Method for establishing energy storage scheduling model of multi-node optical storage direct-flexible system

Abstract

The invention relates to the technical field of optical storage direct-flexible scheduling, and discloses a method for establishing an energy storage scheduling model of a multi-node optical storage direct-flexible system. The method comprises the following steps: acquiring real-time operation data of each node of a system, including a photovoltaic output fluctuation curve, an energy storage charge state, a load demand time sequence, a power grid interaction power limit value and an environmental irradiance change characteristic, and performing multi-dimensional alignment processing on the real-time operation data to form a multi-dimensional operation data set; performing operation state mapping topology construction processing on the data set, and separating out target operation data and state indication data; based on the target operation data, in combination with scheduling priority configuration information of each node, generating a multi-node collaborative energy storage scheduling initial model containing an energy storage charging and discharging time sequence strategy and a power distribution scheme; and identifying node operation abnormal characteristics based on the state indication data, dynamically reconstructing the initial model according to the node operation abnormal characteristics, and outputting the updated energy storage scheduling model to guarantee stable and efficient operation of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic-storage-direct-flexible scheduling, and specifically to a method for establishing an energy storage scheduling model for a multi-node photovoltaic-storage-direct-flexible system. Background Art

[0002] With the rapid development of new energy generation technologies, photovoltaics, as a crucial component of clean and renewable energy, continue to increase their share in the energy system. PV-storage-direct-flexible systems integrate photovoltaic power generation, energy storage, DC power distribution, and flexible control technologies, effectively improving energy efficiency and reducing reliance on traditional power grids. However, in multi-node PV-storage-direct-flexible systems, the operating status of each node varies significantly and is significantly affected by external environmental factors, leading to numerous challenges to the stability and reliability of the overall system operation. Currently, energy storage scheduling methods for PV-storage direct-flexible systems mostly focus on single-node or simple multi-node scenarios, lacking comprehensive consideration of the complex operating environments of multiple nodes. In actual operation, the photovoltaic output of each node exhibits significant fluctuations due to changes in ambient irradiance. The state of charge of the energy storage device changes dynamically during the charging and discharging process, and load demand also exhibits temporal variations. Furthermore, grid interaction power is subject to strict limits. This real-time operating data comes from diverse sources and dimensions. If it is not effectively integrated and processed, it can easily lead to deviations or conflicts between the data, thereby affecting the accuracy of energy storage scheduling decisions. Existing scheduling models often fail to fully consider the differences in scheduling priorities among nodes during their construction, making it difficult to achieve coordinated, optimized scheduling across multiple nodes. This can lead to insufficient or excessive energy supply at some nodes, reducing the overall energy efficiency of the system. Furthermore, when an operational anomaly occurs at a node in the system, existing models lack the ability to quickly identify and dynamically adjust, making it impossible to promptly restructure the scheduling strategy. This can cause the anomaly to spread and impact the stable operation of the entire system. These issues make it difficult for multi-node PV-storage direct-flexible systems to fully realize their advantages in practical applications, limiting their promotion and application in large-scale energy systems. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for establishing an energy storage scheduling model for a multi-node solar-storage direct-flexible system to solve the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention provides a method for establishing an energy storage scheduling model for a multi-node PV-storage direct-flexible system, the method comprising: Acquire real-time operating data of each node in the PV-storage-direct-flexible system and perform multi-dimensional alignment processing on the real-time operating data to form a multi-dimensional operating data set. The real-time operating data includes the PV output fluctuation curve, energy storage charge state, load demand timing, grid interaction power limit, and ambient irradiance variation characteristics; Performing operation status mapping topology construction processing on the multi-dimensional operation data set to separate target operation data for scheduling model construction and status indication data for scheduling abnormal response; Based on the target operation data and in combination with the scheduling priority configuration information of each node, an initial energy storage scheduling model for multi-node collaboration is generated, wherein the initial energy storage scheduling model includes an energy storage charging and discharging timing strategy and a power allocation scheme; Based on the status indication data, abnormal node operation characteristics during the real-time operation data update process are identified, the energy storage scheduling initial model is dynamically reconstructed according to the abnormal node operation characteristics, and an updated energy storage scheduling model is output.

[0005] Preferably, the performing operation status mapping topology construction processing on the multi-dimensional operation data set includes: Convert the multi-dimensional operation data set after aligning the time, node and device dimensions into a scheduling state mapping matrix; Decoupling the state characteristics of the scheduling state mapping matrix to obtain a steady-state characteristic matrix and a transient characteristic matrix, wherein the steady-state characteristic matrix represents the normalized basic structure of the scheduling model, and the transient characteristic matrix represents the abnormal state characteristics corresponding to power limit exceeding, energy storage attenuation, and radiation mutation; The steady-state characteristic matrix is ​​used as the target operating data, and the transient characteristic matrix is ​​used as the state indication data.

[0006] Preferably, generating an initial model for multi-node coordinated energy storage scheduling includes: Extracting the photovoltaic output prediction segment, energy storage charge state interval, load demand peak and valley characteristics, and grid interaction constraint boundaries from the steady-state characteristic matrix; Configuring an optimized weight coefficient for each node based on the scheduling priority configuration information; According to the optimization weight coefficient, a multi-objective fusion calculation is performed on the photovoltaic output forecast segment, the energy storage charge state interval, the load demand peak and valley characteristics and the grid interaction constraint boundary to generate the energy storage charging and discharging timing strategy and power allocation plan.

[0007] Preferably, generating an initial model for multi-node coordinated energy storage scheduling further includes: Configure flexible regulation targets based on the business attributes of the PV-storage direct-flexible system, including maximizing PV absorption, balancing energy storage life, peak-load shaving and valley-filling, and optimizing grid interaction costs; Performing hierarchical identification processing on the flexible adjustment target to determine the primary optimization target, the secondary optimization target and the boundary constraint target; A key adjustment parameter set is selected based on the primary optimization objective and the secondary optimization objective, and a scheduling impact quantitative analysis of the key adjustment parameter set is performed.

[0008] Preferably, performing the scheduling impact quantitative analysis of the key adjustment parameter set includes: extracting an adjustable parameter range from the steady-state characteristic matrix; Performing parameter sensitivity mapping on the primary optimization objective and the secondary optimization objective to generate a target impact quantization gradient; Constructing an interactive influence relationship matrix of all target items in the flexible adjustment target, wherein the interactive influence relationship matrix includes target coordination relationships and target conflict relationships; Calculating sensitivity eigenvalues ​​of the adjustable parameter set based on the target impact quantization gradient and the interaction impact relationship matrix; A parameter optimization constraint set is established according to the sensitivity eigenvalues.

