Photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis

Abstract

This invention discloses a photovoltaic (PV) energy storage collaborative optimization scheduling method based on industrial data analysis, comprising: collecting industrial energy operation data to form a standardized industrial operation dataset; constructing industrial operation sample feature vectors to form an industrial sample set; constructing an improved kernel Gaussian mixture ridge regression model to obtain corresponding component-level prediction results; constructing component-level schedulable capability vectors and constructing a capacity function; constructing a working condition collaborative fusion module based on Choquet integral to obtain a comprehensive scheduling decision index; and solving for energy storage charging and discharging power scheduling instructions and PV power allocation scheduling instructions based on the comprehensive scheduling decision index. This invention, by introducing the kernel Gaussian mixture ridge regression model and the Choquet integral method, achieves stable and executable collaborative optimization scheduling of PV output and energy storage charging and discharging in industrial scenarios.

Description

Technical Field

[0001] This invention relates to the field of industrial data analysis technology, and in particular to a photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis. Background Technology

[0002] In industrial energy scenarios, photovoltaic power generation and energy storage systems, as important distributed energy units, have been widely used to reduce electricity costs, smooth load fluctuations, and improve energy efficiency. With the expansion of industrial load scale and the increasing complexity of energy consumption structures, industrial energy operation exhibits significant time-varying, uncertain, and multi-condition characteristics, with complex coupling relationships between power output, energy storage status, and load demand. Traditional industrial energy dispatching often relies on rule-based control, deterministic models, or single prediction results for decision-making. It typically assumes that the distribution of operating conditions is relatively stable or can be described by a single model, making it difficult to fully characterize the differences in the distribution of multi-source operating data under different conditions and their impact on dispatching results. When faced with intensified power output fluctuations and tightening load constraints, this can easily lead to large fluctuations in dispatching results and high execution risks.

[0003] Some existing technologies attempt to introduce data-driven methods to predict and model photovoltaic output, load demand, or energy storage status, and conduct collaborative scheduling research based on this. Existing methods typically fit historical operating data using statistical regression, machine learning, or probabilistic models, and directly use the prediction results as scheduling input parameters. However, because industrial operating data often exhibits multi-distribution superposition characteristics, there are nonlinear interactions between prediction results under different operating conditions. Existing methods mostly use additive weighting or static fusion methods, making it difficult to simultaneously consider the synergistic effects of multi-condition prediction results and scheduling execution boundary constraints at the scheduling level. This results in insufficient scheduling stability and limited risk perception capabilities.

[0004] Therefore, how to provide a photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a photovoltaic (PV) energy storage collaborative optimization scheduling method based on industrial data analysis. This invention utilizes industrial energy operation data analysis methods, kernel Gaussian mixture ridge regression modeling, and a non-additive collaborative fusion method to uniformly model and make scheduling decisions regarding the complex correlation between PV output, energy storage operating status, and industrial load demand in industrial scenarios. By performing multi-condition distribution modeling on industrial energy operation data, it characterizes the changing characteristics of available PV output, industrial load demand, and energy storage dispatchability under different operating conditions. Furthermore, it introduces a non-additive collaborative fusion method to comprehensively calculate the collaborative influence relationships between multi-condition prediction results. Combined with scheduling feasibility boundary constraints, it generates executable scheduling decision indicators, achieving collaborative optimization scheduling of PV power allocation and energy storage charging and discharging power. This invention can improve the stability and executability of scheduling results under multi-condition and uncertain operating conditions, enhance the ability to perceive operational risks, and reduce the adverse effects of scheduling fluctuations. It has the advantages of strong adaptability, high scheduling reliability, and good suitability for industrial application scenarios.

