Design method and device of event-triggered incentive mechanism based on radial basis function neural network in power market

Abstract

The application discloses a kind of event trigger incentive mechanism design method and device based on radial basis function neural network in power market, the method includes: based on radial basis function neural network design not dependent on the incentive mechanism of private information of electric power user in power market;Innovatively, the design method of discretization incentive coefficient based on event trigger is proposed to save limited resources;Setting parameter guarantees that system state is always bounded and convergent;The effectiveness of the scheme is verified by simulation example.The method of the application can effectively regulate the power consumption behavior of electric power user without revealing the privacy of electric power user, and realize the optimal state regulation of power market.

Description

Technical Field

[0001] This invention relates to the field of optimization control of non-cooperative game systems, and specifically to a design method and apparatus for an event-triggered incentive mechanism based on radial basis function neural networks in the electricity market. Background Technology

[0002] With the rapid development of intelligent society, the coordination and optimization of multi-agent systems has become a core research topic in academia and engineering. In non-cooperative game systems, each participant acts to maximize their own interests. This self-interested decision-making model can optimize benefits from an individual perspective, but from a holistic perspective, it often leads to a decline in overall social benefits. The classic "tragedy of the commons" vividly illustrates this phenomenon: in the allocation of public resources lacking unified management and constraints, individuals, driven by self-interest, recklessly demand resources. Although this satisfies their own needs in the short term, it rapidly depletes public resources, ultimately harming the long-term interests of the entire group. This fully demonstrates the contradiction between individual and collective interests in non-cooperative games. In the electricity market environment, electricity users, as independent decision-makers, often adjust their electricity consumption behavior based on their own electricity cost function. This non-cooperative game behavior may lead to excessive consumption of power generation resources, grid congestion, and a series of other problems, resulting in a decline in overall social performance.

[0003] To improve the overall performance of non-cooperative game systems and align individual decisions with the socially optimal goal, various incentive mechanisms have been proposed. Some methods employ pricing strategies, such as setting resource usage prices, to guide electricity users to adjust their electricity consumption behavior, thus aligning individual actions towards the social optimum. Other methods rely on mandatory regulation, with managers directly intervening in the electricity users' decision-making process. However, these strategies all have significant limitations. They often depend on managers accurately grasping the cost or benefit functions of all electricity users. While this may be feasible in small-scale systems, in large-scale systems, due to the large number of participants and complex relationships, acquiring and processing such a massive amount of information is virtually impossible.

[0004] With increasing awareness of privacy protection, some studies have begun to explore designing incentive mechanisms to balance the needs of individual privacy protection and system optimization under conditions of limited information. While these methods alleviate the information acquisition problem to some extent, they typically require frequent adjustments to incentive coefficients to adapt to system changes. Frequent parameter adjustments not only incur high computational costs but also increase the burden on communication networks, leading to a surge in data transmission and even network congestion. Furthermore, some studies have proposed discretized incentive coefficient design schemes, attempting to reduce computational complexity and communication costs by adjusting incentive coefficients in stages. Unfortunately, the applicability of these schemes is mostly limited to discrete-time systems. In practical continuous-time system scenarios, due to the continuous and real-time nature of system state changes, these methods are difficult to meet the needs of the actual electricity market. Consequently, there is still a lack of truly effective research results on the optimization of incentive coefficients for continuous-time systems under discrete conditions in the electricity market. Summary of the Invention

[0005] To address the aforementioned challenges in designing incentive mechanisms in the electricity market, this invention proposes a method and apparatus for designing an event-triggered incentive mechanism based on a radial basis function neural network (RBN). By employing a RBN to estimate unknown dynamics, it eliminates the need to obtain the private revenue function information of electricity users, thus protecting the privacy of electricity users in the electricity market. Furthermore, compared to continuous incentive coefficients, the event-triggered incentive mechanism design method reduces the system's communication bandwidth requirements and minimizes the consumption of limited computing and communication resources.

[0006] The technical solution adopted to achieve the purpose of this invention is as follows: a method for designing an event-triggered incentive mechanism based on a radial basis function neural network in the electricity market, comprising the following steps:

[0007] Step 1: Establish a socially optimal state problem model for user electricity consumption behavior in the electricity market, parameterize the state of electricity users, i.e., their electricity consumption behavior, and give the target socially optimal state function.

[0008] Step 2: Construct the decision-making model of users in the electricity market as a gradient descent model, and construct the incentive coefficients and incentive functions;

[0009] Step 3: Design an incentive mechanism based on radial basis function neural network to construct a dynamic decision-making mechanism that influences users' electricity consumption behavior without obtaining information on the private revenue function of electricity users;

[0010] Step 4 proposes an event-triggered discretization excitation coefficient update mechanism, which reduces the consumption of limited communication resources by designing triggering event conditions;

[0011] Step 5: Construct a convergence analysis framework for the incentive mechanism. By adjusting parameters, ensure that the electricity consumption behavior of each user in the electricity market can eventually reach the social optimal equilibrium, and exclude Zeno behavior to ensure the feasibility of the incentive design.

