Wind, light and water storage integrated scheduling method and system based on big data

Abstract

The invention provides a wind-light-water storage integrated scheduling method and system based on big data, and relates to the technical field of energy management, and the method comprises the steps: extracting a driving factor of wind-light-water storage data based on a generalized dynamic factor model, and carrying out the dynamic modeling of the driving factor, and obtaining an output state feature graph of a wind power plant and a photovoltaic power station; constructing an edge dependence structure based on a Copula function to obtain a wind-light-water combined space structure diagram; generating an energy storage response strategy sub-graph according to the wind-light-water combined space structure graph, the charge state of the energy storage system and the device degradation parameters; according to the energy storage response strategy subgraph and the load data and fluctuation data of the regional load center, singular spectrum analysis is carried out to extract periodic load characteristics, and a multi-energy output path field is obtained; and performing fuzzy approximate reasoning on the path field to obtain an optimal joint scheduling sequence. According to the method, the whole-process closed-loop optimization of the dispatching of the wind-light-water storage system is realized, and the method has the advantages of high prediction precision, high response speed and high dispatching flexibility.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management. Specifically, it relates to a comprehensive scheduling method and system for wind-solar-hydro-storage based on big data. Background Technique

[0002] With the rapid development of renewable energy, wind power, photovoltaic power, hydropower and energy storage systems have gradually built a multi-energy complementary integrated energy network, becoming an important part of building a new power system. However, there are many uncertainties and coupled non-linear problems in the operation and scheduling process of the wind-solar-hydro-storage system: wind speed and light have high volatility, hydropower regulation depends on hydrological conditions, and the energy storage system is limited by the state of charge and device life. These factors together lead to an increase in the difficulty of joint scheduling of multi-energy systems. Traditional scheduling methods usually adopt static optimization or independent control strategies based on linear prediction. For example, linear programming, dynamic programming or empirical rules are used to schedule each energy subsystem separately. Such methods are difficult to cope with the rapid fluctuations of wind and light resources and the heterogeneity of multi-source data, lack a unified modeling mechanism and cross-domain collaboration ability. Especially in the scenario of high proportion of new energy penetration, it often leads to lagging response of scheduling strategies, decline in system stability and insufficient efficiency of energy storage resource allocation.

[0003] Therefore, there is an urgent need for an advanced method that can fully integrate big data-driven, deep learning modeling and system collaborative control to improve the prediction accuracy, response flexibility and operation stability of the scheduling system. Summary of the Invention

[0004] The purpose of the present invention is to provide a comprehensive scheduling method and system for wind-solar-hydro-storage based on big data to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows: In the first aspect, the present application provides a comprehensive scheduling method for wind-solar-hydro-storage based on big data, including: Obtain wind-solar-hydro-storage data, the state of charge of the energy storage system, device degradation parameters, load data and fluctuation data of the regional load center; Extract the driving factors in the wind farm and photovoltaic power station data from the wind-solar-hydro-storage data based on the generalized dynamic factor model, and perform dynamic modeling on the driving factors to obtain the output state characteristic map of the wind farm and photovoltaic power station; Based on the Copula function, construct an edge dependence structure for the output state characteristic map and the hydropower data in the wind-solar-hydro-storage data, and input this structure into the Markov random field model to obtain the wind-solar joint space structure map; According to the wind-solar joint space structure map, the state of charge of the energy storage system and the device degradation parameters, perform distributed tensor decomposition to construct an energy tensor representation, and then embed the energy tensor representation into a dynamic game graph to generate an energy storage response strategy sub-graph; According to the energy storage response strategy sub - graph, the load data and fluctuation data of the regional load center, perform singular spectrum analysis to extract periodic load characteristics, and then input into a spiking neural network to predict the power consumption sequence, obtaining a multi - energy output path field; Perform fuzzy approximate reasoning on the path field to extract stability rules, and input them into a probabilistic context - free grammar model to generate a dynamic scheduling strategy, obtaining an optimal joint scheduling sequence.

[0005] In a second aspect, the present application also provides a wind - solar - hydro - energy - storage integrated scheduling system based on big data, including: An acquisition unit, configured to acquire wind - solar - hydro - energy - storage data, the state of charge of the energy storage system, device degradation parameters, load data and fluctuation data of the regional load center; An extraction unit, configured to extract driving factors in the wind farm and photovoltaic power plant data from the wind - solar - hydro - energy - storage data based on a generalized dynamic factor model, and perform dynamic modeling on the driving factors to obtain an output state feature map of the wind farm and photovoltaic power plant; A construction unit, configured to construct an edge - dependence structure for the output state feature map and the hydro - power data in the wind - solar - hydro - energy - storage data based on the Copula function, and input this structure into a Markov random field model to obtain a wind - solar joint spatial structure map; A decomposition unit, configured to perform distributed tensor decomposition according to the wind - solar joint spatial structure map, the state of charge of the energy storage system and device degradation parameters to construct an energy tensor representation, and then embed the energy tensor representation into a dynamic game graph to generate an energy storage response strategy sub - graph; An analysis unit, configured to perform singular spectrum analysis according to the energy storage response strategy sub - graph, the load data and fluctuation data of the regional load center to extract periodic load characteristics, and then input into a spiking neural network to predict the power consumption sequence, obtaining a multi - energy output path field; A scheduling unit, configured to perform fuzzy approximate reasoning on the path field to extract stability rules, and input them into a probabilistic context - free grammar model to generate a dynamic scheduling strategy, obtaining an optimal joint scheduling sequence.

[0006] The beneficial effects of the present invention are as follows: Aiming at the problems of weak modeling ability, low coupling degree and poor stability of the existing methods, the present invention constructs an intelligent scheduling mechanism with multi - stage and multi - model collaborative fusion. It not only improves the modeling accuracy of the operating characteristics of the multi - source energy system, but also enhances the real - time performance and stability of the scheduling strategy, providing theoretical support and engineering paths for constructing a new type of power system with multi - energy integration and high - efficiency coordination. And the present invention realizes the full - process closed - loop optimization of "driving modeling - probabilistic reasoning - low - rank extraction - dynamic game - load prediction - intelligent strategy generation" for the scheduling of the wind - solar - hydro - energy - storage system, and has technical advantages such as high prediction accuracy, strong modeling ability, fast response speed, low operation risk and high scheduling flexibility.

[0007] Other features and advantages of the present invention will be described in the subsequent specification, and in part, will become apparent from the specification, or will be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative efforts.

[0009] Figure 1 It is a schematic flowchart of the integrated dispatching method for wind, light, water, and energy storage based on big data described in the embodiments of the present invention; Figure 2 It is a schematic structural diagram of the integrated dispatching system for wind, light, water, and energy storage based on big data described in the embodiments of the present invention.