[0009] Preferably, the output updated energy storage scheduling model includes: Setting control boundaries for the key adjustment parameter set and creating an initial solution space based on the current operating state; Performing fitness evaluation on the solutions in the initial solution space, and generating an optimization direction vector and a gradient optimization step size in combination with the parameter optimization constraint set and the fitness evaluation result; Iteratively updating the initial solution space based on the optimization direction vector and the gradient optimization step size; The updated energy storage scheduling model is output according to the iterative update result.

[0010] Preferably, the identifying abnormal node operation characteristics during the real-time operation data update process based on the status indication data includes: Monitoring abnormal marking entries of the transient characteristic matrix, and associating the node operation deviation, energy storage attenuation rate, and radiation mutation intensity corresponding to the abnormal marking entries; Generate cross-node abnormal correlation features based on the node operation deviation, energy storage decay rate and radiation mutation intensity; When the cross-node abnormal correlation feature exceeds a preset threshold, a multi-node collaborative rescheduling strategy is triggered, and the power allocation plan and energy storage charging and discharging timing strategy are updated according to the transient feature matrix.

[0011] Preferably, the triggering of the multi-node collaborative rescheduling strategy includes: Extracting the node number, abnormal time period and abnormal type feature vector of the abnormality marked in the transient feature matrix; configuring a node coupling weight coefficient for the eigenvector; Performing cluster analysis on the scheduling correlation characteristics of the abnormal node group according to the node coupling weight coefficient to generate a first abnormal cluster and a second abnormal cluster; The first abnormal cluster represents a scheduling anomaly caused by PV output fluctuations, and the second abnormal cluster represents a scheduling anomaly caused by load mutations or energy storage failures. The first abnormal cluster and the second abnormal cluster are distinguished by the node scheduling deviation.

[0012] Preferably, the iterative updating of the initial solution space based on the optimization direction vector and the gradient optimization step size includes: Establish an iterative path identification for each solution, and classify the path status according to the fitness value of each iteration; Identifying the path update state within a preset iterative evaluation window and generating a gradient optimization morphology classification, wherein the gradient optimization morphology classification includes a convergent morphology, an exploratory morphology, and a divergent morphology; A solution space search optimization management is performed based on the gradient optimized morphological classification.

[0013] Preferably, the performing solution space search optimization management based on the gradient optimized morphological classification includes: A gradient proxy model is configured for the convergence pattern, and the optimization trend direction is predicted by the gradient proxy model; Configuring an optimized tabu window for the divergent morphology, and eliminating invalid search directions through the optimized tabu window; Based on the optimization trend direction and the optimization taboo window, a refined iteration is performed on the convergent morphological solution, a mixed direction iteration is performed on the exploratory morphological solution, and a random restart iteration is performed on the divergent morphological solution.

[0014] Compared with the prior art, the present invention has the following beneficial effects: By comprehensively acquiring and multi-dimensionally aligning the real-time operating data of each node in the photovoltaic-storage-direct-flexible system, it is possible to effectively integrate data from diverse sources and dimensions, such as photovoltaic output fluctuation curves, energy storage charge status, load demand timing, grid interaction power limits, and environmental irradiance change characteristics, to form a unified multi-dimensional operating data set. This avoids the adverse effects of data deviations or conflicts on scheduling decisions, and provides a more reliable data foundation for the subsequent scheduling model construction. During the construction and processing of the operating status mapping topology, this method accurately separates the target operating data used for scheduling model construction from the status indicator data used for scheduling anomaly responses, achieving classified data utilization. This classification processing method, on the one hand, allows the target operating data to be more focused on the core requirements of the scheduling model, reducing interference from irrelevant data, and helping to improve the efficiency and accuracy of the initial energy storage scheduling model. On the other hand, the separate separation of status indicator data provides clear data support for the subsequent identification of node operation anomaly characteristics, making the anomaly identification process more targeted. The initial model for multi-node coordinated energy storage scheduling, generated based on target operating data and combined with each node's scheduling priority configuration, fully considers the importance of different nodes in the system and the differences in their operational requirements. By rationally formulating energy storage charge and discharge timing strategies and power allocation schemes, it is possible to achieve energy collaborative optimization among nodes, ensuring a rational energy distribution across them and avoiding imbalances in energy supply and demand at certain nodes. This, in turn, improves the energy efficiency of the entire PV-storage-direct-flexible system and promotes the overall economic efficiency of the system. Based on status indicator data, this method can promptly identify abnormal node operating characteristics during real-time operating data updates and dynamically reconfigure the initial energy storage scheduling model based on these abnormal characteristics. This dynamic adjustment capability enables the scheduling model to quickly adapt to changes in node operating status. When a node experiences an anomaly, the model can promptly adjust the charge and discharge timing strategies and power allocation schemes to prevent the anomaly from significantly impacting overall system operation, thus ensuring the stability and reliability of the multi-node PV-storage-direct-flexible system. Furthermore, the dynamic reconfiguration mechanism provides greater flexibility in the scheduling model, enabling it to adapt to varying environmental conditions and operating scenarios, further expanding the application scope of multi-node PV-storage-direct-flexible systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a working principle diagram of the method for establishing an energy storage scheduling model for a multi-node PV-storage direct-flexible system according to the present invention; Figure 2 A flowchart of a process for building a topology for a running state map; Figure 3 A flow chart for generating an initial model for multi-node collaborative energy storage scheduling; Figure 4 Output the updated flow chart of the energy storage scheduling model; Figure 5 Flowchart for triggering the multi-node collaborative rescheduling strategy. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] See also Figure 1 The present invention provides a method for establishing an energy storage scheduling model for a multi-node PV-storage direct-flexible system, the method comprising: Real-time operating data from each node in the PV-storage direct-flexible system is acquired. This data covers PV output fluctuations, energy storage state of charge, load demand timing, grid interaction power limits, and ambient irradiance variations. This real-time operating data is then aligned in multiple dimensions to form a multi-dimensional operating dataset. This process includes timestamp alignment, node identification unification, and data sampling frequency normalization, ensuring consistency and comparability across time and space for data from different sources. Subsequently, an operating status mapping topology is constructed on the multi-dimensional operating dataset. This process separates the dataset into two components: target operating data for scheduling model construction and status indicator data for scheduling anomaly response. Based on the target operating data and combined with the pre-configured scheduling priority information for each node, an initial multi-node coordinated energy storage scheduling model is generated. This model includes a specific energy storage charge and discharge timing strategy and a power allocation scheme among nodes. Furthermore, based on the status indicator data, abnormal node operating characteristics that occur during the real-time operating data update process are identified, including power limit violations and sudden state changes. Based on the identified abnormal features, the generated initial energy storage scheduling model is dynamically reconstructed, its strategies and plans are adjusted, and finally an updated energy storage scheduling model is output to adapt to changes in the real-time operating status of the system.

