A method and system for optimizing power grid fluctuation based on air conditioning power prediction

Abstract

The present invention relates to a method and system for optimizing power grid fluctuations based on air-conditioning power prediction, comprising the following steps: S1: collecting historical power consumption data, including historical weather data and user power consumption behavior data, and calculating data on air-conditioning usage in the user area; S2: selecting key weather features that affect the use of air-conditioning, and building an air-conditioning power prediction model based on the calculated historical air-conditioning power to predict future air-conditioning power requirements; S3: optimizing power grid load based on the predicted future air-conditioning power requirements and the DQN algorithm model; S4: using a heuristic algorithm to optimize energy scheduling, maximizing the utilization of renewable energy and energy storage systems, and using user behavior analysis to provide personalized incentives to adjust peak power loads. The present invention effectively optimizes power fluctuations caused by the use of air-conditioning and improves the stability and efficiency of the power grid.

Description

Technical Field

[0001] The present invention relates to the field of power grid optimization, and in particular to a power grid fluctuation optimization method and system based on air conditioning power prediction. Background Art

[0002] Industrialization, urbanization, and digitalization have accelerated the rapid development of the national economy and the improvement of people's living standards. The penetration rate of air conditioners has increased year by year. As of 2021, the domestic air conditioner ownership per 100 households was 131.2 units, totaling 540 million units. Air conditioning power consumption has become one of the main factors for fluctuations in electricity consumption in summer and winter, and fluctuations cause interference with the electricity consumption curve.

[0003] In order to effectively cope with the increasing load demand of the power grid, especially in the case of high utilization of air conditioning equipment, grid fluctuation optimization has become a key area. Moreover, in complex power systems, the air conditioning calculation process may be affected by many factors and lead to inaccuracies. These inaccuracies may come from incomplete data, limitations of model selection, uncertainty of user behavior, and changes in the external environment. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a method and system for optimizing power grid fluctuations based on air conditioning power prediction, so as to effectively optimize the power fluctuations caused by the use of air conditioners and improve the stability and efficiency of the power grid.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for optimizing power grid fluctuation based on air conditioning power prediction includes the following steps:

[0007] S1: Collect historical power consumption data, including historical weather data and user power consumption behavior data, and calculate the data on air conditioning usage in the user area;

[0008] S2: Select key weather features that affect air conditioning use, and build an air conditioning power prediction model based on the measured historical air conditioning power consumption to predict future air conditioning power demand;

[0009] S3: Based on the predicted future air conditioning power demand, the grid load is optimized based on the DQN algorithm model;

[0010] S4: Use heuristic algorithms to optimize energy scheduling, maximize the use of renewable energy and energy storage system utilization, and use user behavior analysis to provide personalized incentives and adjust peak power loads.

[0011] Furthermore, the data on air conditioning usage in the user area is calculated as follows:

[0012] Construct a regional air conditioning power measurement model to automatically and intelligently identify typical days for air conditioning power measurement: collect power data for each period; weather data: including temperature, humidity, light, wind speed; date information: weekdays, weekends, holidays; event information: special events that affect power consumption in the region;

[0013] Use interpolation to fill missing data and statistical methods to detect and correct abnormal data points;

[0014] Extract seasonal features, time features, and event marking features;

[0015] Use the K-means clustering algorithm to cluster historical data to identify typical date types;

[0016] The power curve return model based on the power natural growth rate adjustment factor eliminates the abnormal fluctuations in air-conditioning power caused by special events, and realizes quasi-real-time online calculation of air-conditioning power in all regions and industries at all times. Specifically:

[0017] Calculate the trend of air conditioning power consumption based on long time series data and use regression method to determine the growth rate;

[0018] Through event recognition, the abnormal increments are separated, and the relationship between the air-conditioning power consumption and the identified features is modeled using a machine learning regression model to calculate the stable air-conditioning power consumption after elimination;

[0019] The air conditioning power consumption after eliminating geographical and environmental influences is adjusted to the basic power consumption curve, which is generated by adjusting factors;

[0020] Based on the adjusted data, the true power consumption is calculated to express the expected normal air-conditioning power consumption curve.

[0021] Furthermore, S2 is as follows:

[0022] The most relevant weather variables are selected, including temperature and humidity, and binary encoding is used to represent weekdays, weekends, and holidays. A training dataset is constructed based on the measured air conditioning power consumption.

