive mechanism design module and provides state space definition for the incentive optimization module based on reinforcement learning.

[0130] (3) The multi-level incentive mechanism design module provides action space definition for the incentive optimization module based on reinforcement learning and incentive rules for the trust assurance module based on blockchain.

[0131] (4) The incentive mechanism optimization module based on reinforcement learning outputs the optimized incentive parameters, which are executed by the multi-level incentive mechanism design module, and the results are recorded by the trust assurance module based on blockchain.

[0132] (5) The user response data and incentive allocation results recorded by the blockchain-based trust assurance module are fed back to other modules for cost model updates, user grouping adjustments, and reinforcement learning training.

[0133] During actual operation, the system forms a complete closed loop of "data collection - analysis and processing - strategy optimization - execution supervision - and effect feedback." As the system operates, user data continues to accumulate, and the models and parameters of each module are updated and optimized, continuously improving system performance.

[0134] In one or more embodiments, an incentive method of an incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles includes:

[0135] Step S10: Calculate the response cost of each user based on the Internet of Vehicles data and user behavior data;

[0136] Step S20: Divide each user into different groups based on the Internet of Vehicles data and user behavior data;

[0137] Step S30: designing a multi-level incentive mechanism based on the user's response cost and grouping results;

[0138] Step S40: Optimizing the incentive parameters of each level of incentive mechanism based on a reinforcement learning method;

[0139] Step S50: Record user response data and incentive mechanism allocation results based on blockchain technology.

[0140] For the detailed steps of each of the above steps, please refer to the relevant description of the previous embodiment and will not be repeated here.

[0141] Compared with the closest prior art, the present invention has the following significant advantages:

[0142] In terms of user cost assessment: Existing solutions primarily consider the direct economic costs of electricity price differences, ignoring hidden costs such as the impact of charging delays on user convenience and additional battery loss. In contrast, this embodiment establishes a comprehensive cost assessment model that incorporates travel convenience losses, battery cycle loss, and the value of charging time. This more accurately quantifies the user's true response costs, making incentive pricing more reasonable and fair, and improving user acceptance and satisfaction.

[0143] Regarding incentive strategy design: Existing solutions employ a simple categorized subsidy mechanism, lacking detailed consideration of individual user differences. This embodiment, however, leverages multi-dimensional feature analysis and machine learning techniques to achieve precise user segmentation and personalized incentive design, enabling more efficient allocation of incentive resources. Experimental verification shows that compared to a unified incentive strategy, this embodiment's differentiated incentive strategy can increase response rates by approximately 25-30% under the same incentive budget, significantly improving incentive efficiency.

[0144] Regarding incentive mechanisms, existing solutions primarily focus on short-term economic incentives, neglecting the cultivation of long-term user engagement. This embodiment, however, builds a multi-tiered incentive system that integrates economic incentives, point rewards, and social reputation. This system not only satisfies users' short-term economic interests but also fosters their inherent motivation for long-term engagement. Experiments have shown that the introduction of multi-tiered incentives increases user sustained engagement rates by approximately 20%, significantly enhancing the long-term effectiveness of the incentive mechanism.

[0145] In terms of incentive optimization: This embodiment utilizes more advanced technical means. Existing crowdsourcing mechanisms rely on active user submissions, are susceptible to strategic behavior, and have limited adaptability to market changes. This embodiment, however, constructs an adaptive optimization framework based on deep reinforcement learning, capable of continuously optimizing incentive strategies based on dynamic changes in system requirements and user response behavior. Experimental results show that compared to a fixed incentive strategy, this embodiment's adaptive optimization method can reduce total incentive costs by approximately 15-20% while maintaining the same level of response effect, significantly improving optimization efficiency.

[0146] Regarding the trust mechanism: This embodiment provides more reliable technical support. Existing technologies generally overlook issues of user privacy protection and incentive fairness, increasing user concerns about participation. However, this embodiment, based on blockchain technology and differential privacy, establishes a trust mechanism that balances transparency, fairness, and privacy protection, effectively enhancing user confidence in participation. User survey data shows that the introduction of this mechanism has increased user trust by approximately 35%, significantly increasing participation willingness.