[0012] Further, step 1 includes:

[0013] Step 1-1: Establish the model and define the electricity market. The set of electricity users is Electricity users' electricity consumption behavior is represented as ,in Indicates electricity user Energy consumption;

[0014] Steps 1-2: Establish the following incentive mechanism:

[0015] (1)

[0016] in, Representing electricity users The cost function, Managers act through distributed controllers Apply to electricity users Incentive measures, Representing electricity users The cost function after applying the incentive;

[0017] Steps 1-3 define the following social optimal conditions:

[0018] (2)

[0019] in, This represents the globally optimal electricity consumption state in the electricity market, meaning that the electricity consumption behavior of each electricity user minimizes the overall cost. This is information available to managers.

[0020] Further, step 2 includes:

[0021] Step 2-1, construct the excitation coefficients and excitation functions as follows:

[0022] (3)

[0023] in, Representing electricity users The incentive coefficient satisfies For piecewise constants, Representing electricity users The excitation function;

[0024] Step 2-2: Define electricity users in the electricity market Electricity consumption behavior The gradient response over time is as follows:

[0025] (4)

[0026] in, The negative gradient of the original cost function is represented by the gradient descent response dynamics, which describes the nature of individuals in the electricity market who focus only on their immediate own gains and ignore the overall interests.

[0027] Steps 2-3 establish the following strong monotonicity conditions:

[0028] For any state satisfy

[0029] (5)

[0030] in, The gradient vector represents the cost function for electricity users. The coefficient is a strong monotonicity coefficient; according to the strong monotonicity condition, there exists a unique Nash equilibrium point in the unincentivized electricity market. ,in, In addition to electricity users Considering the electricity consumption behavior of all users except those mentioned above, and the Nash equilibrium point satisfies:

[0031] (6)

[0032] Strong monotonicity guarantees the existence of a unique Nash equilibrium point in the electricity market; simultaneously, due to the applied incentives... right It is linear, so applying an excitation The system still meets the requirements - Strong monotonicity, that is, the system still satisfies equation (5) at this time, therefore, the system still has a unique Nash equilibrium point.

[0033] Furthermore, a radial basis function neural network is introduced into the design of the incentive mechanism. When electricity users refuse to share the true cost function due to privacy concerns, the neural network is used to estimate the gradient of their cost function, thereby protecting user privacy. Step 3 includes:

[0034] Step 3-1, construct the radial basis function neural network approximation model:

[0035] (7)

[0036] in, For the ideal weight vector, To approximate the error and satisfy , It is a bounded constant. The activation function is Gaussian. It is the number of neurons in a radial basis function neural network, and

[0037] (8)

[0038] It is the variance (width) of the Gaussian function. It is the expectation (mean) of the Gaussian function. Ideal weight vector. exist It is a continuous function and A constant must exist if it belongs to a bounded compact set, but its actual value is not needed during algorithm design; it is only necessary to know that a constant exists. satisfy ;

[0039] Step 3-2, Design a continuous adaptive control law:

[0040] (9)

[0041] in, It is about controlling the gain. For the ideal weight vector The estimated weight vector, The robustness coefficients satisfy , The parameters are time-varying and satisfy the following conditions: , It is a positive constant. It is the power user under the system's globally optimal condition. Electricity consumption behavior;

[0042] Step 3-3, Design and estimate weights Over time The adaptive update law:

[0043] (10)

[0044] in, and It is a time-varying adjustment parameter. It is a weight constraint compensation term, used to ensure that the following conditions are met. Under the conditions It always holds true;

[0045] Steps 3-4: Design time-varying adjustment parameters , and weight constraint compensation term The renewal law:

[0046] (11)

[0047] (12)

[0048] (13)

[0049] in, It is a positive constant, parameter and satisfy , Case 1 represents ,

[0050] Case 2 represents .

[0051] Furthermore, step 4 includes:

[0052] Step 4-1, Define the event trigger time sequence Update rate:

[0053] (14)

[0054] in, Define electricity users Incentive coefficient Controller update time, It is the conditional function triggered by the event. Based on formula (14), the segmented excitation coefficient update rule is designed as follows:

[0055] (15)

[0056] in, The continuous control signal is designed for step 2; considering the limited computing and communication resources in practice, segmented excitation coefficients are easier to implement in the electricity market than continuous excitation coefficients.

[0057] Step 4-2, construct the event trigger condition function:

[0058] (16)

[0059] in, Indicates continuous control signal With segmented excitation coefficient The difference, and All are positive numbers, auxiliary variables satisfy ;

[0060] Step 4-3, Design auxiliary variables Renewal Law:

[0061] (17)

[0062] in, The attenuation coefficient is... To adjust the parameters, recursive analysis revealed that the function value of the event triggering condition function is less than 0 after each event trigger, therefore the event triggering condition function will not appear. This prevents the event from triggering again; additionally, by introducing auxiliary variables... This can effectively prevent the system from exhibiting Zeno behavior.

