Multi-agent collaborative optimization scheduling method and device based on dynamic rules of peak regulation market

Abstract

The application discloses a kind of multi-agent collaborative optimization scheduling method and device based on peak regulation market dynamic rule formulation, according to resource physical characteristics and regulation flexibility, resource in multi-agent is finely classified, and multi-dimensional classification system is established, to provide data basis for rule parameter dynamic design;For each resource, establish accurate description model, quantify its feasible region and cost characteristics in peak regulation market;Establish the dynamic association between resource and peak regulation market rule, embed rule parameters into optimization model, and construct the optimal rule model of power market containing multi-agent;Build double-layer collaborative optimization strategy, solve the model and output the optimal rule of current peak regulation market, realize market rule dynamic optimization and resource collaborative scheduling.The application breaks through the static limitation of traditional peak regulation market rule, realizes the dynamic matching of rule parameters and resource characteristics through multi-agent collaborative optimization, effectively improves new energy consumption capacity and market operation efficiency.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of new energy power, and particularly relates to a multi-agent collaborative optimization scheduling method and device based on dynamic rules of peak regulation market. The method is particularly aimed at the collaborative scheduling problem of peak regulation resources in the power market, and finds the optimal constraint rules (minimum duration and minimum capacity) through an algorithm to balance the market clearing efficiency and economic benefits. BACKGROUND

[0002] With the large-scale access of high-proportion new energy to the power system, the output fluctuation and anti-peak regulation characteristics of intermittent power sources such as wind power and photovoltaic power are significantly intensified, leading to the continuous expansion of the system net load peak-valley difference and the complex characteristics of multi-time and space scale coupling of peak regulation demand.

[0003] The traditional peak regulation market mechanism generally relies on static rule constraints (such as fixed minimum duration and unified capacity threshold), and its inherent defects are increasingly prominent: firstly, the static rule has insufficient adaptability, and the existing threshold parameters cannot dynamically respond to real-time working conditions such as new energy output prediction deviation, load demand mutation and network congestion, leading to the mismatch between market clearing capacity and actual peak regulation demand, and even causing wind and light curtailment or standby capacity redundancy in extreme scenarios; secondly, the optimization calculation efficiency is low, and for the peak regulation market involving multiple agents, the mixed integer programming model has strong coupling between resource constraints and market rules, and the problem dimension increases exponentially with the number of agents, so the traditional centralized solver faces the "dimension disaster" bottleneck, and it is difficult to meet the timeliness requirements of intra-day rolling clearing; thirdly, the economic target is disconnected with the rule parameter, and the fixed threshold separates the quantitative correlation between the rule parameter and the market core target, which may either exclude flexible resources due to high threshold, reducing market liquidity, or cause disordered competition of resources due to low threshold, increasing system operation cost. Although existing researches try to improve the model uncertainty handling capability through robust optimization or scenario analysis method, they are still limited to single-layer optimization framework, and do not study the core contradiction between dynamic optimization of rule parameters and multi-agent distributed collaboration, so it is difficult to realize the collaborative improvement of global economy, safety and calculation efficiency. Therefore, it is urgent to build a peak regulation market rule generation method that takes into account dynamic adaptability, high calculation efficiency and economic optimality, to support the flexible and reliable operation of new power systems.

[0004] In summary, the current peak regulation market rule generation method based on multi-agent collaborative optimization has many deficiencies, and it is difficult to meet the needs of efficient operation and optimized transactions of new power systems. Therefore, there is an urgent need for a dynamic rule generation method that is more accurate, comprehensive, considers resource collaboration and market factors, to promote the development of ancillary service markets and promote the consumption of new energy and the stable operation of power systems. SUMMARY

[0005] In view of the problems of poor dynamic adaptability, low calculation efficiency and economic inefficiency caused by traditional fixed rules in the peak regulation market participated by multiple agents, the application provides a multi-agent collaborative optimization scheduling method and device based on dynamic rule making of the peak regulation market, generates optimal rule parameters through an optimization algorithm, and reduces market clearing complexity.