[0023] Construct an air conditioning power prediction model based on the SARIMA model;

[0024] Based on the trained model, use cross-validation methods (such as K-fold or time series-specific CV methods) to verify model stability; and check the model residuals to ensure that they meet the white noise assumption and have no obvious patterns or trends.

[0025] Furthermore, an air conditioning power prediction model is constructed based on the SARIMA model, as follows:

[0026] Obtain historical air conditioning power consumption data based on the data of air conditioning usage in the user area;

[0027] By drawing a time series graph, determine whether preliminary differencing is needed to eliminate seasonal trends;

[0028] Temperature and humidity are taken as exogenous variables X to capture their impact on electricity usage;

[0029] Use ACF and PACF plots to determine possible values ​​of non-seasonal (p, q) and seasonal (P, Q); where p is the order of the autoregressive term, q is the order of the moving average term, P is seasonal autoregressive, and Q is seasonal moving average;

[0030] Based on the SARIMAX model, exogenous variables are introduced;

[0031] ;

[0032] in, represents the seasonal autoregressive polynomial, represents the seasonal autoregressive part, B is the lag operator, and s represents the seasonal period; Represents a non-seasonal autoregressive polynomial, which is used to represent the autocorrelation between series; represents the non-seasonal difference term; d represents the number of differences; represents the seasonal difference term; D represents the number of seasonal differences; Y t is the target variable, i.e., the power consumption of air conditioner at time t; is the seasonal moving average polynomial, which represents the influence of seasonal random errors; is the non-seasonal moving average polynomial, which represents the influence of non-seasonal random errors; is the error term; X t Exogenous variable matrix; β is the coefficient of the exogenous variable;

[0033] AIC and BIC were used to select the model order, and the optimal seasonal and non-seasonal orders were determined through cross-validation and grid search; the normality of the residuals was verified by the Shapiro-Wilk test.

[0034] Furthermore, AIC and BIC were used to select the model order, and the optimal seasonal and non-seasonal orders were determined through cross-validation and grid search. The normality of the residuals was verified by the Shapiro-Wilk test, as follows:

[0035] Predefine possible p,d,q,P,D,Q grids; fit each parameter combination using the SARIMAX model;

[0036] Calculate AIC and BIC;

[0037] ;

[0038] ;

[0039] Where k is the number of parameters in the model; is the maximum likelihood estimate of the model; n is the sample size;

[0040] Select the parameter combination with the lowest AIC / BIC;

[0041] Perform time series cross validation under the selected parameter combination to ensure the stability of the model on different training sets;

[0042] The normality of the residuals was verified using the Shapiro-Wilk test to ensure that there was no deviation from the assumptions:

[0043]

[0044] in, x (i) is the sorted sample value; a i is the calculated coefficient combination; is the sample mean;

[0045] W indicates that the residuals are close to a normal distribution.

[0046] Furthermore, the DQN algorithm model is specifically:

[0047] The state space S includes the actual measured current load L of the power grid t , Future air conditioning power demand D obtained based on the prediction model t , the current charging state or remaining energy storage capacity Et of the energy storage system;

[0048] The state vector is represented as:

[0049] S t =[L t ,D t ,E t ] ;

[0050] Action space A, including: adjusting the power generation P g ; Enable backup power generation equipment P b ; Energy storage system charging and discharging (P ch ,Pdis ) ;

[0051] Actions are represented as a vector:

[0052] A t =[P g ,P b ,Pch,P dis ] ;

[0053] The reward function is defined as:

[0054]

[0055] C op Operating costs, including fuel costs, backup generation costs, and energy storage costs; Represents the penalty term for violation of the constraint condition, λ is the penalty coefficient;

[0056] The Q network model uses a fully connected neural network model, which includes an input layer, several hidden layers, and an output layer;

[0057] Network input: state space vector S t ;

[0058] Network output: For each possible action, output the corresponding Q value.

[0059] Furthermore, the power grid load is optimized based on the DQN algorithm model, as follows:

[0060] Initialize the experience buffer H, randomly initialize Q network parameters θ and target network parameters θ′;

[0061] From the current state S t Start by selecting Action A t Get the next state S t+1 and return R t , and store it in H;

[0062] Uniformly sample a batch of historical records (S, A, R, S′) from H;

[0063] Among them, R: is the reward, which means the reward obtained by the agent after taking action A from state S, and S′ is the next state;

[0064] Update the loss function using the least squares method:

[0065] ;

[0066] in, is the discount factor; Indicates the calculation of expected value; Indicates that the current Q value network estimates that it is in state S t Next take action A t Q value; a represents any action;

[0067] Every several steps, the Q network parameter θ replaces the target network parameter θ′;

[0068] Update the environment to the new state, in S t+1 Repeat the above strategy.