[0063] Furthermore, step 5 includes:

[0064] Step 5-1: Adjust parameters to ensure that the energy consumption of electricity users is always bounded; provide the function.

[0065] (18)

[0066] To aid in research on the boundedness of user energy consumption in the electricity market under policy subsidies, among which, , , It is the Nash equilibrium point of the system without an incentive mechanism. It is a positive constant. Represent a A dimensional column vector whose elements are all equal to ;

[0067] If you adjust the parameters , , , and , making existence satisfy as well as Then we can obtain the electricity consumption behavior of electricity users. and auxiliary variables Always bounded, and satisfies

[0068] (19)

[0069] in, , , It is the difference between the Nash equilibrium point and the global optimal state of the system under no-excitation conditions. The electricity consumption behavior of power users can be obtained through (19). and auxiliary variables The upper and lower limits ensure that the electricity consumption of users in the electricity market always remains within a bounded range;

[0070] Step 5-2: Adjust parameters to match user electricity consumption behavior. asymptotic convergence to the socially optimal state of electricity consumption To maximize the overall benefits of the electricity market; give the function

[0071] (20)

[0072] The study aims to analyze changes in user electricity consumption behavior under incentives, including... It is the user's electricity consumption behavior Social optimal power consumption state The difference between them, positive numbers Influencing parameters The exponential convergence speed, It is to estimate the weights. and ideal weight vector The error between the two. Adjusting the parameters while satisfying the previous parameter constraints. , , , and , making

[0073] (twenty one)

[0074] (twenty two)

[0075] If this holds true, then the user's electricity consumption behavior It can asymptotically converge to the socially optimal state of electricity consumption. This means that the overall benefits of the electricity market will gradually be maximized over time, and can be obtained.

[0076] (twenty three)

[0077] According to Grönwall's inequality, we can obtain

[0078] (twenty four)

[0079] That is, user electricity consumption behavior Therefore, the exponential convergence is... of, Therefore, the estimation region of a radial basis function neural network can be chosen as:

[0080] (25)

[0081] Step 5-3, based on the event triggering condition (16) and auxiliary variables The update rate (17) was analyzed and found to be...

[0082] (26)

[0083] Among them, positive numbers satisfy (26) This ensures that the system does not exhibit Zeno behavior, thus guaranteeing the practical feasibility of segmented incentives in the electricity market.

[0084] An event-triggered incentive control device based on radial basis function neural network in an electricity market includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement the above-mentioned design method of event-triggered incentive mechanism based on radial basis function neural network in the electricity market.

[0085] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned design method for an event-triggered incentive mechanism based on a radial basis function neural network in the electricity market.

[0086] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned design method for an event-triggered incentive mechanism based on a radial basis function neural network in the electricity market.

[0087] As can be seen from the above technical solution, the event-triggered incentive mechanism design method based on radial basis function neural networks in the power market provided by this invention includes: estimating the unknown dynamics of the system through radial basis function neural networks, without requiring power users to disclose their own revenue functions and other private information in the power market, and designing the incentive mechanism without relying on sensitive individual data, thus further protecting the privacy and security of power users. The online training mode updates the neural network parameters, eliminating the need for pre-prepared large amounts of historical data, and adjusting parameters only based on the current input data and model state, effectively reducing data storage costs. Simultaneously, this mode gradually optimizes parameters during model operation, avoiding large-scale iterative training, and is particularly suitable for large-scale datasets and complex models, improving training efficiency. Compared to the traditional method of continuously adjusting incentive coefficients, the event-triggered mechanism only updates the incentive coefficients when preset triggering conditions are met, effectively reducing the system's communication bandwidth requirements and reducing the consumption of computing and communication resources, achieving the expected goals with less resource investment. In terms of system performance optimization, this invention ensures that users' electricity consumption behavior can be regulated and converged to the socially optimal electricity consumption state, rather than only converging to the neighborhood of the socially optimal state, achieving an equilibrium state that combines stability and accuracy.

[0088] Furthermore, this invention also provides a corresponding device for designing an event-triggered incentive mechanism based on radial basis function neural networks in the electricity market, further enhancing the practicality of the aforementioned method. This device has corresponding advantages. Attached Figure Description

[0089] To clearly present the embodiments and related technical solutions of the present invention, the accompanying drawings required for the description of the embodiments are briefly introduced below. The following drawings are only related to the embodiments of the present invention, and those skilled in the art can obtain other related drawings without creative effort based on these drawings.

[0090] Figure 1 This is a flowchart illustrating a design method for an event-triggered incentive mechanism based on a radial basis function neural network in the electricity market, provided as an embodiment of the present invention.

[0091] Figure 2 This is a schematic diagram of a non-cooperative game in the electricity market provided in an embodiment of the present invention.