[0006] The application aims to realize the following technical solutions: in a first aspect, the application provides a multi-agent collaborative optimization scheduling method based on dynamic rule making of the peak regulation market, including the following steps:

[0007] S1, finely classifying resources in the multiple agents according to resource physical characteristics and regulation flexibility, and establishing a multi-dimensional classification system;

[0008] S2, establishing an accurate mathematical description model for each type of resource, quantifying the feasible region and cost characteristics of the resource in the peak regulation market;

[0009] S3, establishing a dynamic association between the resource and the peak regulation market rule, and constructing an optimal rule model of the power market containing multiple agents based on the rule parameters;

[0010] S4, based on the mathematical description model constructed in step S2 and the optimal rule model of the power market constructed in step S3, constructing a double-layer collaborative optimization strategy, solving and outputting the optimal rule of the current peak regulation market, and realizing dynamic optimization of the market rule and collaborative scheduling of the resource.

[0011] Further, the specific process of step S1 is as follows: the internal resources of the agent are divided into uncontrollable resources, semi-controllable resources and controllable resources; each type of resource is assigned a dynamic label including a peak regulation capacity margin and a response delay time, and the label value is updated in real time to construct a resource label matrix; real-time resource data is collected to construct a time sequence feature vector including photovoltaic power, fan speed and energy storage SOC data.

[0012] Further, the specific process of step S2 is as follows:

[0013] 1) for controllable resources, an uncertainty model of new energy is established, and a robust optimization is used to describe the resource output interval;

[0014] 2) a dynamic constraint model of energy storage resources is established, a relationship between depth of discharge (DOD) and cycle life is established, and an energy storage model considering charging and discharging efficiency and capacity limitation is established;

[0015] 3) for semi-controllable resources, an indirect response model between resources is established, a relationship between user power and electricity price is established, and a load model based on historical data induction of user behavior and electricity demand law is established;

[0016] 4) For uncontrollable resources, historical data is used to train the prediction model. After obtaining the baseline power curve, a probability distribution model is used to describe the prediction uncertainty.

[0017] Furthermore, the rule parameters for establishing a dynamic relationship between resource and peak-shaving market rules include minimum duration and minimum capacity. The optimal rule model for the electricity market is a two-layer optimization model for the electricity market involving agents, which includes a rule generation layer and an internal optimization layer. The dynamic rules are adjusted daily based on forecast data and historical operating data.

[0018] Furthermore, the rule generation layer is as follows:

[0019] Based on the global optimality rule parameters for power grid trading:

[0020] in, The minimum duration for an agent to participate in the peak-shaving market. To minimize capacity, and to balance grid computational complexity, overall participation, and grid stability, an objective function is designed to maximize the objective:

[0021]

[0022] in, The total participation capacity of internal intelligent agents participating in the electricity market. Historical participation capacity benchmarks for internal agents participating in the electricity market; This refers to the demand satisfaction rate of the power grid, which is the ratio of the cleared winning bid capacity to the actual demand. , , Weighting coefficients for different objectives; The computational complexity of market clearing.

[0023] Furthermore, the internal optimization layer is as follows:

[0024] Each agent is modeled individually, and a mixed-integer programming model is constructed, embedding dynamic rule constraints:

[0025]

[0026]

[0027]

[0028]

[0029] in, Indicates the internal resources of the intelligent agent. The sum of revenue from power generation, revenue from energy storage discharge, and revenue from selling electricity to the load at any given moment; is the electricity price of the agent, is the electricity price of the agent to the load; represents the sum of the generation cost and the energy storage charging and discharging cost of the system at time, is the total power of the internal new energy, is the energy storage adjustment cost, is the power of the energy storage at t time, and N is the total number of resources participating in the peak regulation market; by accumulating the income and cost of each time and maximizing the accumulated value, the economic benefit maximization of the agent in the given time period is realized.