[0006] The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to embodiments of the present invention includes: Collect industrial energy operation data in industrial scenarios, preprocess the industrial energy operation data, and form a standardized industrial operation dataset; Based on standardized industrial operation datasets, feature vectors of industrial operation samples are constructed, and the characteristics and coupling relationships of photovoltaic, energy storage and industrial load operation status are encapsulated in a structured manner to form an industrial sample set; An improved kernel Gaussian mixture ridge regression model is constructed. The improved Gaussian mixture model is used to fit the industrial sample set to obtain the set of operating condition components and their corresponding membership probabilities. The corresponding industrial sample features are mapped to the kernel feature space and ridge regularization constraints are introduced. Regression modeling is performed on photovoltaic available output, industrial load demand and energy storage dispatchable capacity respectively to obtain the corresponding component-level prediction results. Based on the component-level prediction results, a component-level schedulable capacity vector is constructed, and a capacity function is constructed based on historical industrial operation data and the characteristics of operation status changes within the current scheduling cycle. Based on the capacity function and component-level schedulable capability vectors, a working condition coordination fusion module based on Choquet integral is constructed to perform non-additive fusion on the component-level schedulable capability vectors of each operating condition component to obtain comprehensive scheduling decision indicators. Based on the comprehensive scheduling decision indicators, the corresponding energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions are obtained and then issued for execution.

[0007] Optionally, the industrial energy operation data specifically includes power output data on the electric side, energy storage state of charge data, energy storage charging and discharging power data, real-time power data of industrial load, load change rate data, and operation constraint parameter data.

[0008] Optionally, the preprocessing of industrial energy operation data specifically includes timestamp unification, missing value completion, outlier removal, and dimension normalization.

[0009] Optionally, forming the industrial sample set includes: At each scheduling time step, the industrial energy operation data at the corresponding time is read from the standardized industrial operation dataset and used as the basic operation data for the current time step; Based on standardized industrial energy operation data, the operation status data reflecting the photovoltaic output level and energy storage operation status are aggregated and packaged to form photovoltaic energy storage operation status characteristics, and the operation status data reflecting industrial electricity demand and its changing characteristics are aggregated and packaged to form industrial load operation status characteristics. Based on the operating status characteristics of photovoltaic energy storage and industrial load, a coupling relationship feature is constructed to characterize the mutual influence between photovoltaic output, energy storage regulation capability and industrial load constraint intensity. The photovoltaic energy storage operation status characteristics, industrial load operation status characteristics, and coupling relationship characteristics under the same scheduling time step are uniformly spliced ​​together to generate the industrial operation sample feature vector of the corresponding time step. The industrial operation sample feature vectors under each scheduling time step are then collected in chronological order to form an industrial sample set.

[0010] Optionally, obtaining the corresponding component-level prediction result includes: An improved kernel Gaussian mixture ridge regression model is constructed to extract the distribution feature data of each industrial operation sample feature vector at each scheduling time step. The Gaussian mixture model is used to perform fitting processing on the distribution feature data. A component splitting and merging evolution gate is introduced to perform adaptive splitting and merging update processing on the operating condition components to generate a set of operating condition components. Based on the set of operating condition components, the membership probability of each industrial operating sample corresponding to each operating condition component is calculated, and each industrial operating sample is assigned to the corresponding operating condition component according to the size of the membership probability, forming a subset of industrial samples corresponding to each operating condition component. For each industrial sample subset corresponding to each operating condition component, a component boundary shell structure is introduced. Based on the operating constraint parameter data and energy storage operating boundary conditions, a set of boundary samples close to the operating constraint boundary is generated from each industrial sample subset, thus obtaining the component boundary shell and the component main sample set. For each operating condition component, kernel function mapping is performed on the main sample set and the boundary shell of the component to obtain the corresponding kernel feature space representation. Ridge regularization constraint is introduced in the kernel feature space. The target quantity corresponding to the industrial energy operation data is used as the regression object. Regression modeling is performed on the available photovoltaic output, industrial load demand and energy storage dispatchability to obtain the corresponding component-level regression results under each operating condition component. Based on the component-level regression results, a component-level prediction result description set containing the prediction center value, the executable boundary descriptor, and the prediction stability descriptor is generated for each operating condition component, forming the corresponding component-level prediction result.