[0092] Figure 3 For a period of time, the indicator function of users in the electricity market The changes.

[0093] Figure 4 For error Curve showing how it changes over time.

[0094] Figure 5 Electricity consumption behavior of electricity market users How it changes over time.

[0095] Figure 6 This is a schematic diagram of an event-triggered excitation control device based on radial basis function neural network in the power market, provided in an embodiment of the present invention. Detailed Implementation

[0096] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that the embodiments described herein are merely representative implementations of the present invention and not all possible implementations. Various modifications and improvements that can be achieved by those skilled in the art through conventional technical means based on an understanding of the core technical ideas of the present invention should be considered to fall within the protection scope of the present invention.

[0097] This invention provides a method for designing an event-triggered incentive mechanism based on a radial basis function neural network in the electricity market, such as... Figure 1 As shown, it includes the following steps:

[0098] S101. Establish a social optimal state problem model for user electricity consumption behavior in the electricity market, parameterize the state of electricity users, i.e., their electricity consumption behavior, and give the target social optimal state function.

[0099] S102. Construct a gradient descent model for the decision-making model of electricity users in the electricity market, and construct the incentive coefficients and incentive functions;

[0100] S103. Design an incentive mechanism based on radial basis function neural network to construct a dynamic decision-making mechanism that influences the electricity consumption behavior of electricity users without obtaining information on the private revenue function of electricity users.

[0101] S104. An event-triggered discretization excitation coefficient update mechanism is proposed, which reduces the consumption of limited communication resources by designing triggering event conditions.

[0102] S105. Construct a convergence analysis framework for the incentive mechanism, and ensure that the electricity consumption behavior of each user in the electricity market can eventually reach the social optimal equilibrium by adjusting the parameters, and exclude Zeno behavior to ensure the feasibility of the incentive design.

[0103] In the event-triggered incentive mechanism design method based on radial basis function neural networks in the electricity market provided in this embodiment of the invention, the ability of radial basis function neural networks to estimate unknown dynamics of the system is utilized. When designing the incentive mechanism, there is no need for electricity users in the electricity market to disclose their own revenue functions or other privacy-sensitive information, effectively protecting user privacy. Simultaneously, in the neural network parameter update method, this invention adopts an online training mode, eliminating the need for pre-prepared large amounts of historical data. Parameters are adjusted only based on the current input data and model state, effectively reducing data storage costs. Furthermore, this mode gradually optimizes parameters during model operation, avoiding large-scale iterative training, which is particularly suitable for large datasets and complex models, improving training efficiency. Regarding the incentive coefficient design, compared to the traditional method of continuously adjusting incentive coefficients, the event-triggered mechanism updates the incentive coefficients only when preset trigger conditions are met. This design effectively alleviates the pressure on communication bandwidth and reduces the consumption of computing and communication resources, achieving the system optimization goal with lower resource costs. In addition, through reasonable parameter adjustment, this invention can ensure that users' electricity consumption behavior can be regulated and converged to the socially optimal electricity consumption state, rather than merely remaining in the vicinity of the optimal state. The resulting equilibrium state possesses both stability and accuracy.

[0104] Furthermore, in a specific implementation, in the above-mentioned event-triggered incentive mechanism design method based on radial basis function neural network in the electricity market provided in the embodiments of the present invention, step S101 parameterizes the electricity consumption behavior of electricity users in the electricity market, specifically including:

[0105] Establish a state-space model and define the state-space model in the electricity market. A group of electricity users is Electricity user behavior is represented as ,in Indicates electricity user Energy consumption; establish the following incentive mechanism:

[0106]

[0107] in, Representing electricity users The cost function, Managers act through distributed controllers Apply to electricity users Incentive measures, Representing electricity users The cost function after applying incentives. The social optimal condition is defined as follows:

[0108]

[0109] in, This represents the globally optimal electricity consumption state in the electricity market, meaning that the electricity consumption behavior of each electricity user minimizes the overall cost. This is information available to managers.

[0110] Furthermore, in the above-mentioned event-triggered incentive mechanism design method based on radial basis function neural network in the electricity market provided in the embodiments of the present invention, step S102, which constructs a decision model and linear incentive for electricity users under a strongly monotone game, specifically includes:

[0111] The excitation coefficients and excitation functions are constructed as follows:

[0112]

[0113] in, Representing electricity users The incentive coefficient satisfies For piecewise constants, Representing electricity users The excitation function is given. The gradient response dynamics of the power user are as follows:

[0114]

[0115] in, This represents the negative gradient of the original cost function for electricity users. Gradient descent response dynamics describes the nature of the electricity market where individuals focus only on their immediate self-interest while ignoring the overall benefit. Furthermore, the system satisfies the strong monotonicity condition:

[0116]

[0117] in, The gradient vector represents the cost function for electricity users. It is a strong monotonic coefficient.