[0030] Further, the rule generation layer dynamically generates optimal market rule parameters based on an improved multi-objective particle swarm algorithm, including minimum duration and minimum capacity ; specifically as follows:

[0031] Discrete velocity update formula:

[0032]

[0033] Discrete position update formula:

[0034]

[0035] The auxiliary function is defined as:

[0036]

[0037] wherein, represents the velocity value of the particle in dimension at the th iteration, and the initial velocity is randomly initialized in the interval ; is an inertia weight parameter, which is a scalar coefficient controlling the influence strength of the historical velocity on the current velocity; and are cognitive and social coefficients, respectively; and are independent random variables uniformly distributed in the interval ; represents the individual historical optimal position of the particle in dimension ; represents the guiding position of the current group in dimension ; represents the current position of the particle in dimension at the th iteration; is the displacement is the adjustment function; is the boundary constraint function.

[0038] Further, each agent including new energy, energy storage and adjustable load in the internal optimization layer independently solves a mixed integer programming model (MILP), responds to the upper layer rule constraint and feeds back the economic index, realizes the interaction of upper and lower layer parameters through an alternating direction multiplier method (ADMM), and ensures the global optimality and the efficiency of distributed calculation.

[0039] In a second aspect, the application further provides a multi-agent collaborative optimization scheduling device based on dynamic rule making of a peak regulation market, comprising a memory and one or more processors, the memory storing executable code, and the processor implementing the multi-agent collaborative optimization scheduling method based on dynamic rule making of a peak regulation market when executing the executable code.

[0040] In a third aspect, the application further provides a computer readable storage medium having a program stored thereon, and the program, when executed by a processor, implements the multi-agent collaborative optimization scheduling method based on dynamic rule making of a peak regulation market.

[0041] Compared with the prior art, the application has the following beneficial effects:

[0042] (1) The application enhances the dynamic adaptability of the power grid rules, realizes the self-adaptive optimization of the peak regulation market rule parameters (minimum duration and minimum capacity) through the dynamic rule generation layer and the real-time data rolling refresh mechanism, effectively responds to the new energy output fluctuation and load demand change, and improves the response ability of the system to complex scenarios.

[0043] (2) The application can realize the win-win of economic benefits of the power grid and the agent, combines the fine resource cost model (energy storage life loss and adjustable load price response) and the multi-agent collaborative optimization, maximizes the resource income and reduces the overall peak regulation cost, and significantly improves the energy utilization efficiency.

[0044] (3) The application improves the calculation efficiency, adopts a double-layer decoupling architecture (upper layer multi-objective particle swarm algorithm + lower layer distributed ADMM), separates the rule generation and resource scheduling calculation, reduces the model complexity, and supports the intra-day high-frequency optimization.

[0045] (4) The application guarantees the stability of the power grid, embeds the power grid demand satisfaction index and the robust regulation mechanism, preferentially meets the peak regulation demand in critical periods, and enhances the system operation reliability and equipment life. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is a flowchart of the peak regulation market dynamic rule making method based on multi-agent collaborative optimization in the embodiment of the application.

[0047] Figure 2 Participation curve under different market rules.

[0048] Figure 3 Demand satisfaction curve under different market rules.

[0049] Figure 4 Figure is a structural schematic diagram of the multi-agent collaborative optimization scheduling device based on the dynamic rules of the peak regulation market in the embodiment of the present application. DETAILED DESCRIPTION

[0050] The present application will be further described and explained with the aid of the accompanying drawings and specific embodiments.

[0051] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described and explained in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0052] The present application will use the terms commonly used by those skilled in the art to describe various aspects of the illustrative embodiments to convey the essence of their work to other skilled persons in the art. However, it is obvious to those skilled in the art that some alternative embodiments can be practiced using parts of the described aspects. For the purpose of explanation, specific numbers, materials and configurations are described to provide a thorough understanding of the illustrative embodiments. However, it is obvious to those skilled in the art that alternative embodiments can be implemented without specific details. In other cases, some well-known features are omitted or simplified in order not to obscure the illustrative embodiments.