[0010] In the figure: 701, acquisition unit; 702, extraction unit; 703, construction unit; 704, decomposition unit; 705, analysis unit; 706, dispatching unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present invention provided in the following drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0012] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0013] Embodiment 1: This embodiment provides a comprehensive scheduling method for wind, light, water, and storage based on big data.

[0014] See Figure 1 , the figure shows that this method includes steps S1, S2, S3, S4, S5, and S6.

[0015] Step S1: Obtain wind, light, water, and storage data, the state of charge of the energy storage system, device degradation parameters, load data and fluctuation data of the regional load center; It can be understood that in this step, it is first necessary to comprehensively perceive and collect data on various energy systems of wind, light, water, and storage and their operating environments to form a big data foundation of multi-source heterogeneity. Among them, the data of the wind power and photovoltaic parts not only include the conventional output power, but also include the original meteorological driving data such as wind speed, wind direction, solar irradiance, and temperature, etc., so as to extract driving factors and perform dynamic modeling in the follow-up; the data of the hydropower part covers reservoir water level, incoming water volume, historical regulation curves, etc., to depict the dispatchable boundary of water energy. At the same time, the state of charge (SOC) of the energy storage system is an important input for dynamic energy balance in the scheduling process. The voltage, current, and capacity evolution curves are obtained through high-frequency sampling by the BMS (battery management system), and the device degradation parameters are estimated in combination with the operating cycle, number of cycles, and temperature degradation characteristics to ensure the effectiveness of subsequent response strategies under the consideration of aging constraints. In addition, the load data of the regional load center covers the real-time load curve of the whole day, and its fluctuation characteristics such as peak load, low valley period, and load elasticity are extracted through historical data training, which are used to guide the formulation of load-side regulation strategies. This step not only significantly enhances the system's perception ability of the energy state and environmental dynamics, but also provides multi-dimensional constraints and state references for the subsequent complex modeling process, thereby improving the accuracy and practicality of the overall scheduling model.

[0016] Step S2: Extract the driving factors in the wind farm and photovoltaic power station data from the wind, light, water, and storage data based on the generalized dynamic factor model, and perform dynamic modeling on the driving factors to obtain the output state characteristic diagrams of the wind farm and photovoltaic power station; It can be understood that in this step, by fusing statistical dimensionality reduction and non-linear dynamic modeling, the time-varying driving characteristics of wind and light resources are effectively extracted and represented, providing a structured and computable state expression framework for subsequent wind-hydro coupling modeling and energy storage response strategy generation, thereby significantly improving the modeling accuracy and dynamic tracking ability of the system for volatile renewable energy. In this step, step S2 includes steps S21, S22, and S23.

[0017] Step S21: Perform generalized dynamic factor model processing according to the wind speed sequence of the wind farm and the irradiance time series data of the photovoltaic power station. Among them, the main factor sequence representing the coupled driving of wind and light resources is obtained by maximizing the eigenvalue decomposition of the covariance matrix to extract the co-varying factors of the driving system; It is understandable that this step processes the time series data of the main meteorological driving factors on which wind power and photovoltaic systems depend, namely wind speed and solar irradiance, using a generalized dynamic factor model. The purpose is to extract the most representative co-variation structure from the high-dimensional and noise-complex original sequence. Specifically, the generalized dynamic factor model constructs a covariance matrix with time-delay characteristics and performs eigenvalue decomposition on this matrix, maximizing the variance explained by the eigenvectors corresponding to the first few eigenvalues, thereby extracting the low-dimensional latent factors that dominate the changes in wind and light resources. The generalized dynamic factor model can capture the co-evolution characteristics of variables with cross-time lags, so it shows stronger modeling ability when dealing with time-varying trends such as the periodic fluctuations of wind speed and the morning and evening changes in irradiance. The finally generated main factor sequence not only retains the main dynamic change trends of the original meteorological data in time but also reflects the common coupling relationship of wind and light resources among different stations in space, providing a key driving basis for subsequent non-linear interval division and state transition modeling.

[0018] Step S22: Perform threshold vector autoregressive modeling based on the main factor sequence. By setting the wind speed volatility and light mutation rate as threshold variables to divide the non-linear state intervals, obtain the interval-dependent dynamic expression of wind and light output; It is understandable that this step constructs a threshold vector autoregressive model based on the main factor sequence extracted in the previous stage to deeply explore the non-linear dynamic evolution characteristics of wind and light resources under different perturbation situations. In this step, the wind speed volatility and light mutation rate are set as threshold variables to identify and divide the different state intervals in which wind and light output are located, such as the stable operation area, the fluctuation area, or the extreme perturbation area, thereby establishing a segmented and state-dependent regression structure. Specifically, the threshold vector autoregressive model fits an independent vector autoregressive process within each threshold interval, making the response behavior of wind and light output to historical factors change significantly as the state interval switches. This interval division mechanism effectively captures the common non-linear jump, mutation, and other dynamic characteristics in the operation of the wind and light system. At the same time, the model has an adaptive structure selection ability and can determine the optimal threshold point in a sample-driven manner, thus avoiding the model bias caused by artificial setting. The interval-dependent expression of wind and light output obtained in this way can not only retain the global trend information but also reflect the sensitivity and response inertia of the system to perturbations in the local state.

[0019] Step S23: Perform state embedding graph construction based on this dynamic expression. By using the state transition direction of time series nodes within the threshold interval to generate a directed graph structure, obtain the wind and light output state characteristic graph containing gated connections and non-linear interaction characteristics.

[0020] It can be understood that in this step, the state transition relationship in the time series is further mapped into a graph structure to construct the state feature graph of the wind and light output. Among them, by analyzing the state transition direction of each node in the time series (i.e., the wind and light output state at a certain moment) within the threshold interval, a directed graph reflecting the time evolution path is generated. In the specific implementation, the nodes represent the state performances in different nonlinear intervals at different time points, the direction of the edges represents the evolution trend from the current state to the next state, and the weights of the edges are characterized based on dynamic features such as the state change rate and the fluctuation amplitude. In addition, in order to fully reflect the regulation effect of the threshold mechanism on the system state, gated connections are introduced in the graph structure to represent the discontinuous response experienced by the system when jumping across intervals, and at the same time, the potential interaction between wind speed and light is retained through the design of the nonlinear coupling term between nodes. This graph structure not only has the unified expression ability of time causality and state responsiveness, but also can further identify the key state paths and fluctuation propagation paths through graph theory analysis methods.