[0018] Example 1: See Figure 2The implementation involves converting a multi-dimensional operational dataset into a dispatch state mapping matrix, which is then decoupled into steady-state and transient feature matrices. This process begins with further structuring of the completed multi-dimensional aligned data. Although the multi-dimensional operational dataset has been aligned along the time, node, and device dimensions, it still represents a discrete set of data points and needs to be converted into a matrix format more conducive to global analysis and feature extraction. The dispatch state mapping matrix is ​​constructed using a two-dimensional table structure with time as the row and node-device combinations as the column. Each row represents a unified timestamp derived from the global time series determined during the alignment process, with the interval determined by the system's data collection frequency. Each column represents a node and its specific device type, such as "PV panel at node A," "energy storage unit at node B," or "load at node C." Each element in the matrix is ​​populated with the specific operational parameter value for the corresponding time and node-device combination. These values ​​undergo a scale check before entry to ensure consistency of physical meaning. For example, power units are all in kilowatts and state of charge is in percentages, but normalization is not yet performed. The construction of this matrix successfully solidifies the panoramic operating status of the system within a specific time period in the form of a two-dimensional data structure, providing an operation object for subsequent feature separation.

[0019] Decoupling the state characteristics of the dispatch state mapping matrix is ​​a core step. Its purpose is to separate components representing the system's long-term stable operating trends from those representing short-term fluctuations and abnormal events. This process draws on the principle of separating fundamental and harmonics in signal processing, but is applied to multidimensional operational data. A sequence analysis method combining sliding window mean filtering and residual calculation is employed. For each data sequence (i.e., each column) in the matrix, a sliding window of appropriate length is defined. This window length should cover most normal operating fluctuation cycles while being significantly shorter than the system's primary diurnal or seasonal cycles. The arithmetic mean of the data within each window is calculated, and this mean sequence serves as the steady-state trend line for that data sequence. The difference between the original data sequence and this steady-state trend line constitutes the transient fluctuation component of the sequence. For data with strong periodicity, such as photovoltaic output and load, the periodic component is stripped before performing this operation to more accurately extract the non-periodic steady-state trend. The steady-state trend lines of all data columns together constitute the steady-state characteristic matrix, while the transient fluctuation components of all data columns together constitute the transient characteristic matrix.

[0020] The formation of the steady-state characteristic matrix represents a quantitative expression of the system's fundamental operating structure. Each value in this matrix represents a "baseline" operating level at the corresponding time and node, excluding short-term disturbances. Before being delivered as target operating data to the subsequent model-building module, it undergoes normalization. Normalization is performed independently on each node-device type's data column. Its purpose is to eliminate variations caused by different physical dimensions and numerical ranges, confining all data to a unified, dimensionless numerical range, typically mapped between zero and one. The maximum and minimum values ​​used for normalization are not the absolute extremes of the data column, but rather the upper and lower limits of the reasonable operating range determined by the device's nameplate parameters, historical operating statistics, or safe operating procedures. For example, the energy storage state of charge is normalized based on its minimum and maximum allowable states of discharge and charge, rather than its physical limits of zero and one hundred. After normalization, the steady-state characteristic matrix is ​​converted into a structured data set that purely represents the relative operating status level of each node. Its numerical value directly reflects the position of the node's operating status relative to its normal range at that moment, which greatly facilitates the configuration and calculation of multi-objective weights in the subsequent optimization algorithm.

[0021] The processing path for the transient characteristic matrix is ​​distinct from the steady-state matrix. Its core value lies in capturing and quantifying abnormal events that deviate from the "baseline" state during system operation. The values ​​in this matrix represent the difference between the raw data and the steady-state trend. They may contain positive and negative values, and the magnitude of the values ​​directly reflects the amplitude of the fluctuations. To more clearly identify anomalies, these transient fluctuation components require feature extraction and labeling. During implementation, anomaly thresholds are set for each type of operating parameter. These thresholds are typically adaptive and determined based on a multiple of the standard deviation of the parameter's historical fluctuation statistics. When the absolute value of a value in the transient matrix exceeds the threshold, the corresponding data point is marked as an anomaly. Furthermore, for certain specific anomalies, further feature calculation is required. For example, for energy storage units, the focus is not only on the SOC fluctuation of a single point, but also on the rate of change of the SOC over a period of time—the decay or growth rate. This is calculated by analyzing the trends of multiple consecutive points in the transient sequence. For ambient irradiance, the focus is on the severity of the change, the mutation intensity, which is characterized by calculating the change per unit time. All of these anomaly flags, rate calculations, and intensity characterizations are integrated into the metadata of the transient signature matrix or presented as a parallel flag matrix, collectively forming status indication data. This status indication data acts like a system diagnostic report, clearly indicating when, where, and on which device the anomaly occurred, what type of anomaly occurred, and its severity.

[0022] The separate construction of steady-state and transient characteristic matrices physically establishes a dual description system of the system's operating state: a "baseline" and "deviation." The steady-state matrix depicts the ideal operating trajectory of the system and serves as the target state that the scheduling model strives to achieve and maintain. The transient matrix, on the other hand, monitors and reports any deviations from this ideal baseline in real time. This separation allows the subsequent model construction and exception response modules to function effectively and efficiently. The model construction module can focus on developing the optimal scheduling plan based on a clear, noise-free steady-state "blueprint," while the exception response module, based on precise, quantitative "diagnostic" reports, can quickly determine whether and how to intervene and adjust the established plan. The relative independence and synergy of these two components are crucial foundations for the efficient and robust scheduling capabilities of the entire approach. The entire implementation relies on a deep understanding of the system's operating principles and the appropriate application of data processing technologies. Ultimately, the raw, complex operational data stream is refined into two distinct and clearly structured high-level data products, driving subsequent intelligent scheduling decisions.