[0053] Please refer to Figure 1 In a preferred embodiment of the present application, a multi-agent collaborative optimization scheduling method based on dynamic rules of the peak regulation market is provided, which comprises the following steps:

[0054] S1, according to the resource physical characteristics and the regulation flexibility, the resources in the multi-agent are classified in detail, a multi-dimensional classification system is established, and a data basis is provided for dynamic design of rule parameters.

[0055] Controllable resources: such as new energy, energy storage system, linear / nonlinear constraint model is used to accurately characterize its climbing rate, capacity limit and other physical characteristics.

[0056] Semi-controllable resources: including electric vehicle cluster (EV), interruptible load, a response potential model based on user behavior characteristics needs to be established.

[0057] Non-controllable resources: resources not controlled by agents such as part of the load, historical data is used for prediction.

[0058] A dynamic label is given to each type of resource (such as peak shaving capacity margin, response delay time, etc.), and the label value is updated in real time to build a resource label matrix. Real-time resource data (such as photovoltaic power, fan speed, and energy storage SOC) is collected to build a time series feature vector. For uncontrollable resources, the upper and lower limits are both predicted values:

[0059]

[0060] wherein, is the maximum adjustment capacity of the ith resource, is the minimum adjustment capacity of the ith resource, is the prediction error of the ith resource, is the efficiency coefficient of the ith resource, is the response delay of the ith resource.

[0061] S2, an accurate description model is established for each type of resource to quantify its feasible region and cost characteristics in the peak shaving market;

[0062] For controllable resources, an uncertainty model of new energy is established, and a robust optimization is used to describe the resource output interval:

[0063]

[0064] wherein, is the total power of internal new energy, is the predicted value of the resource, According to the historical quantile of the resource prediction error, the uncertainty of the resource is quantified.

[0065] A dynamic constraint model of energy storage resources is established, and the relationship between the depth of charge and discharge (DOD) and the cycle life is proposed:

[0066]

[0067] wherein, is the cycle life, is the depth of charge and discharge, and is the correlation coefficient.

[0068] It is converted into a peak shaving cost term:

[0069]

[0070] wherein, is the energy storage adjustment cost, is the total cost of energy storage configuration, is the power of energy storage at time t.

[0071] For semi-controllable resources, an indirect response model between resources is established to establish the relationship between user power and price, that is, when the price is high, the user is in a power reduction state; when the price is low, the user is in a power increase state:

[0072]

[0073] At the same time , Satisfy:

[0074]

[0075]

[0076] In the formula, P(t) represents the power consumption of the user at time t, and P(t) represents the model parameters obtained by training historical data at time t, , and P(t) represents the upper limit of the power consumption of the user at time t, the lower limit of the power consumption and the reference power, , and P(t) represents the high price standard, the low price standard and the reference price of the user at time t, P(t) represents the price set by the agent for the user at time t.

[0077] For uncontrollable resources, after obtaining the reference power curve by training the prediction model with historical data, a probability distribution model is used to describe the prediction uncertainty.

[0078] Considering the intermittency and volatility of new energy power generation and the charging and discharging characteristics of energy storage devices, the scheduling strategy inside the agent is formulated. Through optimization model calculation, the energy utilization efficiency is improved, the phenomenon of abandoned wind and light is reduced, and the stable operation of the system is guaranteed.

[0079] S3, establish the dynamic correlation between resources and peak regulation market rules, embed dynamic rule parameters (minimum duration, minimum capacity) into the optimization model, and build a multi-agent optimal rule model for the electricity market. The agent participates in the double-layer optimization model of the electricity market, which includes the rule generation layer and the internal optimization layer. The dynamic rule parameters are adjusted based on the predicted data and historical operation data rolling refresh every day.

[0080] Upper layer (rule generation layer): form rule parameters according to global optimization of power grid transactions:

[0081] Among them, is the minimum duration (15min) of the agent participating in the peak regulation market, The minimum capacity (MW). To balance the grid calculation complexity, overall economy and grid stability, the objective function is designed to maximize the target:

[0082] Wherein, is the total participation capacity of the internal agent participating in the electricity market, is the historical participation capacity benchmark value of the internal agent participating in the electricity market; is the demand satisfaction of the grid, i.e. the ratio of the winning capacity after clearing to the actual demand; , , is the weight coefficient of different objectives; is the market clearing calculation complexity.

