Community-level energy utilization and multi-user privacy protection system

Abstract

The invention discloses an intelligent electric meter privacy protection and cost optimization system based on a multi-user shared battery. The system comprises a plurality of intelligent electric meters, a plurality of energy management units and a centralized energy storage device. Through a multi-agent reinforcement learning algorithm, the system can perform dynamic optimization on the basis of the real-time power consumption data and the battery state of the user so as to optimize energy utilization and protect the privacy of the user. The function of the privacy protection module is realized through a discriminator network, and the network guides the energy management unit to adjust the battery operation strategy by measuring the mutual information between the reading of the intelligent electric meter and the real load of the user, thereby reducing the exposure risk of the power utilization habit of the user. The system is suitable for a multi-user scene, and can efficiently manage energy resources and improve the user privacy protection level.

Description

Technical Field

[0001] The present invention relates to an energy management system in the field of smart grids, and in particular to a collaborative control system of community-level multi-user smart meters, energy management units and energy storage devices. Background Art

[0002] With the rapid development of smart grid technology in my country, more and more households and communities have begun to widely use smart meters and energy storage devices (such as household batteries) to optimize electricity efficiency and reduce electricity bills. Smart meters in existing technologies usually provide billing services by monitoring and recording users' electricity usage data, and to a certain extent help users adjust their electricity usage behavior according to electricity price fluctuations. However, most existing systems only rely on real-time data from a single meter, and fail to give full play to the synergy of multiple meters and energy storage devices in the community, and are thus unable to cope with complex electricity price changes.

[0003] For example, the Chinese patent document with application number 202410795815.7 discloses a smart meter control system that can realize the basic single-user electricity usage plan adjustment function, but due to the lack of advanced optimization algorithms, especially in a multi-user, multi-device community environment, it is difficult to achieve the same energy utilization effect. In addition, the existing system also has shortcomings in multi-user privacy protection, such as the inability to take into account users with different electricity usage habits, and there is a risk of privacy leakage.

[0004] For example, although some existing collaborative control methods have been applied to local control of smart grids, they usually fail to fully consider the overall optimization of the community and the special needs of user privacy protection. Existing technologies make it difficult to ensure the optimization of collaborative charging and discharging strategies between multiple users and multiple devices, and it is also difficult to achieve power management of energy utilization while ensuring privacy.

[0005] Therefore, a new smart grid energy management system is needed to coordinate the control of smart meters and energy storage equipment on a community basis, which can not only optimize energy utilization but also effectively protect users' electricity privacy. Summary of the invention

[0006] The existing smart meter and energy management system has the following technical problems in a community environment with multiple users and a single energy storage device: First, when dealing with complex changes in electricity prices, the existing system cannot fully utilize the synergy of multiple meters and one energy storage device in a community, which will inevitably lead to unsatisfactory optimization effects in saving electricity costs; second, there is a lack of effective privacy protection mechanism, which is easy to expose users' electricity behavior patterns and poses a risk of privacy leakage; third, the existing multi-agent collaborative control algorithm fails to fully consider the privacy needs of individual users, making it difficult to achieve efficient privacy protection while saving electricity costs. The present invention aims to deploy a multi-agent adversarial reinforcement learning algorithm in the community and energy management solution to achieve collaborative control of multiple smart meters and energy storage devices, effectively solving the above problems, thereby optimizing electricity costs and providing user privacy protection functions.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] A community-level energy utilization and multi-user privacy protection system includes: multiple energy management units, multiple smart meters, a centralized energy storage device, and a processing device that assists the energy management unit in completing privacy protection.

[0009] This system deploys a new multi-agent reinforcement learning algorithm to optimize the charging and discharging strategies of multiple energy management units in the community in a multi-user collaborative environment.

[0010] Wherein, the energy management unit corresponds to the smart meter and deploys the agent in the multi-agent reinforcement learning algorithm;

[0011] The smart meter is used to monitor the user's electricity load, electricity price and battery charging and discharging status in real time, and interact with the system;

[0012] The centralized energy storage device is a battery, and the intelligent agent protects user privacy and saves electricity costs by controlling its charging and discharging behavior.

[0013] The processing device that assists the energy management unit in completing privacy protection deploys a centralized critic network to evaluate the strategies of each energy management unit. The advantage of this is that it can improve the coordination between multiple energy management units through global information, thereby effectively improving the stability and efficiency of the entire system when dealing with complex multiple energy management units to collaborate in completing energy utilization and privacy protection.