[0023] Example 2: See Figure 3 The implementation focuses on extracting key operational characteristics from the steady-state characteristic matrix and generating an initial energy storage scheduling model based on multi-objective fusion calculations. This also involves the stratification of flexible regulation objectives and the selection of key parameter sets. This process begins with an in-depth analysis of the constructed steady-state characteristic matrix. As a structured data set that has been cleaned, aligned, and normalized, the steady-state characteristic matrix contains core information about the system's operational status over the next period of time. The extraction operation is not simply data reading, but rather the targeted identification and segmentation of fields with specific physical meanings. Extracting the PV output forecast range relies on analyzing the PV node data columns in the matrix. Combining recent historical data with weather forecast information, a time series prediction algorithm is used to derive the expected power generation curve for several future time intervals. This curve reflects the expected changes in PV power generation capacity. Determining the energy storage state-of-charge range requires reading the current normalized state value of each energy storage node and, based on the maximum and minimum allowable charge and discharge thresholds provided by the battery management system, calculating the safe state-of-charge range for each energy storage unit at the current moment. This range is a dynamically changing window. Peak and valley characteristics of load demand are identified through statistical analysis of the load data columns in the matrix. Clustering algorithms or rule-based judgments are used to identify peak and valley periods of electricity demand, as well as their typical power levels, as shown in historical and recent data, thereby predicting the shape of future load curves. The grid interaction constraint boundaries are a set of relatively fixed external conditions, but key constraints such as time-of-use electricity prices, maximum purchasable power, and maximum return power can be analyzed from the relevant fields of the matrix. These boundary conditions define the framework for energy exchange between the system and the external grid.

[0024] After the above feature extraction is completed, scheduling priority configuration information is introduced to guide subsequent optimization. This priority information is a set of strategic parameters predefined by the system operator that quantifies the importance of different nodes in the global optimization process. Priority can be determined based on a variety of factors, such as a node connected to critical loads requiring higher power reliability; a node with higher unit energy storage costs requiring more precise control of its cycle times; or nodes located in areas with more abundant sunlight resources, giving them a higher priority for photovoltaic consumption. These qualitative strategies are converted into specific optimization weight coefficients and assigned to each node. The weight coefficient is a positive real number whose absolute value reflects the contribution of the node to the objective function, while the relative proportion of the weights of different nodes determines the allocation of optimization resources among them. A node with a higher weight coefficient incurs a higher cost if its state deviates from the expected value during the optimization process. Therefore, the algorithm tends to prioritize meeting the operational needs of that node or making better use of its resources.

[0025] Multi-objective fusion calculation is the core optimization process, initiated after the aforementioned preparations are complete. This process coordinates multiple, potentially conflicting, system operational objectives extracted from the steady-state characteristic matrix. These objectives typically encompass maximizing local consumption of photovoltaic power generation, maintaining and balancing the energy storage state, smoothing the load curve, and minimizing grid interaction costs. Fusion calculation employs mathematical programming to construct an optimization problem model encompassing multiple objective functions. Optimization weight coefficients are used to construct a weighted sum objective function, transforming the multi-objective problem into a single-objective problem for solution or defining a priority-based optimization order. The calculation process also strictly adheres to the various constraints extracted from the matrix, including dynamic upper and lower limits on the energy storage state of charge, hard constraints on grid power interaction, and rigid load requirements. The choice of optimization algorithm depends on the scale and complexity of the problem and may include linear programming, quadratic programming, or more complex heuristic algorithms. Through iterative solution, this calculation process ultimately outputs a detailed energy storage charging and discharging timing strategy. This strategy is a time-series plan that specifies the actions each energy storage unit will be instructed to perform during each future scheduling period: charging to absorb excess energy, discharging to supplement power shortages, or remaining in idle standby mode. Furthermore, as a quantitative complement to the timing strategy, the power allocation scheme precisely specifies how the total power demand or surplus power will be distributed among the various energy storage units when multiple units need to operate simultaneously. This allocation is typically based on each unit's remaining capacity, rated power, and priority weight.

[0026] Beyond generating the framework for the initial dispatch model, configuring the flexible regulation targets of the PV-storage direct-flexible system based on the specific application scenarios it serves is a crucial task. Flexible regulation targets are the macroscopic effects that the system desires, defined from the user or operator's perspective. They imbue the mathematical model with a business-critical nature. The maximization of PV absorption requires the system to prioritize energy storage and other flexible resources to absorb PV power, reducing "abandoned" solar power and increasing the proportion of renewable energy. The energy storage lifespan equalization objective focuses on using control strategies to prevent overuse of some storage units while leaving others idle, thereby aligning the aging rate of the entire storage system and extending its overall service life. The peak-load shaving objective aims to utilize the energy storage system to discharge during peak grid load periods and charge during off-peak periods, thereby smoothing the overall load curve and alleviating grid pressure. The grid interaction cost optimization objective focuses on electricity expenditures, achieving optimal operational economics by intelligently adjusting power purchase, sales, and self-use strategies during different electricity price periods. There may be synergies between these goals, but there may also be inherent conflicts. For example, during peak electricity price periods, energy storage discharge is needed to achieve peak shaving and reduce electricity purchase costs, but frequent charge and discharge cycles may affect the life of energy storage.

[0027] Faced with multiple potentially competing flexible objectives, hierarchical identification is a critical prerequisite for effective optimization. This process ranks and categorizes the multiple pre-defined flexible objectives based on their importance, taking into account the current system operating conditions, external market environment, and operational strategy. This ranking results in the identification of a primary optimization objective that is central to the current scheduling cycle. For example, during periods of high PV generation and low electricity prices, maximizing PV consumption may be the primary objective; whereas during periods of peak grid demand and extremely high electricity prices, reducing grid interaction costs may become the primary objective. Objectives of lesser importance are classified as secondary optimization objectives, which are optimized while ensuring that the primary objective is met as much as possible. Furthermore, certain mandatory hard conditions, such as ensuring power supply to critical loads and complying with grid power constraints, are identified as boundary constraint objectives. These are typically presented in the optimization model as constraints rather than objective functions. This hierarchical identification transforms a complex multi-objective decision-making problem into a structured problem with a clear order of priorities.

