Energy storage strategy dynamic optimization system and method based on big data

Abstract

A big data-based dynamic optimization system and method for energy storage strategies includes: constructing three types of discrete optimization objectives; performing correlation feature analysis on multi-source data in a sample database to obtain correlation coefficients between discrete feature variables and discrete optimization objectives; filtering discrete feature variables based on correlation coefficients and retaining those that meet the requirements; constructing a multi-source feature prediction model to obtain a sequence of predicted discrete feature variables for the next acquisition cycle; and using this sequence to obtain energy storage strategy analysis based on the energy storage side's charging and discharging power constraints, SOC constraints, objective functions of the three types of discrete optimization objectives, and dynamic weights to obtain the optimal energy storage strategy for the next acquisition cycle. This invention effectively solves the technical pain points of traditional energy storage scheduling methods, such as strong subjectivity in feature selection, insufficient prediction accuracy, single strategy optimization objective, and response lag.

Description

Technical Field

[0001] This invention relates to the field of power system energy storage control technology, specifically to a dynamic optimization system and method for energy storage strategies based on big data. Background Technology

[0002] As a key device for smoothing renewable energy fluctuations, peak shaving and valley filling, and improving grid flexibility, the rationality of grid scheduling strategies directly affects grid operating efficiency and the economics of energy storage systems. Existing technologies, in the optimization of energy storage strategies, generally suffer from several problems: reliance on manual experience in feature selection leads to strong subjectivity; key influencing factors are easily overlooked; and redundant features are difficult to remove, resulting in low prediction accuracy and poor computational efficiency in subsequent models. Furthermore, traditional methods often rely on historical data for static planning, lacking forward-looking and accurate predictions of the grid's "source-load-storage" operating status. This results in low adaptability of the formulated energy storage strategies to actual operating conditions, making it difficult to cope with the fluctuations in renewable energy output on the power generation side and the randomness of demand on the load side. In addition, existing optimization schemes often focus on a single objective, using fixed weights for multi-objective trade-offs. They cannot dynamically adjust the priorities of economics, renewable energy absorption rate, and energy storage lifespan based on the real-time operating status of the grid, easily leading to imbalances such as "emphasizing benefits over lifespan" or "emphasizing absorption over economics." Ultimately, this results in low overall efficiency of the energy storage system, failing to meet the requirements for safe and efficient grid operation under high renewable energy penetration. Summary of the Invention

[0003] To address the aforementioned technical problems, the present invention aims to provide a dynamic optimization method for energy storage strategies based on big data, comprising the following steps: Step s1: Construct three types of discrete optimization objectives, perform correlation feature analysis on multi-source data in the sample database, obtain the correlation coefficient between discrete feature variables and discrete optimization objectives in multi-source data, filter discrete feature variables based on correlation coefficients, and retain discrete feature variables that meet the requirements; Step s2: Construct a multi-source feature prediction model. Extract discrete feature variables from the multi-source data of the current collection period and input them into the multi-source feature prediction model to obtain the discrete feature prediction variable sequence for the next collection period. Step s3: Based on the discrete feature prediction variable sequence, obtain the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights of the three types of discrete optimization objectives on the energy storage side. Based on the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights, perform energy storage strategy analysis to obtain the optimal energy storage strategy for the next acquisition cycle. Step s4: Monitor the discrete characteristic variable sequence of the current acquisition period in real time, and determine whether to conduct energy storage strategy analysis based on the monitoring results.

[0004] Furthermore, a cloud-based monitoring center is constructed, and monitoring terminals are deployed on the power supply-side entities, load-side entities, and energy storage-side entities within the target area. The cloud-based monitoring center communicates with several monitoring terminals in the target area, receives multi-source data collected by each monitoring terminal, marks the collection time, and sets the collection cycle. A sample database is set up within the cloud-based monitoring center to store multi-source data from several historical collection cycles.

[0005] Furthermore, the process of constructing three types of discrete optimization objectives and performing correlation feature analysis on multi-source data in the sample database to obtain the correlation coefficients between discrete feature variables and discrete optimization objectives in the multi-source data includes: Several historical collection periods of multi-source data were extracted from the sample database as samples. All continuous variables in the multi-source data of each sample were extracted and three types of continuous target values ​​were constructed. All continuous variables and continuous target values ​​were standardized and preprocessed. For the range of values ​​of the standardized preprocessed continuous variable, select a threshold point to divide it into k discrete intervals, and map the sample value of each standardized preprocessed continuous variable to the corresponding discrete interval to generate discrete feature variables; Similarly, within the range of values ​​of the continuous target value after standardization and preprocessing, a threshold point is selected to divide the range into z discrete intervals. The sample value of each continuous target value after standardization and preprocessing is mapped to the corresponding discrete interval to generate a discrete optimization target. The number of samples falling into each discrete interval of the discrete feature variable, the number of samples falling into each discrete interval of the discrete optimization objective, and the number of samples falling into both the discrete intervals corresponding to the discrete feature variable and the discrete interval corresponding to the discrete optimization objective are statistically analyzed. Probability normalization is then performed on the number of samples falling into each discrete interval of the discrete feature variable, the number of samples falling into each discrete interval of the discrete optimization objective, and the number of samples falling into both the discrete intervals corresponding to the discrete feature variable and the discrete optimization objective. The marginal distribution coefficients of the discrete feature variable, the marginal distribution coefficients of the discrete optimization objective, and the joint distribution coefficients between the discrete feature variable and the discrete optimization objective are then obtained. The correlation coefficient between the discrete feature variable and the discrete optimization objective is obtained based on the marginal distribution coefficients and the joint distribution coefficients.