[0014] Furthermore, each intelligent agent in the energy management unit deploys an independent actor network to process the user's meter data and formulate a localized energy storage device control strategy to control the meter reading.

[0015] Furthermore, the energy management unit also includes a discriminator network for measuring the mutual information between the smart meter reading and the user's actual load. From the perspective of algorithm structure, the discriminator network and the actor network correspond one to one and appear in pairs.

[0016] Furthermore, the system deploys a centralized critic network to receive meter data and charging and discharging decisions of all users, and optimize the overall battery scheduling strategy based on the global state. The network evaluates and provides feedback on the strategies of each energy management unit based on the global information at the community level, thereby achieving dual optimization of community-level energy utilization management and user-level privacy protection at the community level.

[0017] Furthermore, the multi-agent reinforcement learning algorithm process is as follows:

[0018] Step S1, environment initialization: Initialize the status of all smart meters, the initial battery power, power load and electricity price information (s t ). The actor network V and its parameters φ, the critic network π i and its parameter θ i And the discriminator network F i and its parameter ω i Initialized to random values.

[0019] Step S2, strategy execution:

[0020] At each time step t, the actor network calculates the current partial observations Select Action After interacting with the environment for T time slots, the joint reward sequence (r 0 ,r 1 ,…,r T )(Calculation method see formula (1)).

[0021] The environment is a community where multiple intelligent agents share batteries. The privacy of individual users is protected by controlling the battery charging and discharging behavior. During the interaction, feedback on electricity costs and privacy protection effects is received, and the process of continuously optimizing one's own strategy is carried out.

[0022] The observed values ​​include the actual power and battery energy storage percentage of the agent corresponding to the user at time t.

[0023] The action is the battery charging (discharging) power;

[0024] The centralized critic network evaluates the value of each action according to formula (2);

[0025]

[0026] Among them, the global state is s t , except for agent i, the actions of other agents are (N represents the number of actor networks), discount factor γ = 1, generalized advantage parameter λ A =0.97, is based on the current critic network V φ Calculate the TD-error, Represents the value function estimated by the current critic network, output by the critic network.

[0027] Step S3, privacy feedback calculation: The discriminator network F corresponding to the i-th actor network i According to formula (3), the corresponding actor network (π i ) smart meter readings under control With real load The mutual information between:

[0028]

[0029] Feed this information back to the reinforcement learning process As the reward function r p The value of is used to measure the privacy risk. where g t represents the current electricity price, g t The setting of can flexibly simulate the electricity price model. t =λr c +(1-λ)r p Helps optimize the parameters of the critic network and its corresponding actor network.

[0030] Step S4, loss function optimization: comprehensive actor network The action selection, the evaluation results of the critic network, and the feedback of the discriminator network together constitute the loss function of the overall system. The joint loss function of the actor network and the critic network is shown in formula (4):

[0031]

[0032] where ρ t represents the probability ratio of the new and old strategies under the current action; θ={θ i ; i = 1, 2, …, N}; ∈ is a clip factor that controls the update amplitude of the actor network.

[0033] The loss function of each discriminator network corresponding to the actor network is shown in formula (5):

[0034]

[0035] in Equivalent to the discriminator network F i; j(t) represents a function about the current time step t; is the sigmoid function. By using the gradient descent method, the parameters of the actor and critic networks are optimized so that the system can minimize the risk of privacy leakage while maximizing the cost optimization goal.

[0036] Beneficial Effects

[0037] By deploying multiple energy management units in the system and using modules that provide built-in privacy protection, multi-user energy management units can collaborate highly in complex community scenarios with shared energy storage devices, intelligently optimize battery charging and discharging strategies, and significantly reduce electricity costs at the community level;

[0038] At the same time, after adding the discriminator network, the system can effectively control the mutual information between the meter readings and the user's actual electricity load, thereby ensuring the optimization of electricity bills while achieving user-level privacy protection without the help of a trusted third party.

[0039] In addition, the design of this system allows the weight λ of the reward function to be flexibly adjusted in practical applications, which not only ensures the adaptability of the technical solution but also meets the actual usage needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a schematic diagram of a multi-agent algorithm framework according to an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of system deployment in an embodiment of the present invention;

[0042] Figure 3 This is a load comparison curve diagram of the system simulation deployment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0044] Figure 1 This is a schematic diagram of the improved multi-agent algorithm framework used in the system of the present invention, taking the number of users 2 as an example;

[0045] Figure 2 This is a schematic diagram of the deployment of the system of the present invention, taking 3 users as an example;

[0046] Figure 3 This is a load comparison curve diagram of the simulated deployment of the system of the present invention, taking the number of users 3 as an example.