[0028] Based on clearly defined primary and secondary optimization objectives, the system can select from a wide range of adjustable operating parameters those key parameter sets that are most sensitive to and have the most significant impact on achieving the current core objective. For example, when the primary objective is to maximize PV absorption, parameters such as the maximum charging power limit of the energy storage unit and the activation threshold of the adjustable load become key parameters. When the primary objective is to optimize grid costs, parameters such as the threshold at which time-of-use electricity prices trigger energy storage operation and the planned value of power exchange with the grid become crucial. This screening process relies on a deep understanding of the system's operating mechanisms and may also require the assistance of tools such as the sensitivity analysis described above. After determining the key parameter set, a scheduling impact quantitative analysis is performed on it. This analysis aims to assess the direction and extent of the impact of these key parameters on the primary and secondary optimization objectives when they vary within their allowable ranges. For example, the analysis can be used to determine how much a one-unit increase in the maximum charging power of the energy storage unit would increase PV absorption, while also determining the negative impact on the energy storage lifespan degradation indicator. This quantitative analysis provides a basis for subsequent decision-making on how to adjust these parameters during the optimization process, eliminating the need for blind search and instead enabling targeted exploration guided by prior knowledge. The entire implementation process reflects the progressive transformation from data to features, from features to constraints, from constraints to models, and then from models to strategies, ultimately generating an initial scheduling plan that conforms to the physical operating laws of the system and is close to actual business needs.

[0029] Example 3: See Figure 4 The implementation involves conducting an in-depth quantitative analysis of the scheduling impact of a set of key control parameters. Based on this analysis, an updated energy storage scheduling model is output through an iterative optimization process. This process begins by accurately extracting the permissible adjustment range for each parameter in the set of key control parameters from the constructed steady-state characteristic matrix. This range is not simply a physical limit; rather, it comprehensively considers the safe operating range specified by the equipment manufacturer, the current actual operating conditions of the system, and the reliable operating boundaries demonstrated in historical operations. For example, for a battery energy storage unit, the adjustable range of its maximum allowable charging power parameter may have a lower limit set at the current minimum charging power allowed (to avoid potential damage to the battery caused by low-current charging), and an upper limit strictly limited to the maximum safe charging power calculated in real time by the battery management system. This value may dynamically adjust with changes in the battery's state of charge, temperature, and internal resistance. Each extracted adjustable range is represented as an interval, defining clear boundaries in the solution space for subsequent optimization searches.

[0030] Parameter sensitivity mapping for the primary and secondary optimization objectives is a core component of quantitative analysis. This mapping aims to reveal how small changes in each key tuning parameter affect the values ​​of each optimization objective function. A local sensitivity analysis method based on partial derivatives is employed in this implementation. For each objective function and each key parameter, a small perturbation step is defined near the parameter's current operating point, and the response ratio of the objective function value to the parameter change is calculated. This response ratio forms the elements of the objective impact quantification gradient matrix, whose rows correspond to different optimization objectives and columns to different key tuning parameters. The magnitude of the gradient indicates the strength of the parameter's influence on the objective, while its sign indicates the direction of the influence. A positive gradient value means that increasing the parameter's value will lead to an improvement in the corresponding objective function value (or a deterioration for minimization objectives), and vice versa.

[0031] Another key task is to construct an interaction matrix between all flexible regulation objectives. This matrix aims to capture the complex interrelationships between multiple objectives. Each element of the matrix qualitatively or semi-quantitatively describes the cascading effects of optimizing one objective on another. This relationship is determined primarily through two methods: first, by analyzing the inherent mathematical connections between objective functions based on the derivation of the system's physical model; and second, by statistically analyzing the correlations between objective values ​​under different scenarios through simulation calculations of a large number of design scenarios. The interaction matrix clearly identifies which pairs of objectives have synergistic relationships, meaning that optimizing one objective also promotes the improvement of another, and which pairs of objectives have conflicting relationships, meaning that improving one objective comes at the expense of another. For example, the objective of maximizing photovoltaic absorption and the objective of optimizing grid interaction costs may be synergistic during the daytime electricity price valley but conflicting during the daytime electricity price peak.

[0032] Based on the obtained target impact quantification gradient and interaction impact relationship matrix, an indicator that comprehensively reflects the sensitivity of key adjustment parameters can be calculated, namely the sensitivity eigenvalue. The calculation of this value aims to comprehensively consider the direct impact of a parameter on multiple targets and the indirect impact caused by conflicts or synergies between targets. A feasible comprehensive calculation method is as follows:

[0033] in: Indicates the The larger the sensitivity characteristic value of a key adjustment parameter, the greater the impact of the parameter on the overall target set, and more attention should be paid to it during the optimization process. Indicates the total number of flexible adjustment targets. Indicates the The weight coefficients assigned to the optimization objectives after hierarchical identification, the weight of the main optimization objective is usually higher than that of the secondary optimization objective. Indicates the objective function Relative to the Key parameters The partial derivative of , that is, the corresponding element in the target influence quantization gradient matrix. is a reference factor used to characterize the The net interaction strength between a goal and other goals will be calculated with reference to the interaction influence relationship matrix. If the goal and other main goals are mostly synergistic, this factor may be less than 1 to weaken their influence; if they are mostly conflicting, this factor may be greater than 1 to enhance their influence, thereby reflecting the amplification effect brought about by the goal conflict in the comprehensive sensitivity.

[0034] Based on the calculated sensitivity eigenvalues ​​of each key parameter, a parameter optimization constraint set can be established. This constraint set not only includes the previously extracted physical allowable range, but may also introduce dynamic constraints based on sensitivity analysis. For example, for parameters with extremely high sensitivity eigenvalues, the optimization step size may be limited to a smaller size to avoid the optimization process being too intense and causing system oscillations; or a more conservative operating boundary may be set for it to reduce operational risks. Conversely, for numbers with lower sensitivity eigenvalues, a bolder search within their range can be allowed. The parameter optimization constraint set provides important rule guidance for the subsequent iterative optimization process.