[0021] Step S3: Based on the Copula function, construct an edge dependence structure for the output state feature graph and the hydropower data in the wind, light, water, and storage data, and input this structure into the Markov random field model to obtain the joint space structure graph of wind, light, and water; It can be understood that this step realizes the high-dimensional scalable modeling of the complex spatial correlation and the coupling relationship of the regulation behaviors between multiple energy sources, significantly enhances the representation ability of the dispatching system for the uncertainty conduction path and the multi-source resource coordination mechanism, and provides an accurate spatial dependence background for the construction of the energy tensor and the strategy game. In this step, step S3 includes step S31, step S32, step S33, and step S34.

[0022] Step S31: Perform edge distribution extraction processing according to the output state feature graph and the water level-output regulation data of the hydropower station. Among them, by using kernel density estimation to extract the edge probability distributions of each variable in different operating states, a multi-source edge probability set of the wind, light, and water system is obtained; It can be understood that this step will coordinate the constructed wind power and photovoltaic output state characteristic diagrams with the regulatory operation data of the hydropower station (such as the relationship between water level changes and corresponding output capacity), and then explore the marginal probability distribution of each energy unit under different operating conditions. This process uses a non-parametric kernel density estimation method to model each key variable. Kernel density estimation does not require the distribution type to be set in advance. By sliding the kernel function (such as a Gaussian kernel) on the observed data to form a smooth probability curve, it can more accurately capture the distribution trend of the variable in each operating range. It is particularly suitable for natural resource variables such as wind speed, irradiance, and water level that have non-Gaussian characteristics and obvious fluctuations. In this step, the wind power output state characteristic diagram provides a state sequence of wind speed and irradiance based on threshold division. Under the guidance of these state nodes, the system can extract the marginal probability of wind speed and irradiance under specific conditions in different partitions; at the same time, the regulation data of the hydropower station provides a mapping curve of water level to output capacity, and the marginal distribution under different water storage strategies is also obtained through the kernel density estimation method. Finally, a joint variable set including wind speed, irradiance, water level and their corresponding output is formed, and the marginal probability expression of each variable in its independent dimension is established, collectively referred to as the multi-source marginal probability set of wind-solar-water system.

[0023] Step S32, performing Copula function modeling processing according to the multi-source edge probability set, wherein the nonlinear correlation between wind-solar output and hydropower regulation is fitted by using a t-Copula model with tail dependency capability to obtain an edge dependency structure of the wind-solar-water system; It can be understood that in order to characterize the complex joint behavior pattern between wind power, photovoltaic output status and hydropower regulation capacity, this step introduces the Copula function modeling method based on the multi-source edge probability set obtained in the previous step to achieve effective coupling between edge distribution and joint distribution. In particular, the t-Copula model with tail dependence capability is used for modeling. Compared with the traditional Gaussian Copula, this model can better capture the correlation between various energy sources under extreme conditions (such as high output or low output), and is particularly suitable for peak synchronization or sudden drop resonance phenomena that frequently occur in energy systems.

[0024] The t-Copula model in this step is based on the t-distribution, and its structure allows for modeling strong tail dependencies between variables, that is, when the wind power and photovoltaic systems are in extreme states, whether the hydropower system has the ability to perform corresponding load regulation. During the construction process, first, pseudo-observations are processed for each pair of marginal variables to uniformly map the marginal distributions to the standard uniform space. Then, the correlation matrix of the t-Copula is estimated based on the sample covariance matrix and the degrees of freedom parameter. Finally, the joint distribution characteristics among the power outputs of the three types of energy sources (wind, light, and water) in the original space are restored. This modeling process effectively makes up for the problem of insufficient expressiveness of traditional linear correlation indicators (such as the Pearson coefficient) under non-Gaussian, multi-scale, and multi-modal data.

[0025] Step S33: Perform graph structure conversion processing according to the marginal dependence structure. Specifically, by mapping the Copula dependence measure to the node association potential function in the graph and embedding the association strength threshold, a probability graph model with marginal relationship weights is constructed to obtain the conditional dependence graph structure of wind, light, and water. It can be understood that this step is based on the constructed Copula marginal dependence structure, and further transforms the complex multi-source coupling probability relationship into a graph structure model that can be used for spatial reasoning and joint modeling. During this process, this step maps the dependence strength characterized by the t-Copula - usually expressed in the form of correlation coefficients, Copula density functions, or tail joint probabilities - to the "potential function" in the graph structure, that is, the weight of the connecting edge between nodes, which is used to characterize the probabilistic interaction relationship between each energy subsystem under different states. This potential function not only reflects the direct dependence degree between variables but also retains the statistical significance under extreme power output conditions, thus introducing the true system cooperation characteristics into the graph model.

[0026] Among them, the potential function is defined as follows:

[0027] Among them, represents the edge potential function between node and , representing its joint dependence weight in state , . represents the current value or state of node , represents the current value or state of node , represents an adjustable edge weight scaling coefficient, reflecting the influence degree of this edge in the global potential energy, usually determined by training or experience adjustment. represents the joint density function using the t-Copula. represents the marginal distribution function of variable . Representation variables The marginal distribution function of .

[0028] Furthermore, in order to enhance the stability and recognition ability of the graph structure, this step introduces an association strength threshold mechanism to retain only the node association edges above the set threshold, thereby avoiding the graph structure from being too dense or containing redundant low-correlation relationships. This sparse processing helps to extract the most critical interaction channels in the wind, solar and water systems, so that subsequent Markov random field modeling can focus on the dominant structure of the system behavior. In addition, the nodes themselves can correspond to specific types of energy units (such as a wind farm or a reservoir node), while the edges represent their joint fluctuations or coordinated regulation probabilities in a certain spatiotemporal coupling scenario.

[0029] Step S34: Markov random field modeling is performed according to the wind-solar-water condition dependency graph structure, and a wind-solar-water joint spatial structure graph is obtained by defining a local neighborhood potential energy function and solving the maximum a posteriori estimate of the overall joint distribution.

[0030] It can be understood that this step is based on the constructed wind, solar and water condition dependency graph structure to carry out Markov random field (MRF) modeling, so as to achieve the global spatial joint expression of the multi-source coupling state of the wind, solar and water system. The process first defines the local neighborhood potential energy function based on each node in the graph (such as wind farm output, hydropower regulation status, photovoltaic power generation trend) and its edge weight (the strength of association mapped by the Copula dependency metric), that is, to model the synergistic relationship between its state and the state of adjacent nodes in the local context of each node. This type of potential energy function not only reflects the probability of a single point state, but also integrates the statistical consistency constraints between it and the adjacent nodes, which is a key factor in describing the joint distribution structure.