[0006] Furthermore, the process of selecting discrete feature variables based on correlation coefficients includes: A preset correlation coefficient threshold is set. The correlation coefficient between the discrete feature variable and each discrete optimization objective is compared with the correlation coefficient threshold. If the correlation coefficient between the discrete feature variable and each discrete optimization objective is less than the correlation coefficient threshold, the discrete feature variable is removed. If the correlation coefficient between the discrete feature variable and a discrete optimization objective is greater than or equal to the correlation coefficient threshold, the discrete feature variable is retained.

[0007] Furthermore, the process of constructing a multi-source feature prediction model includes: Extract multi-source data from power supply-side entities, load-side entities, and energy storage-side entities within several historical acquisition periods. Extract discrete feature variables from the multi-source data as training samples to construct a multi-source feature prediction model. Use the training samples to train the multi-source feature prediction model to obtain the trained multi-source feature prediction model. The discrete feature variables of the power supply-side entity, load-side entity, and energy storage-side entity in the current acquisition cycle are input into the multi-source feature prediction model. Based on the multi-source feature prediction model, the discrete feature prediction variable sequence of the power supply-side entity, load-side entity, and energy storage-side entity in the next acquisition cycle is output.

[0008] Furthermore, the process of obtaining the charging and discharging power constraints on the energy storage side includes: Extract the power output sequence and load demand sequence for the next data acquisition cycle from the discrete characteristic prediction variable sequences of the power supply-side entity, load-side entity, and energy storage-side entity in the next data acquisition cycle. The power supply and demand difference between the power source and the load side is obtained in the next acquisition cycle based on the power output sequence of the power source and the power demand sequence of the load side. The charging and discharging power constraints of the energy storage side are obtained based on the power supply and demand difference.

[0009] Furthermore, the process of conducting energy storage strategy analysis to obtain the optimal energy storage strategy includes: Objective functions for three types of discrete optimization objectives are constructed based on the discrete characteristic prediction variable sequences of the power supply entity, load entity, and energy storage entity in the next acquisition cycle. Several energy storage strategies are randomly generated based on the charging and discharging power constraints and SOC constraints of the energy storage side, with each energy storage strategy corresponding to a charging and discharging power sequence. The dynamic weights of the objective functions for the three types of discrete optimization objectives are obtained. A fitness function is constructed based on the objective functions and dynamic weights of the three types of discrete optimization objectives. Chromosome encoding and population initialization are performed on several energy storage strategies to generate an initial population. The optimal energy storage strategy is obtained through a multi-objective genetic algorithm based on the initial population and the fitness function.

[0010] Furthermore, the process of obtaining the dynamic weights of the objective functions for the three types of discrete optimization objectives includes: The correlation coefficients between discrete feature variables and various discrete optimization objectives are compared with the correlation coefficient thresholds. Discrete feature variables with correlation coefficients greater than the correlation coefficient thresholds are marked as correlation features of discrete optimization objectives. Evaluation indices are constructed based on the correlation characteristics of three types of discrete optimization objectives. Fuzzy sub-intervals of the evaluation indices and the power grid state level corresponding to each fuzzy sub-interval are set. The specific values ​​of the evaluation indices are obtained by predicting the variable sequence based on the discrete characteristics of the power source entity, load entity, and energy storage entity in the next acquisition cycle. The membership matrix of the specific values ​​of the evaluation indices to different power grid state levels is obtained through fuzzy comprehensive evaluation. The power grid state level of the evaluation index is obtained based on the membership matrix. The weight adjustment coefficients corresponding to different power grid state levels are preset. The initial weights of the objective functions of the three types of discrete optimization objectives are adjusted according to the weight adjustment coefficients to obtain the dynamic weights of the objective functions of the three types of discrete optimization objectives.

[0011] Furthermore, the process of real-time monitoring of the discrete characteristic variable sequence of the current acquisition period and determining whether to conduct energy storage strategy analysis based on the monitoring results includes: The discrete feature prediction variable sequence of the current acquisition period is used as the judgment standard to obtain the discrete feature variable sequence of the current acquisition period. The discrete feature variable sequence of the current acquisition period is compared with the judgment standard to obtain the discrete feature deviation of each time period. A preset discrete feature deviation threshold is set, and the discrete feature deviation is compared with the discrete feature deviation threshold. If the discrete feature deviation is greater than the discrete feature deviation threshold, the discrete feature variable sequence of the current acquisition period is input into the multi-source feature prediction model. Based on the multi-source feature prediction model, the discrete feature prediction variable sequence of the power supply entity, load entity, and energy storage entity in the remaining time period of the current acquisition period is output. Based on the discrete feature prediction variable sequence of the remaining time period of the current acquisition period, energy storage strategy analysis is performed, and the optimal energy storage strategy for the remaining time period of the current acquisition period is output.