[0118] Furthermore, in the above-described method for designing an event-triggered incentive mechanism based on a radial basis function neural network in the electricity market provided in this embodiment of the invention, step S103 utilizes the radial basis function neural network to estimate the unknown dynamics, specifically including:

[0119] Constructing a radial basis function neural network approximation model:

[0120]

[0121] in, For the ideal weight vector, To approximate the error and satisfy , It is a bounded constant. The activation function is Gaussian. It is the number of neurons in a radial basis function neural network, and

[0122]

[0123] It is the variance (width) of the Gaussian function. It is the expectation (mean) of the Gaussian function. It's worth noting the ideal weight vector. exist It is a continuous function and If a set belongs to a bounded compact set, it must exist, and there exists a constant. satisfy ;

[0124] Design a continuous adaptive control law:

[0125]

[0126] in, It is about controlling the gain. For the ideal weight vector The estimated weight vector, The robustness coefficients satisfy , The parameters are time-varying and satisfy the following conditions: , It is the power user under the system's globally optimal condition. Electricity consumption behavior;

[0127] Design estimated weights The adaptive update law:

[0128]

[0129] in, and It is a time-varying adjustment parameter. It is a weight constraint compensation term, used to ensure that the following conditions are met. Under the conditions It always holds true;

[0130] Design time-varying adjustment parameters , and weight constraint compensation term The renewal law:

[0131]

[0132]

[0133]

[0134] in, It is a positive constant, parameter and satisfy , Case 1 represents ,

[0135] Case 2 represents .

[0136] Furthermore, in the above-mentioned event-triggered incentive mechanism design method based on radial basis function neural network in the power market provided in this embodiment of the invention, step S104 designs segmented incentive coefficients through event triggering. Using event triggering to design the incentive mechanism avoids a series of adverse effects that may arise under continuous incentive coefficients, such as network congestion and increased costs. Specifically, it includes:

[0137] Provide the event trigger time sequence Update rate:

[0138]

[0139] in, Indicates electricity user Incentive coefficient Controller update time, It is the conditional function that triggers the event. According to the formula... The rule for updating the segmented incentive coefficients is as follows:

[0140]

[0141] in, The continuous control signal is designed for step S102. Considering the limited communication and computing resources in practice, segmented excitation coefficients are easier to implement in the electricity market compared to continuous excitation coefficients;

[0142] Construct the event trigger condition function:

[0143]

[0144] in, Indicates continuous control signal With segmented excitation coefficient The difference, and All are positive numbers, auxiliary variables satisfy ;

[0145] Design auxiliary variables Renewal Law:

[0146]

[0147] in, The attenuation coefficient is... To adjust the parameters.

[0148] Furthermore, in the above-mentioned event-triggered incentive mechanism design method based on radial basis function neural network in the electricity market provided in this embodiment of the invention, step S105 ensures that the energy consumption of electricity users is always bounded and that the final electricity consumption behavior can reach the global optimum by reasonably adjusting the parameters, specifically including:

[0149] Adjust parameters , , , and , making existence satisfy as well as This leads to changes in electricity consumption behavior among power users. and auxiliary variables It is always bounded. Based on this, the parameters are further adjusted. , , , and , making

[0150]

[0151]

[0152] If this holds true, then the electricity consumption behavior of electricity users... Able to asymptotically converge to the socially optimal state of electricity consumption That is, the overall benefit of the electricity market will be maximized over time, and the estimation region of the radial basis function neural network can be chosen as:

[0153]

[0154] in,

[0155]

[0156] Furthermore, Zeno behavior does not exist, ensuring the practical feasibility of segmented incentives in the electricity market.

[0157] The technical effects achieved by this invention will be described in detail below with reference to experiments. Figure 2 The non-cooperative game-theoretic multi-agent system for the electricity market shown in this invention may include the following:

[0158] Based on step S101, the cost function form for electricity users is given as follows: ,in, Indicates electricity user Due to electricity consumption Deviating from its ideal power consumption The economic losses incurred during this period For users Sensitivity to fluctuations in electricity consumption Indicates electricity user Electricity expenses, The price of electricity is per unit of electricity consumption, and the price is affected by the total electricity consumption in the electricity market;

[0159] The specific cost functions for the five electricity users are as follows:

[0160]

[0161]

[0162]

[0163]

[0164]

[0165] in, The globally optimal electricity consumption state of the electricity market is calculated to be: The Nash equilibrium point of the electricity market under no-incentive conditions is: The initial state of the system is .