[0031] Among them, the local neighborhood potential energy function is as follows:

[0032] in, Representation Node The local neighborhood potential energy function represents the node state The state set of its neighbor nodes The combined potential between Representation Node The neighbor node state set, Representation Node With neighbor nodes The edge potential function between them describes the dependency between node states. Representation Node and neighbor nodes state variables.

[0033] Next, the Markov Random Field (MRF) forms a global joint probability distribution model through the aggregation of potentials across the entire graph. The system is solved using the Maximum A Posteriori (MAP) method on this distribution, with the goal of finding the most likely combination of the overall system state given the observed data. This approach effectively extends local dependencies to the global optimal state, capturing the spatial coordination potential in the scheduling process of the wind-solar-hydro multi-source system without explicitly establishing time-causal relationships.

[0034] Step S4: Based on the wind-solar-hydro joint spatial structure diagram, the state of charge of the energy storage system, and the device degradation parameters, perform distributed tensor decomposition to construct an energy tensor representation, and then embed the energy tensor representation into a dynamic game graph to generate a sub-graph of the energy storage response strategy. It can be understood that in this step, by integrating the joint spatial structure information, energy storage state, and device degradation into a tensor representation, and then combining the game graph mechanism to refine the response strategy, the generated energy storage response strategy not only has global coupling and time adaptability but also can dynamically reflect the impact of system aging on energy regulation capabilities, thereby providing more reliable and robust strategy support for subsequent load forecasting and scheduling optimization. In this step, step S4 includes steps S41, S42, and S43.

[0035] Step S41: Perform high-dimensional data tensor construction processing based on the wind-solar-hydro joint spatial structure diagram, the time series of the state of charge of the energy storage system, and the device degradation parameters. Specifically, map the spatial nodes, time steps, and device health status to the three-dimensional tensor dimensions to obtain the original energy tensor representing multi-energy interaction and energy storage state. It can be understood that this step uses the wind-solar-hydro joint spatial structure diagram as the topological framework, integrates the time series of the state of charge of the energy storage system and the key device degradation parameters, and performs unified high-dimensional data modeling. The core is to construct a three-dimensional tensor to represent the dynamic interaction of the multi-source energy system. First, in the first dimension of the tensor, the system takes the node positions of wind power, photovoltaic power, hydropower, and energy storage facilities respectively as the "spatial dimension" to ensure that the geographical and structural coupling relationships of the energy flow are effectively encoded; the second dimension is the "time dimension", and the evolution process of the state of charge of the energy storage system is discretized in sequence using a fixed time step (such as 5 minutes or 15 minutes) to capture the dynamic changes of the energy system over time; the third dimension is the "device health dimension", including parameters such as the battery capacity attenuation rate, internal resistance growth trend, and maximum cycle number ratio that characterize the device degradation state, so as to reflect the potential performance constraints of the energy storage system.

[0036] This construction process not only preserves the probabilistic dependence relationship of the spatial linkage of each node of the wind, light and water sources, but also systematically introduces the capacity boundary of energy storage devices in actual operation, enabling the original energy tensor to have a triple interaction structure of space, time and device state. Through unified modeling in tensor form, natural fusion of heterogeneous data sources can be achieved, avoiding problems such as difficult time alignment and isolated state modeling in traditional data representation. Technically, the construction of this third-order tensor lays a data foundation for subsequent distributed tensor decomposition, policy inference and response optimization.

[0037] Step S42: Perform distributed tensor decomposition processing according to the original energy tensor. Among them, by jointly using high-order singular value decomposition and non-negative tensor decomposition, extract the low-rank core tensor structure of the spatio-temporal-health factors to obtain a compressed energy tensor representation; It can be understood that this step performs distributed tensor decomposition processing on the constructed original third-order energy tensor, aiming to extract the most representative low-rank feature structure from the complex spatio-temporal-health state interaction, and realize unified modeling of data compression and latent variable extraction. In this step, two types of algorithms, high-order singular value decomposition (HOSVD) and non-negative tensor decomposition (NTF), are jointly used, each providing supplements for different tensor structure characteristics: high-order singular value decomposition (HOSVD) can effectively identify the most important change directions in energy data by decomposing the tensor into multiple orthogonal matrices and a core tensor, so as to capture the evolution trend of the dominant energy flow and the spatial cooperation mode; non-negative tensor decomposition (NTF) introduces non-negative constraints on the premise of maintaining physical interpretability, making the extracted components identifiable in engineering. For example, attributing specific output changes to equipment degradation or load fluctuations.

[0038] In actual execution, in this step, high-order singular value decomposition (HOSVD) is first used for pre-decomposition to obtain a coarse-grained low-rank approximation, and then a non-negative tensor factor update mechanism is introduced in each dimension to improve the physical interpretability and sparsity of the decomposed factors. This process is carried out in parallel under a distributed computing framework, significantly improving the computational efficiency of processing large-scale regional energy tensors. Then, the core tensor obtained by HOSVD is used as the input for non-negative tensor decomposition (NTF) processing. Under the premise of maintaining non-negative constraints, NTF further decomposes the core tensor to extract more physically interpretable factor matrices, corresponding to non-negative factors of space, time and health state respectively. The non-negative constraint ensures the interpretability of the factors. For example, energy flow, health indicators, etc. must be non-negative numbers, which helps the subsequent intuitive understanding of device state and energy interaction. Finally, the low-rank factors obtained by NTF decomposition are used as the compressed representation of the tensor. The compressed energy tensor retains the main energy coupling structure and operation characteristics, while removing redundant or noisy information, thus providing an efficient, low-dimensional and information-rich input basis for subsequent dynamic game strategy generation.

[0039] Step S43: Perform dynamic game graph embedding processing based on the energy tensor representation. Specifically, by introducing an evolutionary strategy network under a multi-agent game mechanism, map the tensor factors to the node strategy payoff vectors, and adopt a graph attention mechanism to adjust the dynamic response of the edge weights to obtain a subgraph of the energy storage response strategy.

[0040] It can be understood that this step is based on the low-dimensional energy tensor representation and further embeds it into the dynamic game graph framework to construct a subgraph of the energy storage response strategy with spatio-temporal evolution characteristics and health perception capabilities. This process first introduces a multi-agent game mechanism, maps various factors in the energy tensor (such as regional load situation, renewable energy output pattern, energy storage health status, etc.) to the node states in the game graph, and then constructs a strategy payoff vector to represent the optional response plans and expected revenue levels of each agent (such as wind farms, photovoltaic power plants, hydropower plants, or energy storage units) in the current environment. By introducing the evolutionary strategy network, the system can capture the non-linear effects and dynamic adjustment paths generated by the strategy games between different agents, so that the strategy evolution is not only driven by the current state but also reflects the historical revenue differences accumulated from long-term collaborative games.