[0069] Furthermore, S4 is specifically:

[0070] The optimization goal is to maximize the use of renewable energy and the utilization of energy storage systems while meeting the electricity needs of users:

[0071] ;

[0072] Among them, P renew,t is the renewable energy power usage at time t; E util,t is the utilization rate of the energy storage system at time t; C t is the total operating cost of the system; α, is the trade-off coefficient; Z is the utilization of renewable energy and the utilization rate of the energy storage system; T represents the total number of time periods;

[0073] Constraints include

[0074] Energy balance constraints:

[0075] ;

[0076] Among them, P load,t The power load demand at time t; P grid,t Indicates the power obtained from the grid; E discharge,t and E charge,t They represent the discharging and charging power of the energy storage system at time t respectively;

[0077] Energy storage capacity constraints:

[0078] E min ≤E t ≤E max ;

[0079] Among them, E min and E max Respectively represent the minimum and maximum capacity limits of the energy storage system;

[0080] Renewable energy output limits:

[0081] 0≤P renew,t ≤P renew, max ;

[0082] Among them, P renew, max Represents the maximum output of renewable energy

[0083] Based on the GA algorithm, find the optimal strategy:

[0084] Randomly generate a set of solutions, each solution consists of a set of decision variables;

[0085] Evaluate the fitness of each solution and calculate its utility based on the objective function;

[0086] The parent individuals are selected according to the fitness value by the roulette wheel selection method;

[0087] Through the crossover operator, two parent individuals are mixed to generate offspring;

[0088] Mutate offspring individuals and introduce randomness to maintain population diversity;

[0089] Repeat selection, crossover, and mutation until the termination condition is reached, such as the maximum number of iterations or convergence to a satisfactory solution.

[0090] Furthermore, user behavior analysis is used to provide personalized incentives and adjust peak power loads, as follows:

[0091] Based on the user electricity consumption behavior data, the users are divided into different behavior pattern groups according to the DBSCAN clustering algorithm;

[0092] Implement personalized incentives based on user group behavior models and monitor user responses and grid loads in real time;

[0093] Dynamically adjust incentives and pricing strategies as user behavior changes and feedback emerges.

[0094] A power grid fluctuation optimization system based on air conditioning power prediction includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the power grid fluctuation optimization method based on air conditioning power prediction as described above.

[0095] The present invention has the following beneficial effects:

[0096] 1. The present invention effectively optimizes the power fluctuations caused by the use of air conditioners and improves the stability and efficiency of the power grid;

[0097] 2. The present invention can realize accurate measurement of regional air conditioning power consumption, integrate key factors affecting air conditioning use, and use the constructed prediction model to more accurately predict future air conditioning power demand, providing reliable basic data support for power grid optimization decision-making and improving the accuracy of overall decision-making;

[0098] 3. The present invention integrates multi-level optimization strategies to enable the power grid to make better decisions on resource allocation, reduce operating costs, improve economic benefits, and optimize peak loads. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0100] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0101] refer to Figure 1 In this embodiment, a method for optimizing power grid fluctuation based on air conditioning power prediction is provided, comprising the following steps:

[0102] S1: Collect historical power consumption data, including historical weather data and user power consumption behavior data, and calculate the data on air conditioning usage in the user area;

[0103] S2: Select key weather features that affect air conditioning use, and build an air conditioning power prediction model based on the measured historical air conditioning power consumption to predict future air conditioning power demand;

[0104] S3: Based on the predicted future air conditioning power demand, the grid load is optimized based on the DQN algorithm model;

[0105] S4: Use heuristic algorithms to optimize energy scheduling, maximize the use of renewable energy and energy storage system utilization, and use user behavior analysis to provide personalized incentives and adjust peak power loads.