[0083] The participation parameter is defined as the total participation capacity of each agent participating in the market The specific calculation is as follows:

[0084]

[0085] The calculation complexity is defined as the combination of the market rules and the limitations of and and their nonlinear effects, and the total calculation complexity can be modeled as:

[0086]

[0087] Wherein, is a constant related to market structure and algorithm efficiency, which needs to be fitted from historical market clearing operation data.

[0088] The purpose of the grid demand satisfaction is to consider the balance of supply and demand in the peak regulation market, and the day-ahead clearing result needs to basically meet the market needs. The grid will clear the capacity reported by each agent and issue the winning amount, i.e. the actual capacity of the agent participating in the market. The grid demand satisfaction parameter represents the matching degree of the winning amount obtained through the market clearing in the day-ahead to the total system peak regulation demand, and is used to appropriately reduce the number of winning units in deep peak regulation, improve the winning amount of single unit, and the load rate of the first bid is temporarily set to not higher than 40% of the rated capacity of the unit. This parameter comprehensively reflects the adjustment effect of market rules on supply and demand balance, and is one of the key optimization objectives in the rule generation layer objective function. The definition is as follows:

[0089] (1) Basic expression

[0090]

[0091] In the formula, represents the resource The mid-peak regulation capacity (MW) in the time period ; The resource The actual schedulable coefficient (considering response delay, prediction error, etc.) in the time period ; , representing the total regulation demand (MW), The predicted demand for the time period ; The total number of time periods in the optimization cycle; The total number of resources participating in the regulation market.

[0092] (2) Add space-time weight

[0093] Space-time weight factor: give different priorities to regulation demand in different time periods and regions

[0094]

[0095] In the formula: The weight of the time period (peak time , valley time ); The capacity effective coefficient of the resource in the region (considering network congestion).

[0096] Lower layer (internal optimization layer): model each agent separately, considering them as rational subjects, build a mixed integer programming model, and embed dynamic rule constraints:

[0097]

[0098]

[0099]

[0100]

[0101] Where, The sum of the electricity sales revenue of the agent to the grid and the electricity sales revenue to the load at time . The electricity sales price of the agent, The electricity sales price of the agent to the load. The sum of the generation cost of the system at time , and the charging and discharging cost of the energy storage. By accumulating the revenue and cost at each time, and maximizing this accumulated value, the economic benefit of the agent in the given time period can be maximized.

[0102] S4, based on the mathematical model constructed in step S2 and the double-layer model constructed in step S3, a double-layer collaborative optimization strategy is constructed, the model is solved and the optimal rules of the current peak regulation market are output, and dynamic optimization of market rules and resource collaborative scheduling are realized.

[0103] The upper layer (rule generation layer) dynamically generates optimal market rule parameters (minimum duration ).

[0104] Market rule parameters (such as minimum duration) are usually integers (for example: 15 minutes, 30 minutes). The traditional MOPSO particle position is a continuous value, which needs to be mapped to a discrete space. Simple rounding may lead to invalid solutions (such as negative numbers or out-of-range), low search efficiency or local optimum.

[0105] The improved multi-objective particle swarm optimization algorithm redesigns the discrete position update rule, designs a special particle position update formula, and directly operates in the discrete space (integer domain).