[0012] The energy storage strategy dynamic optimization system based on big data includes a cloud monitoring center, which is connected to a data analysis module, a data prediction module, a data processing module, and a real-time monitoring module. The data analysis module is used to construct three types of discrete optimization objectives, perform correlation feature analysis on multi-source data in the sample database, obtain the correlation coefficient between discrete feature variables and discrete optimization objectives in multi-source data, and filter discrete feature variables based on the correlation coefficient, retaining discrete feature variables that meet the requirements. The data prediction module is used to build a multi-source feature prediction model. It extracts discrete feature variables from the multi-source data of the current collection period and inputs them into the multi-source feature prediction model to obtain the discrete feature prediction variable sequence for the next collection period. The data processing module is used to obtain the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights of three types of discrete optimization objectives on the energy storage side based on the discrete feature prediction variable sequence. Based on the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights, the module performs energy storage strategy analysis to obtain the optimal energy storage strategy for the next acquisition cycle. The real-time monitoring module is used to monitor the discrete characteristic variable sequence of the current acquisition period in real time, and determine whether to perform energy storage strategy analysis based on the monitoring results.

[0013] The big data-based dynamic optimization method for energy storage strategies disclosed in this invention effectively solves the technical pain points of traditional energy storage scheduling methods, such as strong subjectivity in feature selection, insufficient prediction accuracy, single strategy optimization objective, and lag response, through a full-process technical architecture of discrete feature correlation analysis, accurate prediction of multi-source features, dynamic weight optimization of multi-objectives, and real-time monitoring and correction. It achieves an upgrade of energy storage strategies from "experience-driven" to "data-driven" and from "static planning" to "dynamic adaptation." Specific beneficial effects are as follows: 1. Traditional energy storage optimization methods often rely on manual experience to select feature variables, which can easily overlook key influencing factors or introduce redundant features, limiting the accuracy of subsequent model predictions and optimizations. This invention standardizes and discretizes continuous variables and optimization objectives from multi-source data. Based on marginal distribution coefficients and joint distribution coefficients, it calculates the correlation coefficients between discrete feature variables and discrete optimization objectives, and uses threshold filtering to retain strongly correlated feature variables. This process is entirely based on mathematical quantitative analysis, eliminating subjective interference from manual experience and ensuring that the selected feature variables are highly correlated with the three optimization objectives. Simultaneously, by eliminating weakly correlated and redundant features, it reduces the computational complexity of subsequent models, improves data utilization efficiency, and lays a high-quality data foundation for accurate feature prediction and strategy optimization.

[0014] 2. Traditional energy storage optimization methods often focus on a single objective (such as economic efficiency), neglecting the synergistic optimization of renewable energy absorption rate and energy storage lifespan. This can easily lead to contradictions such as "improved economic efficiency but increased energy storage losses" or "achieving absorption rate targets but low economic benefits." This invention constructs objective functions for three types of discrete optimization objectives and, based on the correlation characteristics between discrete feature variables and optimization objectives, combines fuzzy comprehensive evaluation to classify the power grid operating status levels, thereby dynamically adjusting the weight coefficients of each objective function. This dynamic weighting mechanism breaks the limitations of fixed weights and can adaptively adjust the priority of objectives according to the real-time operating status of the power grid (such as renewable energy penetration rate, energy storage health status, etc.)—prioritizing absorption rate during periods of high renewable energy penetration, prioritizing extending equipment lifespan during periods of energy storage decline, and prioritizing economic efficiency when peak-valley electricity price differences are large. By solving the optimal strategy through a multi-objective genetic algorithm, the synergistic optimization of the three major objectives of economic efficiency, renewable energy absorption rate, and energy storage lifespan is achieved, ultimately maximizing the comprehensive benefits of the energy storage system.

[0015] 3. Traditional energy storage strategies are mostly based on offline planning, lacking the ability to quickly respond to sudden changes in grid operating conditions. When the actual operating conditions deviate significantly from the predicted values, it can easily lead to operational risks such as power imbalance and energy storage overload. This invention sets up a real-time monitoring mechanism that compares the discrete feature variable sequence of the current acquisition period with the predicted sequence, and triggers the strategy correction process based on the deviation threshold. When the deviation exceeds the threshold, it can quickly call the multi-source feature prediction model to output the feature prediction sequence for the remaining time period of the current acquisition period, and re-optimize and generate an appropriate energy storage strategy. This real-time correction mechanism realizes closed-loop dynamic adjustment of the energy storage strategy, which can promptly respond to sudden changes in power output on the power source side and fluctuations in load demand on the load side, avoiding scheduling mismatch problems caused by prediction deviations, and ensuring the power balance of the grid and the safe and stable operation of the energy storage system. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a big data-based dynamic optimization method for energy storage strategies, as described in an embodiment of this application.