[0035] Outputting the updated energy storage dispatch model involves a dynamic, iterative optimization process. Based on the parameter optimization constraint set, clear control boundaries are set for the key adjustment parameter set. All subsequent searches are conducted within the feasible region defined by these boundaries. Based on the system's current real-time operating state, an initial solution space is created. This space contains multiple candidate solutions, either random or generated based on the current operating point. Each solution represents a specific value assignment for a set of key parameters. A fitness evaluation is performed on each candidate solution in the initial solution space. The fitness function is typically a scalar function that comprehensively considers all hierarchical objectives (and their weights) and constraint violation penalties. Its value directly reflects the quality of the parameter solution. Combining the parameter optimization constraint set and the fitness evaluation results, the optimization algorithm generates an optimization direction vector, which indicates which parameter adjustment direction within the current solution space is most likely to improve fitness, as well as a gradient optimization step size, which suggests the magnitude of the parameter adjustment.

[0036] Based on the optimization direction vector and gradient optimization step size, the optimization algorithm iteratively updates the initial solution space. This process usually uses gradient descent algorithms, evolutionary algorithms, or other meta-heuristic algorithms. In each round of iteration, the algorithm generates a new generation of candidate solution groups based on the fitness information and direction guidance of the current group. The new generation of solution groups may include existing solutions fine-tuned along the gradient direction, new solutions generated by random combinations, and random exploration solutions introduced to maintain diversity. The iterative process will continue until the preset termination conditions are met, such as reaching the maximum number of iterations, the fitness improvement is no longer obvious, or a solution that meets all requirements is found. The final iterative update result will output one or more candidate solutions with the best performance. The solution with the highest fitness will be selected, and its corresponding key parameter combination scheme will be embedded in the scheduling strategy, thereby forming and outputting the final updated energy storage scheduling model to guide the actual operation of the system.

[0037] Example 4: See Figure 5 The core of the implementation method is to identify operational anomalies in real time based on status indication data and trigger a multi-node collaborative rescheduling strategy. The operation of this process depends on the continuous monitoring and analysis of the transient feature matrix. As a quantitative record of the deviation of the system operation status from the "baseline", the transient feature matrix has dynamic changes in internal data and contains a variety of abnormal indication signals such as power limit signs and state mutation rates. The system operation monitoring module scans the matrix at a very high frequency, focusing on those data entries marked as abnormal by preset rules. Each abnormal mark is like an alarm signal and needs to be immediately associated with its source. The association operation first locates the specific node identifier and device type corresponding to the abnormal entry, and then extracts the deep feature data related to the abnormal event from the matrix or parallel database. These characteristic data include but are not limited to: node operation deviation, that is, the absolute or relative deviation between the current actual operating parameter value of the node and the baseline value predicted in the steady-state characteristic matrix, which directly reflects the degree of abnormality; energy storage decay rate, targeting the abnormality of energy storage equipment, calculating the abnormal change trend of its charge state or health indicators in the recent time window, and the high or low rate value indicates the urgency of the performance degradation or failure of the energy storage equipment; irradiation mutation intensity, targeting the abnormality related to photovoltaic equipment, quantifying the change amplitude of ambient irradiance in a short period of time, and the size of the intensity value reflects the severity of the change in lighting conditions, which is directly related to the sudden drop or surge in photovoltaic output.

[0038] After successfully correlating and extracting anomaly features from individual nodes, the analysis perspective needs to expand from individual nodes to the overall system, aiming to discover cross-node anomaly correlation patterns. Cross-node anomaly correlation features are generated by analyzing the temporal proximity, spatial proximity, and similarity of anomaly types between anomaly events at different nodes. The system calculates the time difference between anomaly events to determine whether they occur nearly simultaneously; examines the electrical distance between nodes in the network topology to determine their physical proximity; and compares the type feature vectors of each anomaly event to determine whether their root cause is likely to be the same. For example, if multiple geographically close photovoltaic nodes report output fluctuations caused by sudden irradiation changes at approximately the same time, and the fluctuations are consistent (all showing a sudden drop), the system will identify a strong cross-node correlation feature, suggesting that the area may be experiencing cloud cover. The strength of this correlation feature is quantified using a composite metric that takes into account the number of nodes with co-occurring anomalies, a weighted average of the anomaly severity, and the closeness of spatiotemporal coupling.

[0039] When the calculated cross-node anomaly correlation strength exceeds a preset threshold based on historical experience and system safety regulations, the system determines that the anomaly is no longer an isolated local event, but rather a correlated fault or disturbance that could impact the overall system operation. At this point, a multi-node coordinated rescheduling strategy is automatically triggered. The core task of this strategy is to dynamically adjust the currently executing or newly established initial energy storage scheduling model based on the latest transient characteristic matrix containing anomaly information. The rescheduling process first assesses the impact of the anomaly on the current power balance and energy storage plan, then rapidly calculates and outputs an updated power allocation plan and energy storage charge and discharge timing strategy. For example, in the face of the aforementioned regional PV output drop, the rescheduling strategy might instruct energy storage units in unaffected areas to discharge earlier or more aggressively to compensate for the sudden drop in power generation. It might also moderately increase the power ceiling for power purchased from the grid and recalculate the power support allocation ratios among nodes.

[0040] After triggering the rescheduling strategy, for more precise coordinated adjustments, detailed information about all flagged anomalies must be systematically extracted from the transient feature matrix. This information is structured into three core dimensions: the unique ID of the anomaly node, which clearly indicates which node has deviated from normal operation; the specific time period of the anomaly, which precisely defines the time window when the disturbance began and lasted; and the anomaly type feature vector, a data set containing multiple attributes that characterizes the anomaly, such as the anomaly code, deviation value, and trend. This feature vector construction transforms anomalies into more than simple Boolean flags, instead imbuing them with rich, quantifiable, and categorizable information. Assigning node coupling weight coefficients to these anomaly feature vectors is a key step in achieving effective coordinated rescheduling. The coupling weight coefficient quantifies the degree of mutual operational influence between any two nodes. This coefficient is not simply based on physical distance, but rather takes into account electrical connection impedance, historical power interaction data, and control logic dependencies. Nodes with close electrical connections and frequent power interaction will have a higher coupling weight coefficient; conversely, nodes with loose connections or little direct interaction will have a lower coupling weight coefficient. The weight coefficients form a matrix that describes the coupling relationship between all nodes.