[0106] In this embodiment, the data of air conditioning usage in the user area is calculated as follows:

[0107] Build a regional air conditioning power measurement model to automatically and intelligently identify typical days for air conditioning power measurement: collect power data for each period; weather data: including temperature, humidity, light, wind speed; date information: weekdays, weekends, holidays; event information: special events in the region that affect power consumption, such as large-scale events and festivals;

[0108] Use interpolation to fill missing data and statistical methods to detect and correct abnormal data points;

[0109] Extract seasonal features (such as spring, summer, autumn, and winter), time features (time period, morning and evening peaks), and event marking features (such as holiday markings);

[0110] Use the K-means clustering algorithm to cluster historical data to identify typical date types (weekdays, weekends, special event days, etc.);

[0111] The power curve return model based on the power natural growth rate adjustment factor eliminates the abnormal fluctuations in air-conditioning power caused by special events, and realizes quasi-real-time online calculation of air-conditioning power in all regions and industries at all times. Specifically:

[0112] Calculate the trend of air conditioning power consumption based on long time series data and use regression method to determine the growth rate;

[0113] Through event recognition, the abnormal increments are separated, and the relationship between the air-conditioning power consumption and the identified features is modeled using a machine learning regression model to calculate the stable air-conditioning power consumption after elimination;

[0114] The air conditioning power consumption after eliminating geographical and environmental influences is adjusted to the basic power consumption curve, which is generated by adjusting factors;

[0115] Based on the adjusted data, the true power consumption is calculated to express the expected normal air-conditioning power consumption curve.

[0116] In this embodiment, S2 is specifically as follows:

[0117] The most relevant weather variables are selected, including temperature and humidity, and binary encoding is used to represent weekdays, weekends, and holidays. A training dataset is constructed based on the measured air conditioning power consumption.

[0118] Construct an air conditioning power prediction model based on the SARIMA model;

[0119] Based on the trained model, use cross-validation methods (such as K-fold or time series-specific CV methods) to verify model stability; and check the model residuals to ensure that they meet the white noise assumption and have no obvious patterns or trends.

[0120] In this embodiment, an air conditioning power prediction model is constructed based on the SARIMA model, as follows:

[0121] Obtain historical air conditioning power consumption data based on the data of air conditioning usage in the user area;

[0122] By drawing a time series graph, determine whether preliminary differencing is needed to eliminate seasonal trends;

[0123] Temperature and humidity are taken as exogenous variables X to capture their impact on electricity usage;

[0124] Use ACF and PACF plots to determine possible values ​​of non-seasonal (p, q) and seasonal (P, Q); where p is the order of the autoregressive term, q is the order of the moving average term, P is seasonal autoregressive, and Q is seasonal moving average;

[0125] Based on the SARIMAX model, exogenous variables are introduced;

[0126] ;

[0127] in, represents the seasonal autoregressive polynomial, represents the seasonal autoregressive part, B is the lag operator, and s represents the seasonal period; Represents a non-seasonal autoregressive polynomial, which is used to represent the autocorrelation between series; represents the non-seasonal difference term; d represents the number of differences; represents the seasonal difference term; D represents the number of seasonal differences; Y t is the target variable, i.e., the power consumption of air conditioner at time t; is the seasonal moving average polynomial, which represents the influence of seasonal random errors; is the non-seasonal moving average polynomial, which represents the influence of non-seasonal random errors; is the error term; X t Exogenous variable matrix; β is the coefficient of the exogenous variable;

[0128] AIC and BIC were used to select the model order, and the optimal seasonal and non-seasonal orders were determined through cross-validation and grid search; the normality of the residuals was verified by the Shapiro-Wilk test.

[0129] In this embodiment, AIC and BIC are used to select the model order, and the optimal seasonal and non-seasonal orders are determined through cross-validation and grid search; the residual normality is verified by the Shapiro-Wilk test, as follows:

[0130] Predefine possible p,d,q,P,D,Q grids; fit each parameter combination using the SARIMAX model;

[0131] Calculate AIC and BIC;

[0132] ;

[0133] ;

[0134] Where k is the number of parameters in the model; is the maximum likelihood estimate of the model; n is the sample size;

[0135] Select the parameter combination with the lowest AIC / BIC;

[0136] Perform time series cross validation under the selected parameter combination to ensure the stability of the model on different training sets;

[0137] The normality of the residuals was verified using the Shapiro-Wilk test to ensure that there was no deviation from the assumptions:

[0138]

[0139] in, x (i) is the sorted sample value; a i is the calculated coefficient combination; is the sample mean;

[0140] W indicates that the residuals are close to a normal distribution.