[0017] Figure 2 This is a flowchart of a big data-based dynamic optimization system for energy storage strategies, as described in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] like Figure 1 As shown, the dynamic optimization method for energy storage strategies based on big data includes the following steps: Step s1: Construct three types of discrete optimization objectives, perform correlation feature analysis on multi-source data in the sample database, obtain the correlation coefficient between discrete feature variables and discrete optimization objectives in multi-source data, filter discrete feature variables based on correlation coefficients, and retain discrete feature variables that meet the requirements; Step s2: Construct a multi-source feature prediction model. Extract discrete feature variables from the multi-source data of the current collection period and input them into the multi-source feature prediction model to obtain the discrete feature prediction variable sequence for the next collection period. Step s3: Based on the discrete feature prediction variable sequence, obtain the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights of the three types of discrete optimization objectives on the energy storage side. Based on the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights, perform energy storage strategy analysis to obtain the optimal energy storage strategy for the next acquisition cycle. Step s4: Monitor the discrete characteristic variable sequence of the current acquisition period in real time, and determine whether to conduct energy storage strategy analysis based on the monitoring results.

[0020] It should be further explained that, in the specific implementation process, a cloud-based monitoring center is constructed, and monitoring terminals are deployed on the power supply-side entities (traditional power plants, new energy power generation stations, and power grid transmission lines), load-side entities (various load user clusters), and energy storage-side entities (energy storage stations / energy storage clusters) within the target area. The cloud-based monitoring center communicates with several monitoring terminals in the target area, receives multi-source data collected by each monitoring terminal, marks the collection time, and sets the collection period (T=24h). A sample database is set up within the cloud-based monitoring center to store multi-source data from several historical collection periods.

[0021] It should be further explained that, in the specific implementation process, the process of constructing three types of discrete optimization objectives, performing correlation feature analysis on multi-source data in the sample database, and obtaining the correlation coefficients between discrete feature variables and discrete optimization objectives in the multi-source data includes: Multi-source data from several historical collection periods were extracted from the sample database as samples. All continuous variables (such as wind speed, solar irradiance, total load, SOC, peak-valley electricity price, charging and discharging power, etc.) were extracted from the multi-source data of each sample. A single sample was used for... The total number of samples is denoted as N, and three types of continuous target values ​​(economic benefits per unit time) are constructed. New energy consumption rate Energy storage lifespan loss rate A single sample is The total number of samples is consistent with the total number of samples of several continuous variables, which is N). All continuous variables and continuous target values ​​are standardized preprocessed (Z-Score standardization) to eliminate dimensional differences. For the standardized preprocessed continuous variable, select a threshold point to divide the range of values ​​into k discrete intervals, and then divide the sample values ​​of each standardized preprocessed continuous variable into k discrete intervals. Mapping to the corresponding discrete interval involves the following steps: For the sample values ​​of all continuous variables ( , ,..., Sort in ascending order to obtain the sorted sequence. According to the principle of equal sample size, Divide into k intervals (k=1,2,...,j), each interval contains N / k samples. Determine the boundary values ​​for each interval, and assign each continuous sample value... Mapped to the corresponding discrete interval Let the discretized variables be... For example, the sample value range of wind speed x is [0,12] m / s, the number of samples (N=1000), and we take (k=5). Then each interval contains 200 samples, and the discrete intervals are ([0,2.4),[2.4,4.8),[4.8,7.2),[7.2,9.6),[9.6,12]). Discrete feature variables are generated based on the discretized variables. Similarly, within the range of the standardized preprocessed continuous target values, a threshold point is selected to divide the range into z discrete intervals. The sample values ​​of each standardized preprocessed continuous target value are then... Mapping to the corresponding discrete intervals specifically involves dividing the range of continuous target values ​​into z discrete intervals. (z=1,2,...,m), for each sample value Mapped to the corresponding discrete interval Let the discretized target value be... A discrete optimization objective is generated based on the discretized objective value; The statistical discrete characteristic variable falls within each discrete interval Number of samples Discrete optimization objective falls within each discrete interval Number of samples and simultaneously falling within the discrete intervals corresponding to the discrete feature variables Discrete interval corresponding to the discrete optimization objective Number of samples For discrete feature variables falling within each discrete interval Number of samples Discrete optimization objective falls within each discrete interval Number of samples and simultaneously falling within the discrete intervals corresponding to the discrete feature variables Discrete interval corresponding to the discrete optimization objective Number of samples Perform probability normalization (number of samples divided by total number of samples N) to obtain the marginal distribution coefficients of the discrete feature variables. ( ), the marginal distribution coefficient p(y) of the discrete optimization objective ( ) and the joint distribution coefficient p(xy) between discrete characteristic variables and discrete optimization objective ( The correlation coefficient between discrete feature variables and discrete optimization objectives is obtained based on the marginal distribution coefficient and joint distribution coefficient.

[0022] The process of obtaining the correlation coefficient is as follows: ; in, Represents the correlation coefficient. Let be the joint distribution coefficient of the j-th discrete feature variable and the m-th discrete optimization objective. Let be the marginal distribution coefficient of the j-th discrete feature variable. Let be the marginal distribution coefficient of the m-th discrete optimization objective.

[0023] It should be further explained that, in the specific implementation process, the process of screening discrete feature variables based on the correlation coefficient and retaining the discrete feature variables that meet the requirements includes: A preset correlation coefficient threshold is set. The correlation coefficient between the discrete feature variable and each discrete optimization objective is compared with the correlation coefficient threshold. If the correlation coefficient between the discrete feature variable and each discrete optimization objective is less than the correlation coefficient threshold, the discrete feature variable is removed. If the correlation coefficient between the discrete feature variable and a discrete optimization objective is greater than or equal to the correlation coefficient threshold, the discrete feature variable is retained.