[0166] According to step S102, a linear excitation is constructed for each power user. and gradient descent model

[0167]

[0168] Based on step S103, construct the radial basis function neural network approximation model:

[0169]

[0170] in, For the ideal weight vector, To approximate the error and satisfy , It is a bounded constant. The activation function is Gaussian. It is the number of neurons in a radial basis function neural network, and

[0171]

[0172] It is the variance (width) of the Gaussian function. It is the expected value (mean) of the Gaussian function. The number of neurons... Set to 7, the expected value (mean) of the Gaussian function for each neuron is designed as follows: , , , , , , ,in, The variances are all set to Due to the ideal weight vector exist It is a continuous function and It must exist if it belongs to a bounded compact set, given a constant. Make Hengcheng was established;

[0173] Design a continuous adaptive control law:

[0174]

[0175] in, It is about controlling the gain. For the ideal weight vector The estimated weight vector, robustness coefficient satisfy , The parameters are time-varying and satisfy the following conditions: , Both are 5. The initial values ​​are respectively , , , , ;

[0176] Design estimated weights The adaptive update law:

[0177]

[0178] in, and It is a time-varying adjustment parameter. It is a weight constraint compensation term, used to ensure that the following conditions are met. Under the conditions This always holds true. To ensure the initial conditions... Established, Order ;

[0179] Design time-varying adjustment parameters , and weight constraint compensation term The renewal law:

[0180]

[0181]

[0182]

[0183] Among them, control gain Time-varying parameters and satisfy , , The initial value is set to , The initial value is set to Case 1 represents ,

[0184] Case 2 represents .

[0185] Based on step S104, the event trigger time sequence is given. Update rate:

[0186]

[0187] in, Indicates electricity user Incentive coefficient Update time, It is the conditional function triggered by the event. Based on formula (14), the segmented excitation coefficient update rule is designed as follows:

[0188]

[0189] in, The continuous control signal designed for step S102;

[0190] Construct the event trigger condition function:

[0191]

[0192] in, Indicates continuous signal With segmented excitation coefficient The difference, setting parameters , Auxiliary variables satisfy ;

[0193] Design auxiliary variables Renewal Law:

[0194]

[0195] in, The attenuation coefficient is... To adjust the parameters.

[0196] According to step S105, the parameter constraints are verified (if not met, adjustments are made), and it is found that there are... satisfy and Therefore, the energy consumption of electricity users and auxiliary variables It is always bounded. Based on this, the parameters... , , , and Further meet , Therefore, the user's electricity consumption behavior It can asymptotically converge to the socially optimal state of electricity consumption. This means that the overall benefits of the electricity market can be maximized. Furthermore, Zeno behavior does not exist, ensuring the practical feasibility of segmented incentives in the electricity market.

[0197] The simulation results of this experiment are shown in the figure below. Figures 3-5 As shown. For example Figure 2 The illustrated non-cooperative game-theoretic multi-agent system in the electricity market Figure 3 The present invention provides the indication function for each electricity user in the event-triggered incentive mechanism design method based on radial basis function neural network in the electricity market provided above. The changes verified that the designed event triggering mechanism can effectively regulate... Ensure that it meets ; Figure 4 For error The curve showing the change over time, and the simulation results, indicate that the error... The method asymptotically converges to zero, verifying the convergence of the proposed method. Figure 5 Electricity consumption behavior of various electricity users The changes over time show that the electricity consumption behavior of each electricity user eventually stabilizes at the expected equilibrium point with high accuracy, consistent with theoretical analysis.

[0198] This invention uses RBFNN to estimate unknown dynamics of a system online, without requiring users in the electricity market to disclose their own revenue functions or other private information (such as...). This implements an incentive mechanism design that is independent of sensitive data; it adopts an online parameter update strategy, which can adjust the RBFNN weights using only the current input data. This eliminates the need for extensive historical data preparation and batch training, reducing computational costs. Furthermore, this mode progressively optimizes parameters during model execution, avoiding large-scale iterative training, making it particularly suitable for large datasets and complex models, thus improving training efficiency. Additionally, when the indicator function... At that time, the event triggers an update of the incentive coefficient. Compared with the traditional continuous adjustment method, this effectively reduces the number of communications between the manager and the distributed controller, thus reducing resource consumption. Finally, simulations were conducted... Figure 4 , Figure 5 Indicates the user's electricity consumption behavior It can converge to the socially optimal state of electricity consumption. Rather than its neighboring regions.

[0199] Based on the same inventive concept, this invention provides an event-triggered excitation control device for the electricity market based on a radial basis function neural network. This device utilizes the nonlinear approximation capability of the radial basis function neural network and the event-triggered mechanism to achieve intelligent regulation of user electricity consumption behavior in the electricity market. The device's hardware architecture is centered on a memory 50 and a processor 51, which work together to implement the relevant control methods.

[0200] The memory 50 stores the power market event triggering control program based on a radial basis function neural network, and the processor 51 executes the program and performs neural network operations and event triggering judgments. The processor 51 adopts a configurable multi-core architecture, integrating a digital signal processing unit and a programmable logic device. The system connects necessary sensor modules and input / output interfaces 53 via a communication bus 56. Each functional component can be flexibly configured according to actual application requirements, such as including a display screen 52, input / output interfaces 53, a communication interface 54, a power supply 55, and sensors 57. Figure 6 The device structure shown is for illustrative purposes only; components can be added or removed as needed in actual applications.