[0147] In terms of system adaptability: Existing technologies are mostly static solutions targeting specific scenarios, making them difficult to adapt to application environments of varying scales and characteristics. However, this embodiment, through modular design and an adaptive learning mechanism, enables flexible configuration and optimization based on the characteristics of different application scenarios. Its applicability ranges from small-scale communities (with dozens of electric vehicles) to large-scale urban clusters (with tens of thousands of electric vehicles), proving to be a promising candidate for broader application.

[0148] The above describes in detail the specific implementation methods of the present application. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments can be modified and improved without departing from the principles and spirit of the present application, the scope of which is defined by the claims and their equivalents. These modifications and improvements should also be within the scope of protection of the present application.

Claims

1. An incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles, characterized by: The incentive system includes: A user response cost evaluation module, which calculates the response cost of each user based on Internet of Vehicles data and user behavior data; A user grouping module, which divides each user into different groups based on Internet of Vehicles data and user behavior data; An incentive mechanism design module, which designs a multi-level incentive mechanism based on the user's response cost and grouping results; An incentive mechanism optimization module, which optimizes incentive parameters of incentive mechanisms at all levels based on a reinforcement learning method; A trust assurance module records user response data and the allocation results of the incentive mechanism based on blockchain technology.

2. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 1 is characterized in that: The user response cost evaluation module includes: a travel convenience loss assessment unit, wherein the travel convenience loss assessment unit assesses the travel convenience loss based on the user behavior data; A battery health loss assessment unit, wherein the battery health loss assessment unit assesses the battery health loss based on the Internet of Vehicles data; a charging time value evaluation unit, wherein the charging time value evaluation unit evaluates the charging time value loss based on the user behavior data; A weighted integration unit is configured to calculate a user's response cost based on the travel convenience loss, the battery health loss, and the charging time value loss.

3. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 1 is characterized in that: The user grouping module includes: A feature space construction unit, wherein the feature space construction unit constructs static features, behavioral features, and response features based on the Internet of Vehicles data and the user behavior data; A feature extraction unit, wherein the feature extraction unit extracts a feature vector from the static feature, the behavioral feature, and the response feature; A clustering unit performs clustering processing based on the extracted feature vectors and determines the group category of each user according to the clustering processing result.

4. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 3 is characterized in that: The user grouping module further includes a dynamic feature unit, which is used to dynamically update the user's behavioral features.

5. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 1 is characterized in that: The multi-level incentive mechanism includes a real-time monetary incentive mechanism based on cost compensation, a long-term incentive mechanism based on points, and a non-material incentive mechanism based on social reputation.

6. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 5 is characterized in that: The real-time monetary incentive mechanism based on cost compensation includes: A marginal cost pricing unit, wherein the marginal cost pricing unit calculates the marginal cost of user participation based on the response cost; A bid price response curve unit, which establishes price response functions for different user groups based on the user grouping results and predicts user participation rates under different incentive levels; A price discrimination optimization unit is configured to design differentiated incentive prices for different user groups based on user grouping results.

7. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 5 is characterized in that: The incentive mechanism optimization module optimizes the incentive parameters of each level of incentive mechanism based on the reinforcement learning method, including: The optimization problem of the incentive parameters of each incentive mechanism is transformed into a Markov decision process. The state space includes multi-dimensional information such as the current grid load status, the distribution of available electric vehicle resources, historical response status, and market prices. The action space is the combination of incentive parameters for each user group. The reward function is a weighted combination of demand response effect and incentive cost. The deep Q-network algorithm is used to learn the optimal combination of excitation parameters.

8. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 1 is characterized in that: The trust assurance module includes: A differential privacy unit is used to desensitize private data in Internet of Vehicles data and user behavior data.

9. The incentive system for electric vehicle users to participate in demand response based on the Internet of Vehicles according to claim 1 is characterized in that: The trust assurance module includes: A trust evaluation unit is configured to calculate a reputation value based on historical response completion rate, response punctuality, response volume accuracy, and abnormal behavior frequency in the user response data.

10. An incentive method for an electric vehicle user participation in demand response incentive system based on the Internet of Vehicles according to any one of claims 1 to 9, characterized in that: The incentive method includes: Calculate the response cost of each user based on Internet of Vehicles data and user behavior data; Divide users into different groups based on Internet of Vehicles data and user behavior data; Design a multi-level incentive mechanism based on the user's response cost and grouping results; Optimize the incentive parameters of each level of incentive mechanism based on reinforcement learning method; Based on blockchain technology, user response data and incentive mechanism allocation results are recorded.

Images