[0106] Discrete velocity update formula:

[0107]

[0108] Discrete position update formula:

[0109]

[0110] Where the auxiliary function is defined as:

[0111]

[0112] Where, represents the particle in the dimension on the th iteration, which is a real number type, carrying the motion trend information of the particle in the discrete space. The initial speed is usually randomly initialized in the interval; is the inertia weight parameter, which is a scalar coefficient controlling the influence strength of the historical speed on the current speed. A higher value enhances the global exploration ability, and a lower value promotes local development. and are the cognitive coefficient and the social coefficient, respectively, which are positive real number acceleration constants adjusting the tendency degree of the particle to individual experience and group experience learning; and are independent random variables uniformly distributed in the interval , which introduces necessary random disturbance for the update process. represents the particle in the dimension The individual historical optimal position (integer) is used to record the best historical solution of the particle in the Pareto sense; Indicates the current group in dimension The guiding position (integer) is determined from the non-dominated solution set through an external archive selection mechanism (such as crowding distance or grid density); Represents particles In dimensions Upper The current position (integer) of the next iteration. It is displacement The adjustment function, where Extracting displacement direction or -1 Ensure the minimum displacement is 1 unit to avoid search stalling caused by zero displacement; It is a boundary constraint function that forces the position values ​​to lie in the dimension. feasible domain Inside( and (These are the lower and upper bounds for integers, respectively), and out-of-bounds solutions are handled through projection.

[0113] To avoid invalid solutions, improve search efficiency and accuracy, and make it easier to find the optimal or near-optimal combination of integer parameters.

[0114] Each agent (new energy, energy storage, adjustable load, etc.) in the lower layer (internal optimization layer) independently solves the mixed integer programming model (MILP), responds to the upper layer rule constraints, and feeds back economic indicators.

[0115] The alternating direction multiplier method (ADMM) is used to achieve parameter interaction between upper and lower layers, ensuring global optimality and the efficiency of distributed computing.

[0116] The two-layer collaborative optimization process is as follows:

[0117] 1. Initialization: Set initial rule parameters ADMM penalty coefficient Particle swarm parameters.

[0118] 2. Outer layer iteration (upper layer MOPSO):

[0119] Generate candidate solution set for rule parameters .

[0120] For each Call the lower-level ADMM solution process to obtain .

[0121] Calculate fitness and update the particle swarm to select the optimal solution.

[0122] 3. Inner Iteration (Lower-level ADMM):

[0123] Each agent receives the current rule parameters , updates the local constraints.

[0124] Solve the local optimization problem in parallel, upload the particle position to the coordinator.

[0125] The coordinator updates the global variables and judges convergence.

[0126] 4. Termination and output: when the upper MOPSO converges, output the optimal rule parameter set and the corresponding agent peak regulation scheme.

[0127] Application example

[0128] The method described in the application is written using Python, and the GUROBI mathematical optimizer is called to solve the model when solving the model, and the implementation effect is shown for the case data.

[0129] Operating environment:

[0130] 12thGenIntel (R) Core (TM) i5-125003.00GHz, 16GB memory, Microsoft Windows10X64

[0131] GUROBI10.0.2

[0132] Implementation results:

[0133] A coastal city-level power grid has a high proportion of new energy access (wind power accounts for 30%), and the daily load fluctuation is significant, with a peak-valley difference of 1500MW. The traditional peak regulation method relies on flexible transformation of thermal power, and has problems such as high regulation cost and new energy curtailment rate exceeding 8%. The double-layer optimization model proposed in this patent is applied to build an agent model containing wind power, energy storage, and adjustable load, participate in the day-ahead peak regulation market, and realize efficient resource coordination and dynamic rule optimization. In terms of hardware configuration, a dispatch center server is set up, the upper MOPSO algorithm and ADMM coordination module are deployed, and the agent local optimizer is deployed on the edge computing node.

[0134] A power market containing 100 agents is simulated. The particle swarm iteration parameters are set, and the dynamic inertia weight . The target function weight coefficient is set to (economic efficiency), (grid stability), (computing efficiency). The economic efficiency and demand satisfaction degree curve of the power grid under different rules can be referred to Figure 2 Figure 3The greater the rule constraint is, the smaller the two indicators are. The original rule of the peak regulation market is [4, 5], and the optimal rule finally output by the double-layer model is: .

[0135] Compared with the original fixed rule of the peak regulation market, the calculation complexity of the optimized market is greatly reduced, and the difference between the other indicators is not large.