[0041] To enhance the expression ability of the game graph for the coupling strength between different agents, a graph attention mechanism is introduced to dynamically weight and adjust the edge weights between nodes. Specifically, based on the state coupling relationship and strategy differences in the current tensor representation, the system automatically strengthens the influence of the edges of key interactions (such as the connection between the energy storage unit and the high-fluctuation wind-solar nodes) through attention weight learning, while suppressing invalid or redundant connections. This attention-guided game graph structure enables the energy storage response strategy to focus on the energy interaction channels with the greatest potential impact during the modeling process, improving the pertinence and scheduling efficiency of the response strategy.

[0042] Step S5: According to the subgraph of the energy storage response strategy, the load data and fluctuation data of the regional load center, perform singular spectrum analysis to extract periodic load characteristics, and then input them into a pulsed neural network to predict the power consumption sequence to obtain a multi-energy output path field; It can be understood that this step effectively realizes the in-depth mining and accurate prediction of the time-varying law of the load, improves the system's perception and control capabilities of the spatio-temporal dynamics of multi-energy output, and ensures the stability and economy of the integrated scheduling. In this step, Step S5 includes Step S51, Step S52, and Step S53.

[0043] Step S51: Perform singular spectrum analysis processing according to the subgraph of the energy storage response strategy, the load time series, and the volatility index of the regional load center. Specifically, by constructing an extended trajectory matrix of the load data and performing eigenvalue decomposition, extract the periodic eigenvectors and trend components to obtain a set of periodic load characteristics; It is understandable that this step first integrates the regional load time series associated with the energy storage response strategy subgraph with its volatility index, and deeply processes the load data through Singular Spectrum Analysis (SSA). The specific operations include first constructing an extended trajectory matrix of the load data, mapping the one-dimensional load time series to a high-dimensional phase space through window embedding, which captures the hidden temporal structure and non-stationary characteristics in the load data. Subsequently, eigenvalue decomposition (Singular Value Decomposition SVD) is performed on the trajectory matrix to extract the eigenvectors and singular value sequences of the matrix. These eigenvalues represent the energy contributions of different components in the load sequence. By analyzing their distributions, periodic changes (such as daily / weekly load fluctuations), trend components (such as long-term load growth), and random perturbations can be effectively distinguished. Among them, in the reconstruction stage, the first several principal components with high energy ratios are selected to reconstruct the signal, removing high-frequency noise and short-term fluctuations, and finally obtaining a set of periodic load characteristics. This processing not only retains the main patterns of load changes but also strengthens the response correlation of the energy storage system in coping with periodic load changes. Through this processing, not only can the seasonal and intraday load change rules of the regional load center be accurately captured, but also the adjustment effect of the energy storage system response strategy on load fluctuations can be reflected, providing a solid feature basis for subsequent electricity consumption behavior prediction.

[0044] Step S52: Perform spiking neural network modeling processing based on the set of periodic load characteristics. Specifically, by constructing a spiking neural network and using periodic pulse excitation to simulate the cumulative memory effect of historical power patterns on the neuron response path, a regional electricity consumption behavior prediction sequence is obtained. It is understandable that this step uses the extracted set of periodic load characteristics as input to construct a spiking neural network to simulate and predict the evolution process of the electricity consumption behavior of the regional load center. Different from traditional continuous neural networks, spiking neural networks adopt an event-driven spiking mechanism, in which neurons communicate through discontinuous spike signals, which is more in line with the non-stationary and sudden characteristics of actual power loads in time series. The model first constructs a Poisson process based on the periodic load characteristics to simulate periodic pulse excitation, enabling neurons to be excited in a random but statistically constrained manner at different time intervals, generating an input spike train approximating the actual fluctuation rhythm.

[0045] Next, each spiking neuron in the network integrates the input spike signals according to a biologically inspired time accumulation mechanism and triggers a response after crossing the threshold, thus forming a "cumulative memory path" for the historical load state. In particular, a variable decay rate and a memory retention window mechanism are introduced, enabling the network to selectively retain the sensitive response to recent load patterns while suppressing redundant historical data, improving the immediacy and adaptability of the model's response to sudden loads. In addition, a joint loss function of temporal cross-entropy and periodic error is introduced during the training of the network, enabling it to not only fit the numerical trend of load changes but also enhance the ability to identify periodic patterns.

[0046] Step S53: Perform the construction process of the output path field according to the regional electricity consumption behavior prediction sequence and the energy storage response strategy sub - graph. Among them, by introducing the state - action mapping mechanism and the path constraint graph optimization strategy, an optimal energy flow path set driven by both multi - source collaboration and load response is formed to obtain the multi - energy output path field.

[0047] It can be understood that this step combines the regional electricity consumption behavior prediction sequence and the energy storage response strategy sub - graph to construct the multi - energy output path field. The core lies in modeling the matching relationship between dynamic load demand and energy storage response based on the state - action mapping mechanism, and solving the optimal energy path set by combining the graph optimization strategy. First, the regional electricity consumption behavior prediction sequence provides the short - term and medium - term load evolution trends, providing load targets for the energy output scheduling of energy storage, wind, light, water, etc.; while the energy storage response strategy sub - graph forms the optimal response strategies and cost functions of each energy storage unit in different states in the previous dynamic game mechanism. The system constructs a state - action space accordingly, where the "state" is driven by electricity consumption prediction, and the "action" is composed of the adjustment behaviors of energy storage and renewable energy.

[0048] Subsequently, introduce the path constraint graph optimization strategy, that is, embed the path feasibility function based on energy supply - demand balance, equipment health constraints, scheduling cost upper limit, and physical topology boundary in the graph structure. In this optimized graph, each edge represents a certain energy transfer behavior, and each path represents a feasible scheduling link from the energy source (such as photovoltaic, hydropower, energy storage unit) to the load center. This step uses the particle swarm optimization algorithm to find multiple feasible paths with the minimum cost and the most timely response under the current electricity consumption prediction, and dynamically evaluates path redundancy and adjustment margin to ensure the robustness and reversibility of the overall energy flow.

[0049] Step S6: Perform fuzzy approximate reasoning on the path field to extract stability rules, and input them into the probabilistic context - free grammar model to generate a dynamic scheduling strategy, obtaining the optimal joint scheduling sequence.