[0024] It should be further explained that, in the specific implementation process, the process of constructing a multi-source feature prediction model includes: Extract multi-source data from power supply-side entities, load-side entities, and energy storage-side entities within several historical acquisition periods. Extract discrete feature variables from the multi-source data as training samples to construct a multi-source feature prediction model. Use the training samples to train the multi-source feature prediction model to obtain the trained multi-source feature prediction model. The multi-source feature prediction model adopts a bidirectional LSTM structure: the forward LSTM captures the trend features of historical data, and the backward LSTM captures the correlation features of future data (such as the correlation between the next day's load and the peak load of the current day). The bidirectional output is fused through a fully connected layer to improve the accuracy of prediction. The discrete feature variables of the power supply-side entity, load-side entity, and energy storage-side entity in the current acquisition cycle are input into the multi-source feature prediction model. Based on the multi-source feature prediction model, the discrete feature prediction variable sequence of the power supply-side entity, load-side entity, and energy storage-side entity in the next acquisition cycle is output.

[0025] It should be further explained that, in the specific implementation process, the process of obtaining the charging and discharging power constraints and SOC constraints on the energy storage side based on the discrete feature prediction variable sequence includes: Extract the power output sequence of the power source side in the next acquisition cycle (e.g., acquisition cycle T = 24 hours, time granularity) from the discrete characteristic prediction variable sequence of the power source side entity, load side entity, and energy storage side entity in the next acquisition cycle T. The total power output sequence (including the output power of all power sources such as thermal power, hydropower, wind power, and photovoltaic power) and the load-side demand power sequence at t=15 minutes. The source-load supply-demand difference for the next data collection period is obtained based on the power output sequence from the power source side and the power demand sequence from the power load side. , ,in, The source-side power output during time period t. The load power demand during time period t. The power loss of the power grid lines during time period t is obtained through calculations using power grid topology parameters and power flow. When the power grid capacity exceeds 0: the grid has a power surplus, and energy storage can be charged to absorb the surplus power (especially the surplus power from new energy sources). <0: Power shortage in the power grid. At this time, energy storage can discharge to make up for the load demand gap; obtain the charging and discharging power constraints of the energy storage side and set the SOC constraints of the energy storage side according to the difference between source and load supply and demand.

[0026] The charging and discharging power constraints are: ; , , ,in, For charging power, For discharge power, This is the upper limit of charging power on the energy storage side. This is the upper limit of the discharge power on the energy storage side; The SOC constraint is: ; ; in, For the rated capacity of energy storage, This refers to the charge / discharge efficiency.

[0027] It should be further explained that, in the specific implementation process, the objective functions and dynamic weights for three types of discrete optimization objectives are constructed based on the discrete feature prediction variable sequence. The process of analyzing energy storage strategies based on charge / discharge power constraints, SOC constraints, objective functions, and dynamic weights to obtain the optimal energy storage strategy includes: Objective functions for three types of discrete optimization objectives are constructed based on the discrete characteristic prediction variable sequences of the power supply-side entity, load-side entity, and energy storage-side entity in the next acquisition cycle. Several energy storage strategies are randomly generated based on the charging and discharging power constraints and SOC constraints of the energy storage side. Each energy storage strategy corresponds to a charging and discharging power sequence, which satisfies the charging and discharging power constraints and SOC constraints. The dynamic weights of the objective functions for the three types of discrete optimization objectives are obtained. A fitness function is constructed based on the objective functions and dynamic weights of the three types of discrete optimization objectives. Chromosome encoding and population initialization are performed on several energy storage strategies to generate an initial population. The optimal energy storage strategy is obtained through a multi-objective genetic algorithm based on the initial population and fitness function.

[0028] Fitness function: Where Y represents the fitness value, , and They represent , and Dynamic weights, For economic benefits per unit time, For the new energy consumption rate, This refers to the energy storage lifespan loss rate. The objective functions for the three types of discrete optimization objectives are as follows: ; Where t is the scheduling period (e.g., 15 minutes, t=1,2,...,T, where T is 96). This indicates the cumulative calculation over the data collection period. The depreciation factor per unit capacity (yuan / kWh) The change in SOC of energy storage during time period t is dimensionless, positive for charging and negative for discharging. Rated energy storage capacity (kWh). The electricity purchase price from the grid during time period t (yuan / kWh). The energy storage charging power (kW) during time period t. The energy storage discharge power (kW) during time period t. For charging and discharging efficiency, The unit subsidy standard for new energy consumption (yuan / kWh) The charging power (kW) of energy storage for absorbing new energy during time period t is included only when there is a surplus of new energy during the charging period. ; in, It is the output power of new energy sources on the power supply side during time period t. The amount of renewable energy curtailed during time period t (kW); ; in, The deep loss coefficient is fitted using the life decay curve provided by the battery manufacturer. This represents the change in depth of charge and discharge. This is the power impact loss coefficient. This represents the rate of change of energy storage power.