[0047] In the figure, I(·) represents mutual information. represents the action taken by the actor network numbered n in time slot t, represents the local observation of the actor network numbered n in time slot t, X(·) represents the protected private data, represents the advantage function estimated by the central critic network, s t represents the global state of time slot t, r t represents the joint immediate reward for time slot t.

[0048] A community-level energy utilization and multi-user privacy protection system includes: multiple energy management units, multiple smart meters, a centralized energy storage device, and a processing device that assists the energy management unit in completing privacy protection.

[0049] This system deploys a new multi-agent reinforcement learning algorithm to optimize the charging and discharging strategies of multiple energy management units in the community in a multi-user collaborative environment.

[0050] Wherein, the energy management unit corresponds to the smart meter and deploys the agent in the multi-agent reinforcement learning algorithm;

[0051] The smart meter is used to monitor the user's electricity load, electricity price and battery charging and discharging status in real time, and interact with the system;

[0052] The centralized energy storage device is a battery, and the intelligent agent protects user privacy and saves electricity costs by controlling its charging and discharging behavior.

[0053] The processing device that assists the energy management unit in completing privacy protection deploys a centralized critic network to evaluate the strategies of each energy management unit. The advantage of this is that it can improve the coordination between multiple energy management units through global information, thereby effectively improving the stability and efficiency of the entire system when dealing with complex multiple energy management units to collaborate in completing energy utilization and privacy protection.

[0054] The multi-agent reinforcement learning algorithm used in the present invention is improved on the basis of the traditional multi-agent reinforcement learning algorithm. The improved algorithm can provide a system-level intrinsic privacy protection function in the scenario where multiple energy management units collaboratively control energy storage devices.

[0055] Furthermore, each intelligent agent in the energy management unit deploys an independent actor network to process the user's meter data and formulate a localized energy storage device control strategy to control the meter reading.

[0056] Furthermore, the energy management unit also includes a discriminator network for measuring the mutual information between the smart meter reading and the user's actual load. From the perspective of algorithm structure, the discriminator network and the actor network correspond one to one and appear in pairs.

[0057] Furthermore, in addition to the decentralized actor network and discriminator network, a centralized critic network is deployed in the system to receive meter data and charging and discharging decisions of all users, and optimize the overall scheduling strategy of the battery based on the global state. The network evaluates and provides feedback on the strategies of each energy management unit based on the global information at the community level, thereby achieving dual optimization of community-level energy utilization management and user-level privacy protection at the community level.

[0058] In the specific implementation, after receiving the current environmental state, each energy management unit first generates the next action to control the energy storage device through its local actor network. These actions are aggregated to jointly control the charging and discharging of energy storage devices. Subsequently, the actions made by these strategies are sent to the centralized critic network, which comprehensively evaluates the strategies of all energy management units and generates an advantage function. Then guide the actor network to update to a better strategy (π).

[0059] In terms of privacy protection, the discriminator network uses a multi-layer perceptron structure to reflect the mutual information between the smart meter reading and the user's actual load. The mutual information is estimated using a metric based on the KL (Kullback-Leibler) divergence. During the optimization process, the weights of the discriminator network are continuously updated through back propagation to accurately estimate the mutual information between the meter reading and the user's actual load, thereby feeding back the privacy risk of its actions to the actor network.

[0060] In order to achieve multi-objective optimization of community-level energy utilization and user-level privacy protection, this system designs a weighted and single-step reward function as shown in formula (1), which takes the energy utilization effect r c and privacy risks p Different weights λ and 1-λ are assigned respectively, as described in Example 1. This has the advantage that λ can be dynamically adjusted according to actual application requirements, so that the energy management proposed by the present invention can be deployed more flexibly.

[0061] r t =λr c +(1-λ)r p (1)

[0062] In this system, the battery storage unit realizes energy sharing among multiple users through intelligent scheduling. The critic network dynamically adjusts the battery charging and discharging strategy according to the global power demand and electricity price fluctuations to maximize energy utilization efficiency and minimize user electricity costs.

[0063] The actor network and critic network are trained based on an improved multi-agent reinforcement learning algorithm, wherein the critic network conducts a comprehensive analysis of the power data of multiple users to generate a globally optimal battery charging and discharging strategy, ensuring that the system protects user privacy while saving costs.