[0201] To ensure the comprehensive implementation of the technical solution, this invention also provides a computer-readable storage medium. This medium contains a pre-defined computer program that, once executed by a processor, enables an event-triggered incentive control method based on radial basis function neural networks in the electricity market. Its specific operating logic and steps are consistent with the control flow detailed in the previous device embodiments, and therefore will not be repeated. Through the coordinated operation of the storage medium and the processor, this innovative control method can be stably applied to the electricity market, ensuring the feasibility and practicality of the technical solution.

[0202] In summary, this invention provides a method for designing an event-triggered incentive mechanism based on radial basis function neural networks (RBFNNs) in the electricity market. By estimating unknown dynamics of the system using RBFNNs, it eliminates the need for users to disclose their own revenue functions or other private information, thus designing the incentive mechanism without relying on sensitive individual data and further protecting user privacy. The online training mode updates neural network parameters, eliminating the need for extensive historical data preparation and adjusting parameters only based on current input data and model state, effectively reducing data storage costs. Furthermore, this mode progressively optimizes parameters during model execution, avoiding large-scale iterative training, making it particularly suitable for large datasets and complex models, thus improving training efficiency. Compared to traditional methods of continuously adjusting incentive coefficients, the event-triggered mechanism updates incentive coefficients only when preset trigger conditions are met, effectively reducing the system's communication bandwidth requirements and minimizing computational and communication resource consumption, achieving system optimization goals with lower resource consumption. At the system performance optimization level, this invention ensures that users' electricity consumption behavior converges to the socially optimal electricity consumption state, rather than just converging to a neighborhood of the socially optimal state, achieving an equilibrium state that combines stability and accuracy.

[0203] It should be understood that the terms "including" and "comprising" used in this specification are open-ended, and the described technical solutions are not limited to the explicitly listed modules, but should also include other functional components necessary for implementing this algorithm. In actual deployment, the system may integrate additional necessary units such as data processing, communication interfaces, or auxiliary control modules as needed.

Claims

1. A method for designing an event-triggered incentive mechanism based on a radial basis function neural network in a power market, characterized in that, The method comprises the following steps: Step 1, a social optimal state problem model of the power consumption behavior of a user in a power market is established, the state of the power user, i.e. the power consumption behavior, is parameterized, and a target social optimal state function is given; Step 2, a decision model of the user in the power market is constructed as a gradient descent model, and an incentive coefficient and an incentive function are constructed; Step 3, an incentive mechanism based on a radial basis function neural network is designed, the dynamic decision of the incentive function on the power consumption behavior of the power user is constructed without obtaining information of a private profit function of the power user; specifically comprising: Step 3-1, a radial basis function neural network approximation model is constructed: ；（7） wherein, is an ideal weight vector, is an approximation error and satisfies , is a bounded constant, is a set of electricity users in the electricity market, is the electricity consumption behavior of the electricity users, wherein denotes the energy consumption of the electricity user , is a Gaussian activation function, is the number of neurons of the radial basis function neural network, and ；（8） is the variance of the Gaussian function, is the expectation of the Gaussian function; ideal weight vector In is a continuous function and belongs to a bounded and compact set, there exists a constant satisfying ; Step 3-2, a continuous adaptive control law is designed: ；（9） in, It is about controlling the gain. For the ideal weight vector The estimated weight vector, The robustness coefficients satisfy , The parameters are time-varying and satisfy the following conditions: , It is a positive constant. It is the power user under the system's globally optimal condition. Electricity consumption behavior; Step 3-3, Designing the estimated weight over time adaptive update law: ；（10） wherein and are time-varying adjustment parameters, is a weight constraint compensation term for ensuring that always holds under the condition that always holds under the condition that Step 3-4, design estimation weights Time-varying tuning parameters in adaptive update law (10) , And weight constraint compensation term Update law: ；（11） ；（12） ；（13） in, It is a positive constant, parameter , satisfy , Case 1 represents , Case 2 represents ; Step 4, an event-triggered discretization incentive coefficient updating mechanism is proposed, and a trigger event condition is designed to reduce consumption of limited communication resources; Step 5, a convergence analysis framework of the incentive mechanism is constructed, the power consumption behavior of each user in the power market can finally reach a social optimal equilibrium by adjusting parameters, and Zeno behavior is excluded to ensure feasibility of the incentive design.

2. The method of claim 1, wherein, The power consumption behavior of the user in the power market is modeled, and step 1 specifically comprises: Step 1-1: Establish a model and define the electricity market. The set of electricity users is Electricity users' electricity consumption behavior is represented as ,in Indicates electricity user Energy consumption; Step 1-2, the following incentive mechanism is established: ； （1） wherein a cost function of the power consumer, a cost function of the power consumer, an incentive measure imposed by the manager on the power consumer via the distributed controller, an incentive measure imposed by the manager on the power consumer via the distributed controller, a cost function of the power consumer after imposing the incentive, a cost function of the power consumer after imposing the incentive, a cost function of the power consumer after imposing the incentive. Step 1-3, the following social optimal condition is defined: ； （2） wherein, represents the global optimal power consumption state of the power market, i.e. the power consumption behavior of each power consumer at this time minimizes the global cost, which is the information that the manager can obtain.