[0011] Optionally, the construction capacity function includes: Based on the component-level prediction results corresponding to each operating condition component and the standardized industrial operation dataset under the current scheduling time step, the component-level prediction results of energy storage dispatchable capacity corresponding to each operating condition component are mapped to the executable charging and discharging power boundary descriptor. At the same time, the component-level prediction results of industrial load demand are mapped to the load rigidity constraint urgency descriptor, and the component-level prediction results of photovoltaic available output are mapped to the available supply level descriptor, thus obtaining the component-level dispatchable capacity vector corresponding to each operating condition component. Based on historical industrial operation data and standardized industrial operation datasets within the current scheduling cycle, the changes in power output on the power side, the changes in energy storage state of charge, and the changes in industrial load power are jointly extracted and processed to obtain characteristic values ​​of the change in operating state. Based on the characteristic values ​​of the change in operating status and the membership probability of each operating condition component at the current scheduling time step, the contribution relationship is updated, the collaborative contribution relationship between operating condition components in historical industrial operating data is updated, and the capacity function at the current scheduling time step is generated.

[0012] Optionally, the obtained comprehensive scheduling decision indicators include: A working condition coordination and fusion module based on Choquet integral is constructed. The working condition coordination and fusion module includes a working condition scheduling representation unit, a coordination weight adjustment unit, a non-additive fusion unit, and a scheduling index generation unit. The operating condition scheduling characterization unit, based on the component-level schedulable capability vector corresponding to each operating condition component, performs attribute cross-combination encapsulation of the photovoltaic available supply level description, the energy storage regulation capacity boundary description, and the industrial load constraint urgency description to generate the combined scheduling characterization input within the operating condition component, and arranges each operating condition component in an ordered manner to obtain the operating condition component priority sequence. The collaborative weight adjustment unit, based on the priority sequence of operating condition components, the membership probability and capacity function of each operating condition component at the current scheduling time step, introduces a collaborative contribution constraint mapping structure, performs constraint mapping processing on the collaborative contribution relationship between operating condition components, performs suppression mapping and maintenance mapping on collaborative weights, and generates a collaborative weight vector corresponding to each operating condition component. The non-additive fusion unit is based on the combined scheduling representation input and the corresponding collaborative weight vector. Combined with the scheduling-aware Choquet fusion integral, a hierarchical non-additive integral path is introduced. The first layer of non-additive fusion calculation is performed within a single operating condition component to obtain the component fusion value corresponding to each operating condition component. Based on the component fusion value and the operating condition component priority sequence, the second layer of non-additive fusion calculation is performed to obtain the non-additive fusion value sequence. The scheduling index generation unit is based on the non-additive fused value sequence and introduces a scheduling feasibility boundary triggering mapping structure. It associates and maps the non-additive fused value sequence with the energy storage executable charge and discharge boundary, industrial load constraint boundary and power balance boundary. It encapsulates the available photovoltaic output, industrial load demand and energy storage scheduling capability in a unified way to generate a comprehensive scheduling decision index constrained by the scheduling feasibility boundary.

[0013] Optionally, obtaining the non-additive fusion value sequence includes: Based on the combined scheduling representation input, the collaborative weight vector and the capacity function, a component credibility screening channel is introduced to perform the operational condition component credibility screening process. The collaborative weights corresponding to operational condition components with membership probabilities below the threshold are reduced to obtain the target operational condition component set and the corresponding effective collaborative weight vector. Based on the combined scheduling representation input, effective collaborative weight vector and capacity function corresponding to the target operating condition component set, and combined with the scheduling-aware Choquet fusion integral, the first-level non-additive fusion calculation is performed on the photovoltaic available supply level description, energy storage regulation capacity boundary description and industrial load constraint urgency description within a single operating condition component to obtain the component fusion value corresponding to each target operating condition component. Based on the component fusion value corresponding to each target operating condition component and the standardized industrial operation dataset under the current scheduling time step, the scheduling executability pruning process is performed. The component fusion value that exceeds the energy storage executable charging and discharging boundary and the rigid constraint boundary of industrial load is pruned to obtain the executable component fusion value corresponding to each target operating condition component. Based on the fusion values ​​within executable components and the priority sequence of operating condition components, a multi-granularity non-additive integration path is introduced. The second-layer non-additive fusion calculation is performed using a scheduling-aware Choquet fusion integration method that is parallel at both the coarse-grained and fine-grained segment levels, resulting in a sequence of non-additive fusion values.