[0064] The discriminator network calculates the mutual information between the user's smart meter readings and their actual power load, reduces the correlation between meter data and the user's power usage habits, and optimizes the balance between user privacy protection and system scheduling.

[0065] The battery storage unit communicates with the smart meters of multiple users in real time. The system can automatically adjust the charging and discharging strategy according to each user's power load and electricity price fluctuations to ensure that electricity costs are minimized during peak hours and battery energy storage is fully utilized during off-peak hours.

[0066] [Example 1]

[0067] Each household smart meter is equipped with an independent actor network and a corresponding discriminator network, such as Figure 1 The network shown is responsible for outputting battery charging and discharging decisions based on the current environmental conditions.

[0068] Each actor network consists of an input layer, a hidden layer, and an output layer. The input layer receives information including electricity prices, current battery power, and household loads; the hidden layer processes the input information through a nonlinear activation function, and the output layer provides the final charging and discharging actions. By minimizing mutual information, the needs of privacy protection and energy utilization can be taken into account when optimizing the actor network.

[0069] The discriminator network is used to estimate the mutual information between the smart meter readings and the actual load. The paired actor network, with the assistance of the discriminator network, can effectively reduce the sensitive information contained in the meter readings by reducing the mutual information, thereby protecting user privacy.

[0070] The critic network plays an evaluation role in the algorithm, such as Figure 1 As shown, it receives global information including the decisions of all energy management units and the overall environmental status to calculate and feedback the quality index of each smart meter's action. Its core is a deep neural network based on global information, which inputs all meter status information, outputs the quality of each energy management unit's action, and guides the actor network to update according to this value.

[0071] The multi-agent reinforcement learning algorithm used in this embodiment is as follows:

[0072] Step S1, environment initialization: Initialize the status of all smart meters, the initial battery power, power load and electricity price information (s t ). The actor network V and its parameters φ, the critic network πi and its parameter θ i And the discriminator network F i and its parameter ω i Initialized to random values.

[0073] Step S2, strategy execution:

[0074] At each time step t, the actor network calculates the current partial observations Select Action After interacting with the environment for T time slots, the joint reward sequence (r 0 ,r 1 ,…,r T )(Calculation method see formula (1)).

[0075] The environment is a community where two intelligent agents share a battery. The privacy of a single user is protected by controlling the battery charging and discharging behavior. During the interaction, feedback on the power cost and privacy protection effect is received, and the process of continuously optimizing one's own strategy is carried out.

[0076] The observed values ​​include the actual power and battery energy storage percentage of the agent corresponding to the user at time t.

[0077] The action is the battery charging (discharging) power;

[0078] The centralized critic network evaluates the value of each action according to formula (2);

[0079]

[0080] Among them, the global state is s t , except for agent i, the actions of other agents are (N represents the number of actor networks, which is 2 in this example), discount factor γ = 1, generalized advantage parameter λ A =0.97, is based on the current critic network V φ Calculate the TD-error, Represents the value function estimated by the current critic network, output by the critic network.

[0081] Step S3, privacy feedback calculation: The discriminator network F corresponding to the i-th actor network i According to formula (3), the corresponding actor network (π i ) smart meter readings under control With real load The mutual information between:

[0082]

[0083] Feed this information back to the reinforcement learning process As the reward function r p The value of is used to measure the privacy risk. where g t represents the current electricity price, g t The setting of can flexibly simulate the electricity price model, an example is the tiered electricity price. t =λr c +(1-λ)r p Helps optimize the parameters of the critic network and its corresponding actor network.

[0084] Step S4, loss function optimization: comprehensive actor network The action selection, the evaluation results of the critic network, and the feedback of the discriminator network together constitute the loss function of the overall system. The joint loss function of the actor network and the critic network is shown in formula (4):

[0085]

[0086] Among them, ρ t represents the probability ratio of the new and old strategies under the current action; θ={θ i ; i = 1, 2, …, N}; ∈ is a clip factor that controls the update amplitude of the actor network.

[0087] The loss function of each discriminator network corresponding to the actor network is shown in formula (5):

[0088]

[0089] in Equivalent to the discriminator network F i ; j(t) represents a function about the current time step t; is the sigmoid function. By using the gradient descent method, the parameters of the actor and critic networks are optimized so that the system can minimize the risk of privacy leakage while maximizing the cost optimization goal.