[0141] In this embodiment, the DQN algorithm model is specifically:

[0142] The state space S includes the actual measured current load L of the power grid t , Future air conditioning power demand D obtained based on the prediction model t , the current charging state or remaining energy storage capacity Et of the energy storage system;

[0143] The state vector is represented as:

[0144] S t =[L t ,D t ,E t ] ;

[0145] Action space A, including: adjusting the power generation P g ; Enable backup power generation equipment P b ; Energy storage system charging and discharging (P ch ,P dis ) ;

[0146] Actions are represented as a vector:

[0147] A t =[P g ,Pb ,Pch,P dis ] ;

[0148] The reward function is defined as:

[0149]

[0150] C op Operating costs, including fuel costs, backup generation costs, and energy storage costs; It represents the penalty term for violation of constraints (such as maximum load of generator, energy storage capacity, etc.), and λ is the penalty coefficient;

[0151] The Q network model uses a fully connected neural network model, which includes an input layer, several hidden layers (such as ReLU activation), and an output layer;

[0152] Network input: state space vector S t ;

[0153] Network output: For each possible action, output the corresponding Q value.

[0154] In this embodiment, the power grid load optimization is performed based on the DQN algorithm model, as follows:

[0155] Initialize the experience buffer H, randomly initialize Q network parameters θ and target network parameters θ′;

[0156] From the current state S t Start by selecting Action A t Get the next state S t+1 and return R t , and store it in H;

[0157] Uniformly sample a batch of historical records (S, A, R, S′) from H;

[0158] Among them, R is the reward, which means the reward obtained by the agent after taking action A from state S, and S′ is the next state;

[0159] Update the loss function using the least squares method:

[0160] ;

[0161] in, is the discount factor; Indicates the calculation of expected value; Indicates that the current Q value network estimates that it is in state S t Next take action A t Q value; a represents any action;

[0162] Every several steps, the Q network parameter θ replaces the target network parameter θ′;

[0163] Update the environment to the new state, in S t+1 Repeat the above strategy.

[0164] In this embodiment, S4 is specifically:

[0165] The optimization goal is to maximize the use of renewable energy and the utilization of energy storage systems while meeting the electricity needs of users:

[0166] ;

[0167] Among them, P renew,t is the renewable energy power usage at time t; E util,t is the utilization rate of the energy storage system at time t; C t is the total operating cost of the system; α, is the trade-off coefficient; Z is the utilization of renewable energy and the utilization rate of the energy storage system; T represents the total number of time periods;

[0168] Constraints include

[0169] Energy balance constraints:

[0170] ;

[0171] Among them, P load,t The power load demand at time t; P grid,t Indicates the power obtained from the grid; E discharge,t and E charge,t They represent the discharging and charging power of the energy storage system at time t respectively;

[0172] Energy storage capacity constraints:

[0173] E min ≤E t ≤E max ;

[0174] Among them, E min and E max Respectively represent the minimum and maximum capacity limits of the energy storage system;

[0175] Renewable energy output limits:

[0176] 0≤P renew,t ≤P renew, max ;

[0177] P renew, max Represents the maximum value of renewable energy output;

[0178] Based on the GA algorithm, find the optimal strategy:

[0179] Randomly generate a set of solutions, each solution consists of a set of decision variables;

[0180] Evaluate the fitness of each solution and calculate its utility based on the objective function;

[0181] The parent individuals are selected according to the fitness value by the roulette wheel selection method;

[0182] Through the crossover operator (such as single-point crossover, double-point crossover), two parent individuals are mixed to generate offspring;

[0183] Mutate offspring individuals and introduce randomness to maintain population diversity;

[0184] Repeat selection, crossover, and mutation until the termination condition is reached, such as the maximum number of iterations or convergence to a satisfactory solution.

[0185] In this embodiment, user behavior analysis is used to provide personalized incentives and adjust peak power loads, as follows:

[0186] Based on the user electricity consumption behavior data, the DBSCAN clustering algorithm is used to divide users into different behavior pattern groups, such as users with high peak electricity consumption and price-sensitive users;

[0187] Implement personalized incentives based on the behavioral model of user groups, monitor user responses and grid loads in real time, and set different electricity price strategies for different user groups. For price-sensitive users, higher electricity prices can be set during peak hours to encourage them to reduce electricity consumption. Use a point system to give points to users based on the amount of electricity they reduce during peak hours, which can be exchanged for electricity discounts or other services.

[0188] Dynamically adjust incentives and pricing strategies as user behavior changes and feedback emerges.