[0014] Optionally, the solution to obtain the corresponding energy storage charging and discharging power scheduling command and photovoltaic power allocation scheduling command includes: Based on the comprehensive scheduling decision indicators and the standardized industrial operation dataset under the current scheduling time step, the target range of photovoltaic adjustable power, the target range of energy storage charging and discharging, and the target range of industrial load power supply demand are extracted to form a set of collaborative scheduling target parameters. Based on the set of collaborative scheduling target parameters, joint variables are constructed for photovoltaic power allocation and energy storage charging and discharging power, and a set of scheduling constraints is constructed by combining the operational constraint parameter data. The set of scheduling constraints includes boundary constraints for energy storage charge state, boundary constraints for energy storage charging and discharging power, boundary constraints for industrial load power supply, and boundary constraints for photovoltaic output allocation. Based on the set of collaborative scheduling target parameters and the set of scheduling constraints, collaborative optimization is performed on the photovoltaic power allocation and energy storage charging and discharging power to obtain a set of candidate scheduling instructions that meet the constraints at each scheduling time step, and then the target scheduling instruction set is obtained by screening. The target scheduling instruction set includes energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions. The target scheduling instruction set is sent to the photovoltaic-side execution equipment and the energy storage-side execution equipment for execution.

[0015] The beneficial effects of this invention are: This invention proposes a photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis. It comprehensively adopts industrial energy operation data modeling method, kernel Gaussian hybrid ridge regression modeling method and non-additive collaborative fusion method to uniformly characterize and schedule the complex coupling relationship between photovoltaic output, energy storage operation status and industrial load demand in industrial scenarios.

[0016] The method of this invention constructs an operational sample feature vector from standardized industrial energy operation data, identifies and divides different operational condition components, and performs regression modeling on the available photovoltaic output, industrial load demand, and energy storage dispatchability under each operational condition in the kernel feature space to form component-level prediction results under multiple operational conditions. Furthermore, it introduces an operational condition collaborative fusion method based on Choquet integral to perform fusion calculation on the nonlinear interaction relationship between the prediction results of multiple operational conditions, and generates a comprehensive scheduling decision index in combination with industrial operation constraint boundaries to achieve collaborative optimization of photovoltaic power allocation and energy storage charging and discharging power.

[0017] The method of this invention can effectively improve the adaptability of scheduling results to complex operating states under multiple operating conditions, multiple distributions and uncertain operating conditions, avoid scheduling deviations caused by single prediction results, enable scheduling decisions to simultaneously meet operating constraints and execution boundary requirements, enhance the stability, reliability and executability of the scheduling process, reduce the operating risks of industrial energy systems, and is suitable for photovoltaic energy storage collaborative optimization scheduling applications in complex industrial scenarios. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows the photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis proposed in this invention. Figure 2 This is a functional flowchart of the improved kernel Gaussian mixture ridge regression model for the photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis proposed in this invention. Figure 3 This is a schematic diagram of the working condition collaborative fusion module of the photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 , Figure 2 and Figure 3 A photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis includes: Collect industrial energy operation data in industrial scenarios, preprocess the industrial energy operation data, and form a standardized industrial operation dataset; Based on standardized industrial operation datasets, feature vectors of industrial operation samples are constructed, and the characteristics and coupling relationships of photovoltaic, energy storage and industrial load operation status are encapsulated in a structured manner to form an industrial sample set; An improved kernel Gaussian mixture ridge regression model is constructed. The improved Gaussian mixture model is used to fit the industrial sample set to obtain the set of operating condition components and their corresponding membership probabilities. The corresponding industrial sample features are mapped to the kernel feature space and ridge regularization constraints are introduced. Regression modeling is performed on photovoltaic available output, industrial load demand and energy storage dispatchable capacity respectively to obtain the corresponding component-level prediction results. Based on the component-level prediction results, a component-level schedulable capacity vector is constructed, and a capacity function is constructed based on historical industrial operation data and the characteristics of operation status changes within the current scheduling cycle. Based on the capacity function and component-level schedulable capability vectors, a working condition coordination fusion module based on Choquet integral is constructed to perform non-additive fusion on the component-level schedulable capability vectors of each operating condition component to obtain comprehensive scheduling decision indicators. Based on the comprehensive scheduling decision indicators, the corresponding energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions are obtained and then issued for execution.

[0021] In this embodiment, the industrial energy operation data specifically includes power output data on the electric side, energy storage state of charge data, energy storage charging and discharging power data, real-time power data of industrial load, load change rate data, and operation constraint parameter data.