[0050] It can be understood that this step enhances the robustness and generalization ability of the path selection process through fuzzy reasoning, and realizes the probabilistic modeling and structure generation of the dynamic scheduling strategy through the probabilistic context - free grammar model, making the finally output joint scheduling sequence have predictability, interpretability, and real - time adaptability, providing a refined and intelligent operation plan for the multi - energy integration system. In this step, Step S6 includes Step S61, Step S62, and Step S63.

[0051] Step S61: Perform fuzzy approximate reasoning based on the multi-energy output path field. Specifically, a fuzzy relation matrix is constructed and the Takagi-Sugeno fuzzy model is introduced to model the non-exact mapping relationship between the path state and the fluctuation threshold, extract the stability rule set under path perturbations, and obtain the fuzzy rule expression characterizing the stable structure of the path field. It can be understood that this step aims at the non-linear perturbation characteristics and uncertain evolution trends existing in the multi-energy output path field, and uses fuzzy approximate reasoning technology to abstractly model the stability characteristics of the path field. The specific steps are as follows: First, a state feature set is constructed according to the scheduling attributes such as the energy flow density, load matching degree, and time response delay of each path in the path field, and combined with perturbation indicators such as the load volatility and the output variation of wind, light, water, and storage, to form a multi-dimensional input space. On this basis, the Takagi-Sugeno (T-S) fuzzy inference model is introduced, and the non-exact mapping relationship between the path state and the perturbation threshold is characterized by constructing a fuzzy relation matrix, that is, rules of the type "if the path state is A and the perturbation is B, then the stability is C" are embedded into the system, and the activation degrees of multiple fuzzy rules are normalized and combined to dynamically approximate the stability response behavior under path perturbations.

[0052] Among them, the application advantage of the T-S model in this scenario is reflected in its ability to combine and fit the local linear segments between the input state and the output stability, which not only improves the modeling accuracy but also provides a well-explainable stability semantic expression for path perturbation responses. Finally, through clustering and summarizing the fuzzy inference outputs of each path, the system extracts a set of "stability rule sets" in a regularized form. This rule set reflects, in the form of fuzzy rules, which path combinations have high output stability and energy continuity under different perturbation situations, thus laying a foundation for the generation of subsequent scheduling strategy logic.

[0053] Step S62: Perform probabilistic context-free grammar modeling based on the fuzzy rule expression. Specifically, by encoding the stability rules as grammar production rules and assigning path probability weights to them, a probabilistic grammar model for characterizing the time-varying strategy generation structure is constructed to obtain a dynamic scheduling strategy generator. It can be understood that this step systematically organizes the path field stability rules previously obtained through fuzzy approximate reasoning into executable scheduling logic, and the system further introduces probabilistic context-free grammar for modeling. The core idea of this step is to transform the fuzzy rule expression form into the production structure of a formal language. Among them, each "path state - perturbation condition - stability response" rule is encoded as a grammar production rule, for example, S→AB[p], where S is a non-terminal symbol, A and B represent path states or perturbation conditions, and p is the probability of this production rule, reflecting the triggering possibility and actual stability performance of this strategy in the dynamic system.

[0054] In the modeling process, first, semantic elements such as state labels, perturbation factors, and output stability categories in fuzzy rules are mapped to the symbol set of probabilistic context-free grammar. Combining multi-source data such as rule matching frequency, system historical evolution path data, and scheduling response success rate, probability weighting is performed to construct multiple sets of production rules with different selection weights. Subsequently, a dynamic policy generation tree is defined through these weighted grammar production rules, supporting the real-time adaptive expansion of the most likely policy path according to the current operating state and predicted perturbations. Among them, probabilistic context-free grammar not only retains the recursive construction ability of the grammar structure, enabling the scheduling policy to have the advantage of combinatorial expression, but also introduces a probability weighting mechanism, making policy generation have "soft selectivity" and adaptability, and can dynamically adjust the optimal policy sequence according to the current system state. The finally output "dynamic scheduling policy generator" can not only learn the stability semantic structure in the offline stage, but also adaptively decide the scheduling path during real-time operation, combining rule interpretability and response flexibility.

[0055] Step S63: Perform joint policy sequence generation processing according to the dynamic scheduling policy generator, and infer and generate the optimal joint scheduling sequence in the context of the integrated wind-solar-hydro-storage system by traversing the grammar production tree and combining resource state feedback constraints.

[0056] It can be understood that this step takes the current state of the integrated wind-solar-hydro-storage system as the input root node, starts traversing the grammar production tree defined by probabilistic context-free grammar, and continuously introduces feedback information of resource states (such as energy storage state of charge, uncertainty of wind and solar power output, and regulation margin of hydropower) as constraint conditions during the traversal process to perform constraint pruning and dynamic weight adjustment on the expansion path of each production rule.

[0057] In specific operations, the traversal strategy not only considers the probability weights of each production rule, but also integrates the "dynamic context" constructed by resource state feedback, making the scheduling decision no longer based solely on the stability rules learned from history, but performing real-time optimization by combining the current system operation boundary and resource availability. To improve the inference efficiency, the system uses probability pruning + priority queue expansion to avoid wasting computing resources on low-value or conflicting paths, while retaining the policy expansion paths with high probability and high resource matching degree, thereby constructing a joint policy sequence that satisfies "temporal stability + resource coupling coordination + minimum scheduling cost".

[0058] In the process of this policy generation, this step forms a complementarity through the grammar generation advantage of probabilistic context-free grammar and the imprecise reasoning ability of fuzzy rules: the grammar mechanism provides clear structural constraints and combinatorial expression ability, while the feedback constraint mechanism embeds the feasibility boundary of the current environment. The two together ensure that the generated scheduling sequence has high operability and real-time adjustability when facing the complex and changeable source-load-storage coordination relationship.

[0059] In this step, by structuring and serializing the scheduling rules and combining with the resource status feedback dynamic guidance strategy, the system can efficiently generate an optimal joint scheduling strategy sequence with strong pertinence, timely response and excellent coordination in the coupled wind-solar-hydro-storage system, significantly enhancing the operation stability and comprehensive scheduling efficiency of the multi-energy system.

[0060] Embodiment 2: As Figure 2 shown, this embodiment provides a wind-solar-hydro-storage integrated scheduling system based on big data. Refer to Figure 2 The system includes an acquisition unit 701, an extraction unit 702, a construction unit 703, a decomposition unit 704, an analysis unit 705 and a scheduling unit 706.