[0041] Based on the node coupling weight coefficients, the system performs a cluster analysis of scheduling-related characteristics for all groups of nodes experiencing anomalies. Clustering algorithms (such as hierarchical clustering or K-means variants) group anomaly nodes based on the similarity of their anomalies and the strength of their coupling relationships. This analysis typically yields two representative anomaly clusters. The first anomaly cluster consists of nodes whose anomaly characteristics consistently point to PV output fluctuations caused by external environmental changes, such as sudden irradiance changes leading to a sharp drop or rise in output. The second anomaly cluster consists of nodes whose anomaly characteristics primarily manifest as unexpected sudden changes in load demand (such as the unplanned start and stop of large equipment) or signs of failure in the energy storage system itself (such as an abnormal increase in internal resistance or a sharp decrease in available capacity). These two anomaly clusters are distinguished not by subjective judgment but by a calculated quantitative metric—node scheduling deviation. This deviation measures the overall deviation between the node's actual operating point and the expected operating point in the original scheduling plan at the time of the anomaly. Dispatch deviations caused by PV fluctuations typically manifest as unplanned power shortages or surpluses; whereas those caused by sudden load changes or energy storage failures often manifest as unplanned power demand or loss of supply capacity. This fundamental difference enables clustering algorithms to effectively distinguish between them (see Table 1).

[0042] Table 1: Abnormal events and node coupling weights.

[0043]

[0044] Table 1 shows the coupling weights between abnormal events and nodes. The data in this table indicates that around 10:05, a sudden drop in output occurred at photovoltaic node PV-Node-05. This sudden drop in output had a high coupling weight with energy storage node ESS-Node-02, indicating that the energy storage node was a critical source of power support. Another abnormality occurred later: an abnormal increase in internal resistance at energy storage node ESS-Node-08 and a sudden change at load node Load-Node-03 occurred almost simultaneously. Their high coupling weight strongly suggests that they may belong to the same fault event cluster, requiring coordinated response. The entire implementation process represents a complete closed loop, from anomaly perception, feature correlation, global assessment, to coordinated response. This ensures that the system can respond quickly and coordinatedly to local disturbances, maintaining overall stable and efficient operation.

[0045] Example 5: The implementation focuses on the management and adaptive adjustment of the optimization iterative process. Its core lies in the refined classification and guidance of solution space search behavior. This process begins by establishing a separate iteration path identifier for each candidate solution generated during the iterative optimization process. This identifier is not a simple serial number, but a structured record that continuously tracks and records the update history of the solution across iterations. This record includes the solution's fitness value in each generation, the direction and magnitude of change in its parameter combination compared to the previous generation, and the method by which it was generated (for example, through crossover, mutation, or gradient guidance). This historical trajectory allows for dynamic classification of the state of the search path represented by the solution. Classification is primarily based on observing the trend and stability of the fitness value over multiple iterations. If a path's fitness value shows a stable and significant trend toward a better value over multiple iterations, its state is classified as convergent. If the fitness value fluctuates within a certain range, without a clear and sustained direction of improvement, but also without significant performance degradation, its state is classified as exploration. If the fitness value continues to deteriorate over multiple iterations, or its parameter values ​​gradually deviate from the boundaries of the feasible region, its state is classified as divergent.

[0046] Within a preset iterative evaluation window, the path update status of all active solutions is systematically identified and statistically analyzed. This evaluation window is typically set to a fixed number of iterations; for example, after every ten iterations, a comprehensive evaluation of the path status of all solutions in the current solution population is performed. The purpose of this evaluation is not simply to label each path, but more importantly, to generate a holistic understanding of the current search landscape across the entire solution space, known as the gradient optimization morphology classification. This classification describes the stage and characteristics of the current optimization process at a macro level. The dominance of convergent morphologies may indicate that the algorithm is conducting a local, detailed search within a promising region. The prevalence of exploratory morphologies may indicate that the algorithm is in a global exploration phase, extensively evaluating the potential of different regions. The frequent occurrence of divergent morphologies may be a warning sign, indicating that the current search strategy may be flawed or that the algorithm is entering a disadvantageous region.

[0047] Based on gradient optimization morphology classification, the system performs solution space search optimization management, a highly adaptive strategy allocation mechanism. For search paths identified as converging, the system configures a gradient proxy model. This proxy model is a lightweight, computationally efficient mathematical model that learns and fits the fitness function response characteristics of the region near the path to predict the potential effects of further parameter adjustments. Essentially, it uses local models to predict optimization trends, avoiding time-consuming full fitness evaluations for each fine-tuning step and enabling rapid refinement within promising regions. For search paths identified as diverging, the system adopts a more cautious strategy, configuring an optimization taboo window. This window records parameter movement directions and associated parameter regions that cause the path's fitness to deteriorate. For subsequent iterations, updates to the path are prohibited from exploring these directions and regions marked as invalid or even harmful, forcing the path to redirect or steer away from unfavorable search areas.

[0048] A hybrid strategy is adopted for the exploration morphological paths, which are generally the most numerous and have not yet shown a clear convergence or divergence trend, and thus have the potential to explore new areas. Their iterative updates combine multiple sources of information, including the macro-optimization trend direction extracted from the gradient surrogate model of the convergent path (as a heuristic guide), invalid directions excluded from the tabu window of the divergent path (to avoid repeating the same mistakes), and a certain proportion of random exploration components to maintain the diversity of the population and the global exploration ability.

[0049] This classification management mechanism ultimately translates into differentiated iterative update strategies for solutions of different types. For solutions in a convergent state, refined iterations are performed. This means using a small optimization step size and making small, precise parameter adjustments strictly along the local optimization trend predicted by the gradient surrogate model. This aims to fully exploit the potential of the current local optimal solution and achieve the ultimate performance improvement. For solutions in an exploratory state, mixed-direction iterations are performed. The parameter update vector is a composite of multiple directions, potentially based partly on attracting the global optimal solution, partly on random perturbations, and partly on learning from other successful paths. This aims to balance the algorithm's exploration and exploitation capabilities. For solutions in a divergent state, random restart iterations are performed. This is a more aggressive strategy. When a path is confirmed to be continuously diverging and the taboo strategy is ineffective, the solution is not fine-tuned in the current region. Instead, its parameters are randomly reset to a new region of the solution space that has not been fully explored and is not marked as taboo, and the evaluation begins from scratch. This helps the algorithm escape local optima or invalid regions and revitalize the global search. Through this refined management of different strategies for different search states, the efficiency and effectiveness of the entire optimization process are significantly improved, enabling faster approaches to high-quality solutions and effectively avoiding premature convergence or invalid searches, thereby providing a guarantee for outputting a high-performance updated energy storage scheduling model.