[0029] The process of obtaining the optimal energy storage strategy based on the initial population and fitness function using a multi-objective genetic algorithm includes: Energy storage strategies are encoded as chromosomes, for example, using a string of numbers to represent the charge / discharge power sequence information. A certain number of chromosomes are randomly generated to form an initial population, with each chromosome representing a possible energy storage strategy. The fitness function is obtained, and a tournament selection method is used to select chromosomes with higher fitness values ​​from the current population as parents. The parent chromosomes are crossbred, exchanging some genes to generate new offspring chromosomes, simulating biological genetic exchange to generate new production plans. The offspring chromosomes are mutated, randomly changing some genes to increase population diversity and avoid getting trapped in local optima. The above steps are repeated iteratively until the termination condition is met, such as reaching the maximum number of iterations or the fitness value no longer significantly improving. The optimal energy storage strategy is then output and distributed to the energy storage-side entity to execute charging / discharging actions. At the same time, the strategy is synchronized to the power supply-side entity (adjusting the unit output plan) and the load-side entity (guiding demand response).

[0030] It should be further explained that, in the specific implementation process, the process of obtaining the dynamic weights of the objective functions for the three types of discrete optimization objectives includes: The correlation coefficients between discrete feature variables and various discrete optimization objectives are compared with the correlation coefficient thresholds. Discrete feature variables with correlation coefficients greater than the correlation coefficient thresholds are marked as correlation features of discrete optimization objectives. Evaluation indicators are constructed based on the correlation characteristics of three types of discrete optimization objectives. Fuzzy sub-intervals of the evaluation indicators and the corresponding power grid state levels (excellent, good, medium, poor) for each fuzzy sub-interval are set. The membership matrix of the evaluation indicators for different power grid state levels is obtained through fuzzy comprehensive evaluation based on the evaluation indicators. For example, the correlation characteristic of renewable energy absorption rate is the penetration rate of renewable energy output power on the power supply side. , =New energy output power / Total power supply output; the correlation characteristics of energy storage life are: energy storage health state (SOH) and frequency deviation between grid frequency and rated frequency. ; The fuzzy subintervals corresponding to each power grid state level are: Excellent condition: ( <20%), ( < ), ( ≥90%), the power grid is operating stably; Good condition: (20%≤) <30%), ( ≤ < ), ( ≤ <90%; In progress: (30%≤ <40%) ≤ < ), ( ≤ < ); Poor condition: ( ≥40%) or ( ≥ )or( If the power grid operates at a rate of less than 80%, there is a risk to its operation.

[0031] The power grid state level of the evaluation index is obtained based on the membership matrix (the membership degree of the evaluation index to different power grid state levels in the membership matrix is ​​obtained, the power grid state level with the highest membership degree is selected, and the power grid state level with the highest membership degree is used as the power grid state level of the evaluation index). The weight adjustment coefficients corresponding to different power grid state levels are preset, and the initial weights of the objective functions of the three types of discrete optimization objectives are adjusted according to the weight adjustment coefficients to obtain the dynamic weights of the objective functions of the three types of discrete optimization objectives.

[0032] The formula for adjusting the initial weights of the objective functions of the three types of discrete optimization objectives according to the weight adjustment coefficient is as follows: ; in, For the adjustment coefficient, Let be the initial weights, i be the power grid state level, and j be the discrete optimization objective type. After adjustment, they satisfy... + + =1.

[0033] The weighting adjustment coefficients and adjustment results for different power grid state levels are shown in Table 1 below: Table 1

[0034] When the power grid condition is "poor" (e.g.) When the renewable energy consumption rate reaches 45%, the weight of the renewable energy consumption rate is increased to 0.6, giving priority to the consumption of renewable energy and avoiding power curtailment. When the storage SOH = 75% (decay period), the lifetime weight increases to 0.3, limiting deep charge and discharge.

[0035] It should be further explained that, in the specific implementation process, the real-time monitoring of the discrete characteristic variable sequence of the current acquisition period, and the determination of whether to conduct energy storage strategy analysis based on the monitoring results, includes: The discrete feature prediction variable sequence of the current acquisition period is used as the judgment standard. The discrete feature variable sequence of the current acquisition period is obtained, and the discrete feature variable sequence of the current acquisition period is compared with the judgment standard to obtain the discrete feature deviation of each time period (the ratio of the absolute value of the difference between the discrete feature variable and the judgment standard to the judgment standard). A preset discrete feature deviation threshold (3%) is set. The discrete feature deviation is compared with the discrete feature deviation threshold. If the discrete feature deviation is greater than the discrete feature deviation threshold, the discrete feature variable sequence of the current acquisition period is input into the multi-source feature prediction model. Based on the multi-source feature prediction model, the discrete feature prediction variable sequence of the power supply entity, load entity, and energy storage entity in the remaining time period of the current acquisition period is output. Based on the discrete feature prediction variable sequence of the remaining time period of the current acquisition period, the energy storage strategy is analyzed, and the optimal energy storage strategy for the remaining time period of the current acquisition period is output.