[0136]

[0137] The present application effectively solves the problems of rigid rules and high calculation complexity of the traditional peak regulation market through resource classification modeling, rule-market dynamic correlation and optimization algorithm, realizes the collaborative optimization of economy, stability and real-time, and provides a feasible technical solution for a high proportion of new energy power systems.

[0138] Corresponding to the foregoing embodiment of the multi-agent collaborative optimization scheduling method based on the dynamic rules of the peak regulation market, the present application also provides an embodiment of a multi-agent collaborative optimization scheduling device based on the dynamic rules of the peak regulation market.

[0139] Referring to Figure 4 , the embodiment of the present application provides a multi-agent collaborative optimization scheduling device based on the dynamic rules of the peak regulation market, which comprises a memory and one or more processors, the memory stores executable codes, and the processor executes the executable codes to implement the multi-agent collaborative optimization scheduling method based on the dynamic rules of the peak regulation market.

[0140] The embodiment of the multi-agent collaborative optimization scheduling device based on the dynamic rules of the peak regulation market provided by the present application can be applied to any device with data processing capability, which can be a device or apparatus such as a computer. The device embodiment can be realized by software, or by hardware or a combination of software and hardware. Taking software realization as an example, as a logical device, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory for execution by the processor of the device with data processing capability. From the hardware level, as shown in Figure 4 , it is a hardware structure diagram of the device with data processing capability of the multi-agent collaborative optimization scheduling device based on the dynamic rules of the peak regulation market provided by the present application. In addition to the processor, memory, network interface and non-volatile memory shown in Figure 4 , the device with data processing capability in the embodiment usually includes other hardware according to the actual functions of the device with data processing capability, and details are not repeated.

[0141] The implementation process of the functions and roles of each unit in the above device is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0142] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part can be referred to the part of the method embodiment. The device embodiment described above is only illustrative, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the present application according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0143] The embodiment of the application also provides a computer readable storage medium, which stores a program, and the program is executed by a processor to realize the multi-agent collaborative optimization scheduling method based on the dynamic peak regulation market rule.

[0144] The computer readable storage medium can be an internal storage unit of any data processing capable device, such as a hard disk or a memory. The computer readable storage medium can also be an external storage device of any data processing capable device, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. Further, the computer readable storage medium can include both an internal storage unit and an external storage device of any data processing capable device. The computer readable storage medium is used to store the computer program and other programs and data required by the data processing capable device, and can also be used to temporarily store data that has been output or will be output.

[0145] The application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the multi-agent collaborative optimization scheduling method based on the dynamic peak regulation market rule.

[0146] The above embodiments are used to explain and illustrate the application, but not to limit the application. Any modifications and changes made to the application within the spirit and protection scope of the claims fall within the protection scope of the application.