Claims

1. A method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model, characterized in that: include: Acquire real-time operating data of each node in the PV-storage-direct-flexible system and perform multi-dimensional alignment processing on the real-time operating data to form a multi-dimensional operating data set. The real-time operating data includes the PV output fluctuation curve, energy storage charge state, load demand timing, grid interaction power limit, and ambient irradiance variation characteristics; Performing operation status mapping topology construction processing on the multi-dimensional operation data set to separate target operation data for scheduling model construction and status indication data for scheduling abnormal response; Based on the target operation data and in combination with the scheduling priority configuration information of each node, an initial energy storage scheduling model for multi-node collaboration is generated, wherein the initial energy storage scheduling model includes an energy storage charging and discharging timing strategy and a power allocation scheme; Based on the status indication data, abnormal node operation characteristics during the real-time operation data update process are identified, the energy storage scheduling initial model is dynamically reconstructed according to the abnormal node operation characteristics, and an updated energy storage scheduling model is output.

2. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 1 is characterized in that: The performing operation status mapping topology construction processing on the multi-dimensional operation data set includes: Convert the multi-dimensional operation data set after aligning the time, node and device dimensions into a scheduling state mapping matrix; Decoupling the state characteristics of the scheduling state mapping matrix to obtain a steady-state characteristic matrix and a transient characteristic matrix, wherein the steady-state characteristic matrix represents the normalized basic structure of the scheduling model, and the transient characteristic matrix represents the abnormal state characteristics corresponding to power limit exceeding, energy storage attenuation, and radiation mutation; The steady-state characteristic matrix is ​​used as the target operating data, and the transient characteristic matrix is ​​used as the state indication data.

3. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 2, characterized in that: The generating of the initial model for multi-node coordinated energy storage scheduling includes: Extracting the photovoltaic output prediction segment, energy storage charge state interval, load demand peak and valley characteristics, and grid interaction constraint boundaries from the steady-state characteristic matrix; Configuring an optimized weight coefficient for each node based on the scheduling priority configuration information; According to the optimization weight coefficient, a multi-objective fusion calculation is performed on the photovoltaic output forecast segment, the energy storage charge state interval, the load demand peak and valley characteristics and the grid interaction constraint boundary to generate the energy storage charging and discharging timing strategy and power allocation plan.

4. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 3 is characterized in that: The generating of the initial model for multi-node coordinated energy storage scheduling further includes: Configure flexible regulation targets based on the business attributes of the PV-storage direct-flexible system, including maximizing PV absorption, balancing energy storage life, peak-load shaving and valley-filling, and optimizing grid interaction costs; Performing hierarchical identification processing on the flexible adjustment target to determine the primary optimization target, the secondary optimization target and the boundary constraint target; A key adjustment parameter set is selected based on the primary optimization objective and the secondary optimization objective, and a scheduling impact quantitative analysis of the key adjustment parameter set is performed.

5. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 4 is characterized in that: The performing of the scheduling impact quantitative analysis of the key adjustment parameter set includes: extracting an adjustable parameter range from the steady-state characteristic matrix; Performing parameter sensitivity mapping on the primary optimization objective and the secondary optimization objective to generate a target impact quantization gradient; Constructing an interactive influence relationship matrix of all target items in the flexible adjustment target, wherein the interactive influence relationship matrix includes target coordination relationships and target conflict relationships; Calculating sensitivity eigenvalues ​​of the adjustable parameter set based on the target impact quantization gradient and the interaction impact relationship matrix; A parameter optimization constraint set is established according to the sensitivity eigenvalues.

6. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 5, characterized in that: The output updated energy storage scheduling model includes: Setting control boundaries for the key adjustment parameter set and creating an initial solution space based on the current operating state; Performing fitness evaluation on the solutions in the initial solution space, and generating an optimization direction vector and a gradient optimization step size in combination with the parameter optimization constraint set and the fitness evaluation result; Iteratively updating the initial solution space based on the optimization direction vector and the gradient optimization step size; The updated energy storage scheduling model is output according to the iterative update result.

7. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 6, characterized in that: The identifying, based on the status indication data, abnormal node operation characteristics during the real-time operation data update process includes: Monitoring abnormal marking entries of the transient characteristic matrix, and associating the node operation deviation, energy storage attenuation rate, and radiation mutation intensity corresponding to the abnormal marking entries; Generate cross-node abnormal correlation features based on the node operation deviation, energy storage decay rate and radiation mutation intensity; When the cross-node abnormal correlation feature exceeds a preset threshold, a multi-node collaborative rescheduling strategy is triggered, and the power allocation plan and energy storage charging and discharging timing strategy are updated according to the transient feature matrix.

8. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 7, characterized in that: The triggering multi-node collaborative rescheduling strategy includes: Extracting the node number, abnormal time period and abnormal type feature vector of the abnormality marked in the transient feature matrix; configuring a node coupling weight coefficient for the eigenvector; Performing cluster analysis on the scheduling correlation characteristics of the abnormal node group according to the node coupling weight coefficient to generate a first abnormal cluster and a second abnormal cluster; The first abnormal cluster represents a scheduling anomaly caused by PV output fluctuations, and the second abnormal cluster represents a scheduling anomaly caused by load mutations or energy storage failures. The first abnormal cluster and the second abnormal cluster are distinguished by the node scheduling deviation.

9. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 8, characterized in that: The iterative updating of the initial solution space based on the optimization direction vector and the gradient optimization step size includes: Establish an iterative path identification for each solution, and classify the path status according to the fitness value of each iteration; Identifying the path update state within a preset iterative evaluation window and generating a gradient optimization morphology classification, wherein the gradient optimization morphology classification includes a convergent morphology, an exploratory morphology, and a divergent morphology; A solution space search optimization management is performed based on the gradient optimized morphological classification.

10. The method for establishing a multi-node solar-storage direct-flexible system energy storage scheduling model according to claim 9, characterized in that: The performing solution space search optimization management based on the gradient optimization morphological classification includes: A gradient proxy model is configured for the convergence pattern, and the optimization trend direction is predicted by the gradient proxy model; Configuring an optimized tabu window for the divergent morphology, and eliminating invalid search directions through the optimized tabu window; Based on the optimization trend direction and the optimization taboo window, a refined iteration is performed on the convergent morphological solution, a mixed direction iteration is performed on the exploratory morphological solution, and a random restart iteration is performed on the divergent morphological solution.

Images