3. The method of claim 2, wherein, A linear incentive is used to affect the gradient dynamic decision process of the power user under a strong monotonic game, and step 2 specifically comprises: Step 2-1, the following incentive coefficient and incentive function are constructed: ； （3） wherein a representative electric power user a stimulation coefficient, satisfying is a piecewise constant, a representative electric power user a stimulation function; Step 2-2, defining the power consumer behaviour The gradient response dynamics over time are as follows: ；（4） wherein, is the rate of change of the power consumption behavior of the power consumer i over time, denotes the negative gradient of the original cost function of the power consumer ; Step 2-3, the following strong monotonicity condition is established: For any Satisfies ；（5） in, The gradient vector represents the cost function for electricity users. The coefficient is a strong monotonicity coefficient; according to the strong monotonicity condition, there exists a unique Nash equilibrium point in the electricity market without any incentives. ,in, In addition to electricity users Considering the electricity consumption behavior of all users except those mentioned above, and the Nash equilibrium point satisfies: ； （6） The strong monotonicity ensures that there is only one Nash equilibrium point in the power market; Furthermore, the system after the application of the incentive still satisfies strong monotonicity, i.e. the system still satisfies equation (5) at this point, and thus, the system still has a unique Nash equilibrium point.

4. The method of claim 1, wherein, Step 4 specifically comprises: Step 4-1, defining event trigger time series of the update rate: ；（14） wherein The power consumer Incentive coefficient The controller update time, is an event-triggered condition function; the piecewise incentive coefficient update rule is designed according to formula (14) as follows: ；（15） wherein is a continuous control signal; Step 4-2, an event trigger condition function is constructed: ；（16） wherein, denotes a continuous control signal with a piecewise excitation coefficient of the difference, and are normal numbers, auxiliary variables satisfy ; Step 4-3, design of auxiliary variables Update law: ；（17） wherein, is a decay coefficient, is a tuning parameter; by recursive analysis, the function value of the event-triggering condition function will be less than 0 after each event trigger, so that the event-triggering condition function does not occur again, so that the event no longer triggers; by introducing an auxiliary variable , the Zeno behavior of the system is avoided.

5. The method of claim 4, wherein, The parameters are adjusted to ensure that the energy consumption of the power user is always bounded and the final power consumption behavior can achieve global optimization, and step 5 specifically comprises: Step 5-1, the parameters are adjusted to ensure that the energy consumption of the power user is always bounded; a function is given ；（18） to assist in determining boundedness of user energy consumption in electricity markets with policy subsidies, wherein, , , is a Nash equilibrium point of the system without incentive mechanisms, is a positive constant, denotes a column vector with all elements equal to 1; If the adjustment parameters , , , and are such that there exists satisfying and , the user's electricity consumption behavior and the auxiliary variable are always bounded and satisfy ；（19） wherein, , , is the difference between the Nash equilibrium point and the global optimal state of the system without incentives; the upper and lower bounds of the system state and the auxiliary variable are obtained through (19), which ensures that the electricity consumption of users in the electricity market always remains within a bounded range; Step 5-2, adjust parameters so that user's electricity behavior Asymptotically converges to the social optimal electricity state So that the overall benefit of the electricity market is maximized; give the function ；（20） to analyze the change of power user's behavior under incentive, wherein, state error is the difference between user's behavior and the social optimal state , normal number influencing parameter exponential convergence speed, error between estimated weight and ideal weight vector ; under the parameter constraint condition before meeting, adjust parameter , , , and , so that ；（21） ；（22） Always, so that the user's electricity behavior can asymptotically converge to the social optimal electricity state that is, the overall interests of the electricity market will gradually reach maximization over time, and push ；（23） Further according to the Grönwall inequality, the following can be obtained ；（24） i.e. user electricity behavior converges to , where ; therefore, the estimation region of the radial basis function neural network is selected as: ；（25） Step 5-3, based on the event triggering condition (16) and auxiliary variables The update rate (17) was analyzed and found to be... ；（26） where k is a positive constant satisfies ; (26) ensures that the system does not have Zeno behavior, guaranteeing the practical feasibility of the piecewise incentive in the electricity market.

6. An event-triggered incentive control apparatus based on a radial basis function neural network in a power market, characterized by, The computer program is executed by the processor to realize the event-triggered incentive mechanism design method based on the radial basis function neural network in the power market.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the event-triggered incentive mechanism design method based on the radial basis function neural network in the power market.

8. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the event-triggered incentive mechanism design method based on the radial basis function neural network in the power market.

Images