[0090] In the whole system, each smart meter works together through multi-agent reinforcement learning to form an efficient distributed energy management system. After the introduction of the discriminator network, the system maintains high energy efficiency while increasing the user privacy protection function. By deploying a centralized critic network, the adaptability and robustness of the algorithm are increased, and the algorithm calculation amount is reduced by replacing the decentralized critic network with a centralized critic network.

[0091] [Example 2]

[0092] Based on Example 1, this example further provides a method for deploying the system of the present invention in a building scenario, which is suitable for scenarios where multiple users share the same energy storage device (community-level battery). The system implements intelligent scheduling of batteries and user privacy protection through a multi-agent reinforcement learning algorithm.

[0093] like Figure 2 As shown in the figure, each user (smart meter) in the building is connected to the central shared battery through a local smart meter. The user's daily electricity consumption behavior is monitored in real time by the smart meter and transmitted to the energy management unit in the system for decision-making. All users share a community-level battery, and the control strategies of the energy management units at each user level are aggregated at the battery to control the battery behavior, so as to achieve the dual goals of optimizing electricity costs and protecting privacy.

[0094] Specifically, each user's smart meter monitors key information such as the actual household electricity consumption and the current community-level battery status in real time, and transmits it to their respective energy management units for processing. Each energy management unit makes local charging or discharging recommendations based on the local information received. The management of community-level batteries is jointly responsible by the energy management units of all users; at the same time, the critic network receives the decision and global status information of all smart meters, evaluates the charging and discharging requests of each user, and comprehensively considers the electricity demand, electricity price fluctuations, privacy protection goals, etc. of all users in the building, and guides the optimization of the energy management unit from a global perspective. The energy management unit in the system ensures that the user's electricity demand must be met, while taking into account the complex changes in electricity prices, and intelligently charging the battery to maximize the overall economic benefits. In addition, the energy management unit also needs to ensure that the meter readings do not expose the user's real electricity usage habits too much. In this way, the system optimizes energy utilization while ensuring that user privacy will not be easily leaked. Over time, electricity prices and user needs may change. The energy management unit in the system can continuously learn and optimize the local battery charging and discharging strategy to ensure that it can meet the dynamic needs of users and effectively adapt to various electricity fluctuation scenarios.

[0095] Under this system deployment solution, users can achieve intelligent optimization and scheduling of electricity consumption in building scenarios. By sharing a battery, users can effectively share energy resources and reduce energy cost burdens from the community and individuals. The system is not only economical, but also provides users with a sense of security and privacy protection.

[0096] [Example 3]

[0097] Based on Example 1, this embodiment targets the need for user privacy protection by shaping the user load to prevent the outside world from inferring the user's actual electricity usage behavior through the meter readings, such as Figure 3 As shown:

[0098] Figure 3 The left side shows the user's actual electricity load curve, which represents the load situation when the user actually uses electricity.

[0099] like Figure 3 On the right, to protect user privacy, the system uses a discriminator network to process the real load, making the output load curve present irregular noise characteristics. The shaped load curve interferes with the outside world's analysis of the user's electricity usage habits.

[0100] During this process, the discriminator network compares the difference between the user's actual load and the shaped load curve to ensure that privacy protection is achieved by perturbing the actual load. The battery achieves synchronization between privacy protection and electricity cost optimization through charge and discharge scheduling.

[0101] Through this implementation method, the system not only effectively improves the privacy of users' electricity usage behavior, but also ensures the stability of system operation and the efficient use of power resources.

[0102] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. A community-level energy utilization and multi-user privacy protection system, characterized in that: include: Multiple energy management units, multiple smart meters, a centralized energy storage device, and a processing device that assists the energy management unit in protecting privacy; The system deploys a new multi-agent reinforcement learning algorithm to optimize the charging and discharging strategies of multiple energy management units in the community in a multi-user collaborative environment; Wherein, the energy management unit corresponds to the smart meter and deploys the agent in the multi-agent reinforcement learning algorithm; The smart meter is used to monitor the user's electricity load, electricity price and battery charging and discharging status in real time, and interact with the system; The centralized energy storage device is a battery, and the intelligent agent protects user privacy and saves electricity costs by controlling its charging and discharging behavior; The processing device that assists the energy management unit in completing the privacy protection function deploys a centralized critic network to evaluate the strategies of each energy management unit.

2. A community-level energy utilization and multi-user privacy protection system according to claim 1, characterized in that: The centralized critic network receives the meter data and charging and discharging decisions of all users, and optimizes the overall scheduling strategy of the battery based on the global state; the network evaluates and provides feedback on the strategies of each energy management unit based on the global information at the community level, thereby achieving dual optimization of community-level energy utilization management and user-level privacy protection at the community level.