[0061] The acquisition unit 701 is used to acquire wind-solar-hydro-storage data, the state of charge of the energy storage system, device degradation parameters, load data and fluctuation data of the regional load center; The extraction unit 702 is used to extract the driving factors in the wind farm and photovoltaic power plant data from the wind-solar-hydro-storage data based on the generalized dynamic factor model, and perform dynamic modeling on the driving factors to obtain the output state characteristic map of the wind farm and photovoltaic power plant; The construction unit 703 is used to construct an edge dependence structure for the output state characteristic map and the hydropower data in the wind-solar-hydro-storage data based on the Copula function, and input the structure into the Markov random field model to obtain the wind-solar joint space structure diagram; The decomposition unit 704 is used to perform distributed tensor decomposition according to the wind-solar joint space structure diagram, the state of charge of the energy storage system and device degradation parameters to construct an energy tensor representation, and then embed the energy tensor representation into a dynamic game graph to generate an energy storage response strategy sub-graph; The analysis unit 705 is used to perform singular spectrum analysis on the energy storage response strategy sub-graph, the load data and fluctuation data of the regional load center to extract periodic load characteristics, and then input them into a pulse neural network to predict the power consumption sequence to obtain a multi-energy output path field; The scheduling unit 706 is used to perform fuzzy approximate reasoning on the path field to extract stability rules, and input them into a probabilistic context-free grammar model to generate a dynamic scheduling strategy to obtain an optimal joint scheduling sequence.

[0062] It should be noted that regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment related to the method, and will not be elaborated here.

[0063] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention. As mentioned above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or replacements, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A comprehensive scheduling method for wind, light, water and energy storage based on big data, characterized in that Including: Obtain the data of wind, light, water and energy storage, the state of charge of the energy storage system, the device degradation parameters, the load data and the fluctuation data of the regional load center; Based on the generalized dynamic factor model, extract the driving factors in the wind farm and photovoltaic power station data in the wind-light-water-energy storage data, and perform dynamic modeling on the driving factors to obtain the output state characteristic map of the wind farm and photovoltaic power station; Based on the Copula function, construct an edge dependence structure for the output state characteristic map and the hydropower data in the wind-light-water-energy storage data, and input this structure into the Markov random field model to obtain the wind-light-water joint spatial structure map; According to the wind-light-water joint spatial structure map, the state of charge of the energy storage system and the device degradation parameters, perform distributed tensor decomposition to construct an energy tensor representation, and then embed the energy tensor representation into the dynamic game graph to generate an energy storage response strategy subgraph; According to the energy storage response strategy subgraph, the load data and the fluctuation data of the regional load center, perform singular spectrum analysis to extract the periodic load characteristics, and then input them into the pulsed neural network to predict the power consumption sequence to obtain the multi-energy output path field; Perform fuzzy approximate reasoning on the path field to extract stability rules, and input them into the probabilistic context-free grammar model to generate a dynamic scheduling strategy to obtain the optimal joint scheduling sequence.

2. The integrated scheduling method for wind-solar-hydro energy storage based on big data according to claim 1, wherein , Based on the generalized dynamic factor model, extract the driving factors in the wind farm and photovoltaic power station data in the wind-light-water-energy storage data, and perform dynamic modeling on the driving factors, including: Perform generalized dynamic factor model processing according to the wind speed sequence of the wind farm and the irradiance time series data of the photovoltaic power station. Among them, extract the co-varying factors of the driving system by maximizing the eigenvalue decomposition of the covariance matrix to obtain the main factor sequence representing the coupling drive of wind-light resources; Perform threshold vector autoregressive modeling processing according to the main factor sequence. By setting the wind speed volatility and the light mutation rate as threshold variables to divide the nonlinear state interval, obtain the interval-dependent dynamic expression of wind-light output; Perform state embedding graph construction processing according to this dynamic expression. By using the state transition direction of time series nodes within the threshold interval to generate a directed graph structure, obtain the wind-light output state characteristic map including gated connections and nonlinear interaction characteristics.

3. The integrated scheduling method for wind, light, water and energy storage based on big data according to claim 1, wherein , Based on the Copula function, construct an edge dependence structure for the output state characteristic map and the hydropower data in the wind-light-water-energy storage data, and input this structure into the Markov random field model, including: Perform edge distribution extraction processing according to the output state characteristic map and the water level-output regulation data of the hydropower station. Among them, extract the edge probability distribution of each variable in different operating states by using kernel density estimation to obtain the multi-source edge probability set of the wind-light-water system; Perform Copula function modeling processing according to the multi-source edge probability set. Among them, use the t-Copula model with tail dependence ability to fit the nonlinear correlation relationship between wind-light output and hydropower regulation to obtain the edge dependence structure of the wind-light-water system; Perform graph structure conversion processing according to the edge dependence structure. Specifically, construct a probability graph model with edge relationship weights by mapping the Copula dependence measure to the node association potential function in the graph and embedding the association strength threshold, and obtain the wind-solar-hydro conditional dependence graph structure; Perform Markov random field modeling processing according to the wind-solar-hydro conditional dependence graph structure. By defining the local neighborhood potential function and solving the maximum a posteriori estimation of the overall joint distribution, obtain the wind-solar-hydro joint space structure graph.

4. The integrated scheduling method for wind-solar-hydro-storage based on big data according to claim 1, wherein ,Construct an energy tensor representation by performing distributed tensor decomposition according to the wind-solar-hydro joint space structure graph, the state of charge of the energy storage system, and the device degradation parameters, and then embed the energy tensor representation into a dynamic game graph to generate a sub-graph of the energy storage response strategy, including: Perform high-dimensional data tensor construction processing according to the wind-solar-hydro joint space structure graph, the time series of the state of charge of the energy storage system, and the device degradation parameters. Specifically, map the spatial nodes, time steps, and device health status to the three-order tensor dimensions to obtain the original energy tensor representing multi-energy interaction and energy storage status; Perform distributed tensor decomposition processing on the original energy tensor. Specifically, jointly use high-order singular value decomposition and non-negative tensor decomposition to extract the low-rank core tensor structure of the spatio-temporal-health factors, and obtain the compressed energy tensor representation; Perform dynamic game graph embedding processing according to the energy tensor representation. Specifically, introduce an evolutionary strategy network under the multi-agent game mechanism, map the tensor factors to the node strategy payoff vector, and use the graph attention mechanism to adjust the edge weight dynamic response to obtain the sub-graph of the energy storage response strategy.