Claims

1. A multi-agent cooperative optimization scheduling method based on dynamic rules formulated by a peak-shaving market, characterized in that, Including the following steps: S1. Based on the physical characteristics and adjustment flexibility of resources, resources in multi-agent systems are finely classified to establish a multi-dimensional classification system; the internal resources of the agents are divided into uncontrollable resources, semi-controllable resources, and controllable resources; each type of resource is assigned a dynamic label including peak-shaving capacity margin and response delay time, and the label value is updated in real time to construct a resource label matrix; real-time resource data is collected to construct a time-series feature vector, including photovoltaic power, wind turbine speed, and energy storage SOC data; S2 establishes a precise mathematical description model for each type of resource, quantifying its feasible region and cost characteristics in the peak-shaving market; the specific process is as follows: 1) For controllable resources, establish a new energy uncertainty model and use robust optimization to describe the resource output range; 2) Establish a dynamic constraint model for energy storage resources, comprehensively considering the relationship between depth of charge / discharge (DOD) and cycle life, as well as charge / discharge efficiency and capacity limitations; 3) For semi-controllable resources, establish a resource indirect response model, and establish the relationship between user power and electricity price based on user behavior and electricity demand patterns summarized from historical data; 4) For uncontrollable resources, historical data is used to train the prediction model. After obtaining the baseline power curve, a probability distribution model is used to describe the prediction uncertainty. S3 establishes a dynamic relationship between resource and peak-shaving market rules, and constructs an optimal power market rule model with multiple agents based on rule parameters. The rule parameters for establishing the dynamic relationship between resource and peak-shaving market rules include minimum duration and minimum capacity. The optimal power market rule model is a two-layer optimization model with agents participating in the power market, including a rule generation layer and an internal optimization layer. The dynamic rules are adjusted daily based on forecast data and historical operating data. S4, based on the mathematical description model constructed in step S2 and the optimal rule model of the electricity market constructed in step S3, constructs a two-layer collaborative optimization strategy, solves it, and outputs the optimal rule of the current peak-shaving market, realizing dynamic optimization of market rules and collaborative resource scheduling; the rule generation layer dynamically generates optimal market rule parameters based on an improved multi-objective particle swarm optimization algorithm, including minimum duration. and minimum capacity The details are as follows: Discrete velocity update formula: Discrete position update formula: The auxiliary function is defined as: in, Represents particles In dimensions Upper The velocity value of the next iteration, the initial velocity is... Interval random initialization; It is the inertia weight parameter, which serves as a scalar coefficient to control the intensity of the influence of historical velocity on the current velocity; and These are the cognitive coefficient and the social coefficient, respectively. and It is in the interval Independent random variables that are uniformly distributed on the upper bound. Represents particles In dimensions The individual's historical best position; Indicates the current group in dimension The guiding position on; Represents particles In dimensions Upper The current position in the next iteration; It is displacement The adjustment function; It is a boundary constraint function; The internal optimization layer includes each agent of new energy, energy storage and adjustable load independently solving the mixed integer programming model (MILP), responding to upper-layer rule constraints and feeding back economic indicators. The alternating direction multiplier method (ADMM) is used to realize the interaction of parameters between the upper and lower layers, ensuring global optimality and the efficiency of distributed computing.

2. The multi-agent cooperative optimization scheduling method based on dynamic rules of the peak-shaving market as described in claim 1, characterized in that, The rule generation layer is: Based on the global optimality rule parameters for power grid trading: in, The minimum duration for an agent to participate in the peak-shaving market. To minimize capacity; and to balance grid computational complexity, overall participation, and grid stability, design an objective function that maximizes the objective: in, Total participation capacity of internal agents participating in the electricity market the sum of Historical participation capacity benchmarks for internal agents participating in the electricity market; This refers to the demand satisfaction rate of the power grid, which is the ratio of the cleared winning bid capacity to the actual demand. , , Weighting coefficients for different objectives; The computational complexity of market clearing.

3. The multi-agent cooperative optimization scheduling method based on dynamic rules of the peak-shaving market as described in claim 2, characterized in that, The internal optimization layer is as follows: Each agent is modeled individually, and a mixed-integer programming model is constructed, embedding dynamic rule constraints: in, Indicates that the intelligent agent is in The sum of electricity revenue sold to the grid and electricity revenue sold to the load at any given time; It is the electricity price sold by the intelligent agent. It is the electricity price that the intelligent agent sells to the load; Indicates that the system is in The sum of the power generation cost and the energy storage charging and discharging cost at any given moment. The total power of internal new energy sources, To regulate energy storage costs, Let N be the energy storage power at time t, and N be the total number of resources participating in the peak-shaving market. The minimum duration for an agent to participate in the peak-shaving market. To achieve the minimum capacity, the agent maximizes its economic benefits within a given time period by accumulating the gains and costs at each moment and maximizing this accumulated value.

4. A multi-agent cooperative optimization scheduling device based on dynamic rules of a peak-shaving market, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that... When the processor executes the executable code, it implements a multi-agent collaborative optimization scheduling method based on dynamic rules of the peak-shaving market as described in any one of claims 1-3.

5. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements a multi-agent collaborative optimization scheduling method based on dynamic rules of the peak-shaving market as described in any one of claims 1-3.

Images