[0036] like Figure 2 As shown, the energy storage strategy dynamic optimization system based on big data includes a cloud monitoring center, which is connected to a data analysis module, a data prediction module, a data processing module, and a real-time monitoring module. The data analysis module is used to construct three types of discrete optimization objectives, perform correlation feature analysis on multi-source data in the sample database, obtain the correlation coefficient between discrete feature variables and discrete optimization objectives in multi-source data, and filter discrete feature variables based on the correlation coefficient, retaining discrete feature variables that meet the requirements. The data prediction module is used to build a multi-source feature prediction model. It extracts discrete feature variables from the multi-source data of the current collection period and inputs them into the multi-source feature prediction model to obtain the discrete feature prediction variable sequence for the next collection period. The data processing module is used to obtain the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights of three types of discrete optimization objectives on the energy storage side based on the discrete feature prediction variable sequence. Based on the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights, the module performs energy storage strategy analysis to obtain the optimal energy storage strategy for the next acquisition cycle. The real-time monitoring module is used to monitor the discrete characteristic variable sequence of the current acquisition period in real time, and determine whether to perform energy storage strategy analysis based on the monitoring results.

[0037] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A dynamic optimization method for energy storage strategies based on big data, characterized in that, Includes the following steps: Step s1: Construct three types of discrete optimization objectives, perform correlation feature analysis on multi-source data in the sample database, obtain the correlation coefficient between discrete feature variables and discrete optimization objectives in multi-source data, filter discrete feature variables based on correlation coefficients, and retain discrete feature variables that meet the requirements; Step s2: Construct a multi-source feature prediction model. Extract discrete feature variables from the multi-source data of the current collection period and input them into the multi-source feature prediction model to obtain the discrete feature prediction variable sequence for the next collection period. Step s3: Based on the discrete feature prediction variable sequence, obtain the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights of the three types of discrete optimization objectives on the energy storage side. Based on the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights, perform energy storage strategy analysis to obtain the optimal energy storage strategy for the next acquisition cycle. Step s4: Monitor the discrete characteristic variable sequence of the current acquisition period in real time, and determine whether to conduct energy storage strategy analysis based on the monitoring results.

2. The method for dynamic optimization of energy storage strategies based on big data according to claim 1, characterized in that, A cloud-based monitoring center is constructed, and monitoring terminals are deployed on the power supply-side entities, load-side entities, and energy storage-side entities within the target area. The cloud-based monitoring center communicates with several monitoring terminals in the target area, receives multi-source data collected by each monitoring terminal, marks the collection time, and sets the collection cycle. A sample database is set up within the cloud-based monitoring center to store multi-source data from several historical collection cycles.

3. The method for dynamic optimization of energy storage strategies based on big data according to claim 2, characterized in that, The process of constructing three types of discrete optimization objectives, performing correlation feature analysis on multi-source data in a sample database, and obtaining the correlation coefficients between discrete feature variables and discrete optimization objectives in the multi-source data includes: Several historical collection periods of multi-source data were extracted from the sample database as samples. All continuous variables in the multi-source data of each sample were extracted and three types of continuous target values ​​were constructed. All continuous variables and continuous target values ​​were standardized and preprocessed. For the range of values ​​of the standardized preprocessed continuous variable, select a threshold point to divide it into k discrete intervals, and map the sample value of each standardized preprocessed continuous variable to the corresponding discrete interval to generate discrete feature variables; Similarly, within the range of values ​​of the continuous target value after standardization and preprocessing, a threshold point is selected to divide the range into z discrete intervals. The sample value of each continuous target value after standardization and preprocessing is mapped to the corresponding discrete interval to generate a discrete optimization target. The number of samples falling into each discrete interval of the discrete feature variable, the number of samples falling into each discrete interval of the discrete optimization objective, and the number of samples falling into both the discrete intervals corresponding to the discrete feature variable and the discrete interval corresponding to the discrete optimization objective are statistically analyzed. Probability normalization is then performed on the number of samples falling into each discrete interval of the discrete feature variable, the number of samples falling into each discrete interval of the discrete optimization objective, and the number of samples falling into both the discrete intervals corresponding to the discrete feature variable and the discrete optimization objective. The marginal distribution coefficients of the discrete feature variable, the marginal distribution coefficients of the discrete optimization objective, and the joint distribution coefficients between the discrete feature variable and the discrete optimization objective are then obtained. The correlation coefficient between the discrete feature variable and the discrete optimization objective is obtained based on the marginal distribution coefficients and the joint distribution coefficients.

4. The method for dynamic optimization of energy storage strategies based on big data according to claim 3, characterized in that, The process of selecting discrete feature variables based on correlation coefficients includes: A preset correlation coefficient threshold is set. The correlation coefficient between the discrete feature variable and each discrete optimization objective is compared with the correlation coefficient threshold. If the correlation coefficient between the discrete feature variable and each discrete optimization objective is less than the correlation coefficient threshold, the discrete feature variable is removed. If the correlation coefficient between the discrete feature variable and a discrete optimization objective is greater than or equal to the correlation coefficient threshold, the discrete feature variable is retained.

5. The method for dynamic optimization of energy storage strategies based on big data according to claim 4, characterized in that, The process of constructing a multi-source feature prediction model includes: Extract multi-source data from power supply-side entities, load-side entities, and energy storage-side entities within several historical acquisition periods. Extract discrete feature variables from the multi-source data as training samples to construct a multi-source feature prediction model. Use the training samples to train the multi-source feature prediction model to obtain the trained multi-source feature prediction model. The discrete feature variables of the power supply-side entity, load-side entity, and energy storage-side entity in the current acquisition cycle are input into the multi-source feature prediction model. Based on the multi-source feature prediction model, the discrete feature prediction variable sequence of the power supply-side entity, load-side entity, and energy storage-side entity in the next acquisition cycle is output.