[0189] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0190] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0191] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0192] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0193] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for optimizing power grid fluctuations based on air conditioning power prediction, characterized in that: The following steps are involved: S1: Collect historical power consumption data, including historical weather data and user power consumption behavior data, and calculate the data on air conditioning usage in the user area; S2: Select key weather features that affect air conditioning use, and build an air conditioning power prediction model based on the measured historical air conditioning power consumption to predict future air conditioning power demand; S3: Based on the predicted future air conditioning power demand, the grid load is optimized based on the DQN algorithm model; S4: Use heuristic algorithms to optimize energy dispatch, maximize the use of renewable energy and energy storage systems, and use user behavior analysis to provide personalized incentives and adjust peak power loads; The data for calculating the usage of air conditioners in the user area are as follows: Construct a regional air conditioning power measurement model to automatically and intelligently identify typical days for air conditioning power measurement: collect power data for each period; weather data: including temperature, humidity, light, wind speed; date information: weekdays, weekends, holidays; event information: special events that affect power consumption in the region; Use interpolation to fill missing data and statistical methods to detect and correct abnormal data points; Extract seasonal features, time features, and event marking features; Use the K-means clustering algorithm to cluster historical data to identify typical date types; The power curve return model based on the power natural growth rate adjustment factor eliminates the abnormal fluctuations in air-conditioning power caused by special events, and realizes quasi-real-time online calculation of air-conditioning power in all regions and industries at all times. Specifically: Calculate the trend of air conditioning power consumption based on long time series data and use regression method to determine the growth rate; Through event recognition, the abnormal increments are separated, and the relationship between the air-conditioning power consumption and the identified features is modeled using a machine learning regression model to calculate the stable air-conditioning power consumption after elimination; The air conditioning power consumption after eliminating geographical and environmental influences is adjusted to the basic power consumption curve, which is generated by adjusting factors; Based on the adjusted data, the true power is calculated to express the expected normal air conditioning power curve; The DQN algorithm model is specifically: The state space S includes the actual measured current load L of the power grid t , Future air conditioning power demand D obtained based on the prediction model t , the current charging state or remaining energy storage capacity Et of the energy storage system; The state vector is represented as: S t =[L t ,D t ,E t ] ； Action space A, including: adjusting the power generation P g ; Enable backup power generation equipment P b ;Charging and discharging of energy storage system (P ch , P dis ) ; Actions are represented as a vector: A t =[P g ,P b ,Pch,P dis ] ； The reward function is defined as: ； C op Operating costs, including fuel costs, backup generation costs, and energy storage costs; Represents the penalty term for violation of the constraint condition, λ is the penalty coefficient; The Q network model uses a fully connected neural network model, which includes an input layer, several hidden layers, and an output layer; Network input: state space vector S t ; Network output: For each possible action, output the corresponding Q value; The power grid load optimization based on the DQN algorithm model is as follows: Initialize the experience buffer H, randomly initialize Q network parameters θ and target network parameters θ′; From the current state S t Start by selecting Action A t Get the next state S t+1 and return R t , and store it in H; Uniformly sample a batch of historical records (S, A, R, S′) from H; Among them, R is the reward, which means the reward obtained by the agent after taking action A from state S, and S′ is the next state; Update the loss function using the least squares method: ； in, is the discount factor; Indicates the calculation of expected value; Indicates that the current Q value network estimates that it is in state S t Next take action A t Q value; a represents any action; Every several steps, the Q network parameter θ replaces the target network parameter θ′; Update the environment to the new state, in S t+1 Repeat the above strategy.

2. A method for optimizing power grid fluctuations based on air conditioning power prediction according to claim 1, characterized in that: The S2 is specifically as follows: The most relevant weather variables are selected, including temperature and humidity, and binary encoding is used to represent weekdays, weekends, and holidays. A training dataset is constructed based on the measured air conditioning power consumption. Construct an air conditioning power prediction model based on the SARIMA model; Based on the trained model, the cross-validation method is used to verify the model stability; The model residuals were checked to ensure that they met the white noise assumption and had no obvious patterns or trends.