3. A community-level energy utilization and multi-user privacy protection system according to claim 1, characterized in that: The intelligent agent in each of the energy management units deploys an independent actor network to process the user's meter data and formulate a localized energy storage device control strategy to control the meter reading.

4. A community-level energy utilization and multi-user privacy protection system according to claim 3, characterized in that: The energy management unit also includes a discriminator network for measuring the mutual information between the smart meter reading and the user's actual load. From the perspective of algorithm structure, the discriminator network and the actor network correspond one to one and appear in pairs.

5. A community-level energy utilization and multi-user privacy protection system according to claim 4, characterized in that: The multi-agent reinforcement learning algorithm process is as follows: Step S1, environment initialization: Initialize the status of all smart meters, the initial battery power, power load and electricity price information t ; Actor network V and its parameters φ, critic network π i and its parameter θ i And the discriminator network F i and its parameter w i Initialize to random values; Step S2, strategy execution: At each time step t, the actor network calculates the current partial observations Select Action After interacting with the environment for T time slots, we obtain the joint reward sequence (r0, r1, ..., r T ), the calculation method is shown in formula (1); r t =λr c +(1-λ)r p (1) The environment is a community where multiple intelligent agents share batteries, and the privacy of individual users is protected by controlling the battery charging and discharging behavior, and feedback on power cost and privacy protection effect is received during the interaction process, so as to continuously optimize the strategy. The observed values ​​include the actual power and battery energy storage percentage of the agent corresponding to the user at time t; The action is the battery charging / discharging power; The centralized critic network evaluates the value of each action according to formula (2); Among them, the global state is s t , except for agent i, the actions of other agents are N represents the number of actor networks, the discount factor γ = 1, and the generalized advantage parameter λ A =0.97, is based on the current critic network V φ Calculate the TD-error, represents the value function estimated by the current critic network, output by the critic network; Step S3, privacy feedback calculation: The discriminator network F corresponding to the i-th actor network i According to formula (3), the corresponding actor network (π i ) smart meter readings under control With real load The mutual information between: Feed this information back to the reinforcement learning process As the reward function r p The value of is used to measure the privacy risk. where g t represents the current electricity price, g t The setting of can flexibly simulate the electricity price model; t =λr c +(1-λ)r p Helps optimize the parameters of the critic network and its corresponding actor network; Step S4, loss function optimization: comprehensive actor network The action selection, the evaluation results of the critic network, and the feedback of the discriminator network together constitute the loss function of the overall system; the joint loss function of the actor network and the critic network is shown in formula (4): where ρ t represents the probability ratio of the new and old strategies under the current action; θ={θ i ; i = 1, 2, ..., N}; ∈ is a clip factor, which can control the update amplitude of the actor network; The loss function of each discriminator network corresponding to the actor network is shown in formula (5): in Equivalent to the discriminator network F i ; j(t) represents a function about the current time step t; is the sigmoid function; the parameters of the actor and critic networks are optimized through the gradient descent method, so that the system can minimize the risk of privacy leakage while maximizing the cost optimization goal.

6. A community-level energy utilization and multi-user privacy protection system according to claim 4, characterized in that: The battery storage unit in the system realizes energy sharing among multiple users through intelligent scheduling. The critic network dynamically adjusts the battery charging and discharging strategy according to the global power demand and electricity price fluctuations to maximize energy utilization efficiency and minimize user electricity costs.

7. A community-level energy utilization and multi-user privacy protection system according to claim 4, characterized in that: The actor network and critic network are trained based on an improved multi-agent reinforcement learning algorithm, wherein the critic network conducts a comprehensive analysis of the power data of multiple users to generate a globally optimal battery charging and discharging strategy, ensuring that the system protects user privacy while saving costs.

8. A community-level energy utilization and multi-user privacy protection system according to claim 4, characterized in that: The system introduces a privacy-preserving discriminator network, which calculates the mutual information between the user's smart meter readings and their actual power load, reduces the correlation between meter data and the user's power usage habits, and optimizes the balance between user privacy protection and system scheduling.

9. A community-level energy utilization and multi-user privacy protection system according to claim 4, characterized in that: The battery storage unit communicates with the smart meters of multiple users in real time. The system can automatically adjust the charging and discharging strategy according to each user's power load and electricity price fluctuations to ensure that electricity costs are minimized during peak hours and battery energy storage is fully utilized during off-peak hours.

Images