6. The method for dynamic optimization of energy storage strategies based on big data according to claim 5, characterized in that, The process of obtaining the charging and discharging power constraints on the energy storage side includes: Extract the power output sequence and load demand sequence for the next data acquisition cycle from the discrete characteristic prediction variable sequences of the power supply-side entity, load-side entity, and energy storage-side entity in the next data acquisition cycle. The power supply and demand difference between the power source and the load side is obtained in the next acquisition cycle based on the power output sequence of the power source and the power demand sequence of the load side. The charging and discharging power constraints of the energy storage side are obtained based on the power supply and demand difference.

7. The method for dynamic optimization of energy storage strategies based on big data according to claim 6, characterized in that, The process of conducting energy storage strategy analysis to obtain the optimal energy storage strategy includes: Objective functions for three types of discrete optimization objectives are constructed based on the discrete characteristic prediction variable sequences of the power supply entity, load entity, and energy storage entity in the next acquisition cycle. Several energy storage strategies are randomly generated based on the charging and discharging power constraints and SOC constraints of the energy storage side, with each energy storage strategy corresponding to a charging and discharging power sequence. The dynamic weights of the objective functions for the three types of discrete optimization objectives are obtained. A fitness function is constructed based on the objective functions and dynamic weights of the three types of discrete optimization objectives. Chromosome encoding and population initialization are performed on several energy storage strategies to generate an initial population. The optimal energy storage strategy is obtained through a multi-objective genetic algorithm based on the initial population and the fitness function.

8. The method for dynamic optimization of energy storage strategy based on big data according to claim 7, characterized in that, The process of obtaining the dynamic weights of the objective function for three types of discrete optimization objectives includes: The correlation coefficients between discrete feature variables and various discrete optimization objectives are compared with the correlation coefficient thresholds. Discrete feature variables with correlation coefficients greater than the correlation coefficient thresholds are marked as correlation features of discrete optimization objectives. Evaluation indices are constructed based on the correlation characteristics of three types of discrete optimization objectives. Fuzzy sub-intervals of the evaluation indices and the power grid state level corresponding to each fuzzy sub-interval are set. The specific values ​​of the evaluation indices are obtained by predicting the variable sequence based on the discrete characteristics of the power source entity, load entity, and energy storage entity in the next acquisition cycle. The membership matrix of the specific values ​​of the evaluation indices to different power grid state levels is obtained through fuzzy comprehensive evaluation. The power grid state level of the evaluation index is obtained based on the membership matrix. The weight adjustment coefficients corresponding to different power grid state levels are preset. The initial weights of the objective functions of the three types of discrete optimization objectives are adjusted according to the weight adjustment coefficients to obtain the dynamic weights of the objective functions of the three types of discrete optimization objectives.

9. The method for dynamic optimization of energy storage strategy based on big data according to claim 8, characterized in that, The process of real-time monitoring of the discrete characteristic variable sequence of the current acquisition period and determining whether to conduct energy storage strategy analysis based on the monitoring results includes: The discrete feature prediction variable sequence of the current acquisition period is used as the judgment standard to obtain the discrete feature variable sequence of the current acquisition period. The discrete feature variable sequence of the current acquisition period is compared with the judgment standard to obtain the discrete feature deviation of each time period. A preset discrete feature deviation threshold is set, and the discrete feature deviation is compared with the discrete feature deviation threshold. If the discrete feature deviation is greater than the discrete feature deviation threshold, the discrete feature variable sequence of the current acquisition period is input into the multi-source feature prediction model. Based on the multi-source feature prediction model, the discrete feature prediction variable sequence of the power supply entity, load entity, and energy storage entity in the remaining time period of the current acquisition period is output. Based on the discrete feature prediction variable sequence of the remaining time period of the current acquisition period, energy storage strategy analysis is performed, and the optimal energy storage strategy for the remaining time period of the current acquisition period is output.

10. A big data-based dynamic optimization system for energy storage strategies, specifically applied to the big data-based dynamic optimization method for energy storage strategies as described in any one of claims 1 to 9, characterized in that, This includes a cloud-based monitoring center, which has communication connections with data analysis, data prediction, data processing, and real-time monitoring modules. The data analysis module is used to construct three types of discrete optimization objectives, perform correlation feature analysis on multi-source data in the sample database, obtain the correlation coefficient between discrete feature variables and discrete optimization objectives in multi-source data, and filter discrete feature variables based on the correlation coefficient, retaining discrete feature variables that meet the requirements. The data prediction module is used to build a multi-source feature prediction model. It extracts discrete feature variables from the multi-source data of the current collection period and inputs them into the multi-source feature prediction model to obtain the discrete feature prediction variable sequence for the next collection period. The data processing module is used to obtain the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights of three types of discrete optimization objectives on the energy storage side based on the discrete feature prediction variable sequence. Based on the charging and discharging power constraints, SOC constraints, objective functions and dynamic weights, the module performs energy storage strategy analysis to obtain the optimal energy storage strategy for the next acquisition cycle. The real-time monitoring module is used to monitor the discrete characteristic variable sequence of the current acquisition period in real time, and determine whether to perform energy storage strategy analysis based on the monitoring results.

Images