5. The integrated scheduling method for wind-solar-hydro-storage based on big data according to claim 1, wherein ,Perform singular spectrum analysis to extract periodic load characteristics according to the sub-graph of the energy storage response strategy, the load data and fluctuation data of the regional load center, and then input it into a spiking neural network to predict the electricity consumption sequence, including: Perform singular spectrum analysis processing according to the sub-graph of the energy storage response strategy, the load time series of the regional load center, and the volatility index. Specifically, construct an extended trajectory matrix of the load data and perform eigenvalue decomposition to extract the periodic eigenvector and trend component, and obtain the periodic load characteristic set; Perform spiking neural network modeling processing according to the periodic load characteristic set. Specifically, construct a spiking neural network, and use periodic pulse excitation to simulate the cumulative memory effect of the historical power pattern on the neuron response path to obtain the regional electricity consumption behavior prediction sequence; Perform output path field construction processing according to the regional electricity consumption behavior prediction sequence and the sub-graph of the energy storage response strategy. Specifically, introduce a state-action mapping mechanism and a path constraint graph optimization strategy to form an optimal energy flow path set driven by multi-source collaboration and load response, and obtain the multi-energy output path field.

6. A comprehensive scheduling system for wind, light, water and energy storage based on big data, characterized in that, Including: An acquisition unit for acquiring wind-solar-hydro energy storage data, the state of charge of the energy storage system, device degradation parameters, load data and fluctuation data of the regional load center; An extraction unit for extracting the driving factors in the wind farm and photovoltaic power plant data from the wind-solar-hydro energy storage data based on the generalized dynamic factor model, and performing dynamic modeling processing on the driving factors to obtain the output state characteristic graph of the wind farm and photovoltaic power plant; A construction unit, which is used to construct an edge dependence structure for the output state feature map and the hydropower data in the wind-solar-hydro energy storage data based on the Copula function, and input the structure into a Markov random field model to obtain a wind-solar-hydro joint spatial structure diagram; A decomposition unit, which is used to perform distributed tensor decomposition based on the wind-solar-hydro joint spatial structure diagram, the state of charge of the energy storage system, and the device degradation parameters to construct an energy tensor representation, and then embed the energy tensor representation into a dynamic game graph to generate a sub-graph of the energy storage response strategy; An analysis unit, which is used to perform singular spectrum analysis based on the sub-graph of the energy storage response strategy, the load data and the fluctuation data of the regional load center to extract the periodic load characteristics, and then input them into a pulsed neural network to predict the power consumption sequence to obtain a multi-energy output path field; A scheduling unit, which is used to perform fuzzy approximate reasoning on the path field to extract stability rules, and input them into a probabilistic context-free grammar model to generate a dynamic scheduling strategy to obtain an optimal joint scheduling sequence.

7. The integrated dispatching system for wind, light, water and energy storage based on big data according to claim 6, characterized in that The extraction unit includes: A first extraction subunit, which is used to perform generalized dynamic factor model processing on the wind speed sequence of the wind farm and the irradiance time series data of the photovoltaic power station. Among them, the driving system co-variable factor is extracted by maximizing the eigenvalue decomposition of the covariance matrix to obtain the main factor sequence representing the coupling drive of the wind-solar resources; A second extraction subunit, which is used to perform threshold vector autoregressive modeling processing based on the main factor sequence. By setting the wind speed volatility and the light mutation rate as threshold variables to divide the non-linear state interval, the interval dependence dynamic expression of the wind-solar output is obtained; A third extraction subunit, which is used to perform state embedding graph construction processing based on the dynamic expression. By using the state transition direction of the time series nodes within the threshold interval to generate a directed graph structure, a wind-solar output state feature map with gating connections and non-linear interaction characteristics is obtained.

8. The integrated dispatching system for wind, light, water and energy storage based on big data according to claim 6, characterized in that The construction unit includes: A first construction subunit, which is used to perform edge distribution extraction processing based on the output state feature map and the water level-output regulation data of the hydropower station. Among them, the edge probability distribution of each variable in different operating states is extracted by using kernel density estimation to obtain a multi-source edge probability set of the wind-solar-hydro system; A second construction subunit, which is used to perform Copula function modeling processing based on the multi-source edge probability set. Among them, the non-linear correlation relationship between the wind-solar output and the hydropower regulation is fitted by using a t-Copula model with tail dependence ability to obtain the edge dependence structure of the wind-solar-hydro system; A third construction subunit, which is used to perform graph structure conversion processing based on the edge dependence structure. Among them, the Copula dependence measure is mapped into the node association potential function in the graph and the association strength threshold is embedded to construct a probability graph model with edge relationship weights to obtain the wind-solar-hydro conditional dependence graph structure; A fourth construction subunit, which is used to perform Markov random field modeling processing based on the wind-solar-hydro conditional dependence graph structure. By defining the local neighborhood potential function and solving the maximum a posteriori estimation of the overall joint distribution, the wind-solar-hydro joint spatial structure diagram is obtained.

9. The integrated dispatching system for wind, light, water and energy storage based on big data according to claim 6, characterized in that, The decomposition unit includes: The first decomposition subunit is used to perform high-dimensional data tensor construction processing according to the combined wind-solar-hydro spatial structure diagram, the time series of the state of charge of the energy storage system, and the device degradation parameters. Among them, by mapping spatial nodes, time steps, and device health status to the three-order tensor dimensions, an original energy tensor representing multi-energy interaction and energy storage status is obtained; The second decomposition subunit is used to perform distributed tensor decomposition processing according to the original energy tensor. Among them, by jointly using high-order singular value decomposition and non-negative tensor decomposition to extract the low-rank core tensor structure of spatio-temporal-health factors, a compressed energy tensor representation is obtained; The third decomposition subunit is used to perform dynamic game graph embedding processing according to the energy tensor representation. Among them, by introducing an evolutionary strategy network under the multi-agent game mechanism, the tensor factors are mapped to the node strategy payoff vector, and the graph attention mechanism is used to adjust the dynamic response of the edge weights to obtain the energy storage response strategy subgraph.

10. The integrated dispatching system for wind, light, water and energy storage based on big data according to claim 6, characterized in that, The analysis unit includes: The first analysis subunit is used to perform singular spectrum analysis processing according to the energy storage response strategy subgraph, the load time series of the regional load center, and the volatility index. Among them, by constructing an extended trajectory matrix of the load data and performing eigenvalue decomposition, periodic eigenvectors and trend components are extracted to obtain a periodic load feature set; The second analysis subunit is used to perform spiking neural network modeling processing according to the periodic load feature set. Among them, by constructing a spiking neural network and using periodic pulse excitation to simulate the cumulative memory effect of the historical power pattern on the neuron response path, a regional electricity consumption behavior prediction sequence is obtained; The third analysis subunit is used to perform output path field construction processing according to the regional electricity consumption behavior prediction sequence and the energy storage response strategy subgraph. Among them, by introducing a state-action mapping mechanism and a path constraint graph optimization strategy, an optimal energy circulation path set under the dual drive of multi-source collaboration and load response is formed to obtain a multi-energy output path field.

Images