3. The method for optimizing power grid fluctuation based on air conditioning power prediction according to claim 2 is characterized in that: The air conditioning power prediction model based on the SARIMA model is constructed as follows: Obtain historical air conditioning power consumption data based on the data of air conditioning usage in the user area; By drawing a time series graph, determine whether preliminary differencing is needed to eliminate seasonal trends; Temperature and humidity are taken as exogenous variables X to capture their impact on electricity usage; Use ACF and PACF plots to determine possible values ​​of non-seasonal (p, q) and seasonal (P, Q); where p is the order of the autoregressive term, q is the order of the moving average term, P is seasonal autoregressive, and Q is seasonal moving average; Based on the SARIMAX model, exogenous variables are introduced; ； in, represents the seasonal autoregressive polynomial, represents the seasonal autoregressive part, B is the lag operator, and s represents the seasonal period; Represents a non-seasonal autoregressive polynomial, which is used to represent the autocorrelation between series; represents the non-seasonal difference term; d represents the number of differences; represents the seasonal difference term; D represents the number of seasonal differences; Y t is the target variable, i.e., the power consumption of air conditioner at time t; is the seasonal moving average polynomial, which represents the influence of seasonal random errors; is the non-seasonal moving average polynomial, which represents the influence of non-seasonal random errors; is the error term; X t Exogenous variable matrix; β is the coefficient of the exogenous variable; AIC and BIC were used to select the model order, and the optimal seasonal and non-seasonal orders were determined through cross-validation and grid search; the normality of the residuals was verified by the Shapiro-Wilk test.

4. The method for optimizing power grid fluctuation based on air conditioning power prediction according to claim 3 is characterized in that: The AIC and BIC are used to select the model order, and the optimal seasonal and non-seasonal orders are determined through cross-validation and grid search; the residual normality is verified by the Shapiro-Wilk test, as follows: Predefine possible p,d,q,P,D,Q grids; fit each parameter combination using the SARIMAX model; Calculate AIC and BIC; ； ； Where k is the number of parameters in the model; is the maximum likelihood estimate of the model; n is the sample size; Select the parameter combination with the lowest AIC / BIC; Perform time series cross validation under the selected parameter combination to ensure the stability of the model on different training sets; The normality of the residuals was verified using the Shapiro-Wilk test to ensure that there was no deviation from the assumptions: ； in, x (i) is the sorted sample value; a i is the calculated coefficient combination; is the sample mean; W indicates that the residuals are close to a normal distribution.

5. The method for optimizing power grid fluctuation based on air conditioning power prediction according to claim 1, characterized in that: The S4 is specifically: The optimization goal is to maximize the use of renewable energy and the utilization of energy storage systems while meeting the electricity needs of users: ； Among them, P renew,t is the renewable energy power usage at time t; E util,t is the utilization rate of the energy storage system at time t; C t is the total operating cost of the system; α, is the trade-off coefficient; Z is the utilization of renewable energy and the utilization rate of the energy storage system; T represents the total number of time periods; Constraints include Energy balance constraints: ； Among them, P load,t The power load demand at time t; P grid,t Indicates the power obtained from the grid; E discharge,t and E charge,t They represent the discharging and charging power of the energy storage system at time t respectively; Energy storage capacity constraints: AND min ≤E t ≤E max ; Among them, E min and E max Respectively represent the minimum and maximum capacity limits of the energy storage system; Renewable energy output limits: 0≤P renew,t ≤P renew, max ; Among them, P renew, max Represents the maximum value of renewable energy output; Based on the GA algorithm, find the optimal strategy: Randomly generate a set of solutions, each solution consists of a set of decision variables; Evaluate the fitness of each solution and calculate its utility based on the objective function; The parent individuals are selected according to the fitness value by the roulette wheel selection method; Through the crossover operator, two parent individuals are mixed to generate offspring; Mutate offspring individuals and introduce randomness to maintain population diversity; Selection, crossover, and mutation are repeated until a termination condition is reached, which includes a maximum number of iterations or convergence to a satisfactory solution.

6. A method for optimizing power grid fluctuations based on air conditioning power prediction according to claim 5, characterized in that: The user behavior analysis is used to provide personalized incentives and adjust peak power loads, as follows: Based on the user electricity consumption behavior data, the users are divided into different behavior pattern groups according to the DBSCAN clustering algorithm; Implement personalized incentives based on user group behavior models and monitor user responses and grid loads in real time; Dynamically adjust incentives and pricing strategies as user behavior changes and feedback emerges.

7. A power grid fluctuation optimization system based on air conditioning power prediction, characterized in that: It includes a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the power grid fluctuation optimization method based on air conditioning power prediction as described in any one of claims 1 to 6.

Images