[0022] In this embodiment, the preprocessing of industrial energy operation data specifically includes timestamp unification, missing value completion, outlier removal, and dimension normalization.

[0023] In this embodiment, forming an industrial sample set includes: At each scheduling time step, the industrial energy operation data at the corresponding time is read from the standardized industrial operation dataset and used as the basic operation data for the current time step; Based on standardized industrial energy operation data, operational status data reflecting photovoltaic power output levels and energy storage operation status are aggregated and packaged to form photovoltaic energy storage operational status characteristics. Similarly, operational status data reflecting industrial electricity demand and its changing characteristics are aggregated and packaged to form industrial load operational status characteristics. The characteristics of photovoltaic energy storage operation status are as follows: The power output data, energy storage state of charge data, and energy storage charge and discharge power data are read from standardized industrial energy operation data. The power output data of the electric side is extracted by time window. The power output sequence of the electric side within the time window is arranged into a photovoltaic power output sequence in time order. The average value, maximum value, minimum value and absolute value of the difference between adjacent sampling points of the time window are calculated as the output level and fluctuation characteristics. Boundary verification is performed on the energy storage state of charge data. The current state of charge value is read and the difference between the current state of charge value and the state of charge value of the previous scheduling time step is calculated to obtain the state of charge change characterization quantity. At the same time, the charging and discharging direction of the energy storage charging and discharging power data is determined and the current power value, power change rate and continuous charging and discharging duration characterization quantity are extracted. The photovoltaic power output level and fluctuation characteristics, state of charge value and state of charge change characteristics, and charging and discharging power characteristics are spliced ​​and encapsulated in a fixed field order to obtain the photovoltaic energy storage operation status characteristics of the current scheduling time step. The characteristics of industrial load operation are as follows: The real-time power data and load change rate data of industrial load are read from the basic operation data. First, the real-time power data of industrial load is extracted by time window. The load power sequence within the time window is arranged into a load demand sequence in time order. The average value, peak value, valley value and peak-valley difference of the time window are calculated as load demand level and fluctuation characteristics. Consistency verification is performed on the load change rate data. The current load change rate value is read and the deviation between the change rate obtained by the difference between the current load change rate and the adjacent sampling points of the load demand sequence is calculated. This deviation is used as a confidence indicator of the change characteristics. At the same time, the absolute value of the load change rate is extracted as a load ramping strength indicator. The load demand level, fluctuation characteristics, load ramp intensity characteristics, and change characteristic reliability characteristics are concatenated and encapsulated in a fixed field order to obtain the industrial load operation status characteristics at the current scheduling time step. Based on the operating characteristics of photovoltaic energy storage and industrial load, a coupling relationship feature is constructed to characterize the mutual influence between photovoltaic output, energy storage regulation capability, and industrial load constraint intensity, wherein: A coupling relationship feature is constructed to characterize the interaction between photovoltaic power output, energy storage regulation capacity, and industrial load constraint intensity, specifically as follows: The algebraic difference between the photovoltaic power output level and the industrial load demand level is calculated to obtain the supply-demand deviation value. Based on the current state of charge value of energy storage and the boundary value of the chargeable and dischargeable power of energy storage, the adjustable range value of energy storage under the current scheduling time step is calculated to obtain the energy storage adjustment margin value. The supply-demand deviation value and the load constraint urgency value are numerically combined to generate the supply-demand deviation modulation value under the load constraint. The supply-demand deviation value, the energy storage adjustment margin value and the supply-demand deviation modulation value are sequentially spliced ​​and packaged to form the coupling relationship characteristics under the current scheduling time step. The photovoltaic energy storage operation status characteristics, industrial load operation status characteristics, and coupling relationship characteristics under the same scheduling time step are uniformly spliced ​​together to generate the industrial operation sample feature vector of the corresponding time step. The industrial operation sample feature vectors under each scheduling time step are then collected in chronological order to form an industrial sample set.

[0024] In this embodiment, obtaining the corresponding component-level prediction result includes: An improved kernel Gaussian mixture ridge regression model is constructed to extract the distribution feature data of each industrial operation sample feature vector at each scheduling time step. A Gaussian mixture model is then used to fit the distribution feature data. A component splitting and merging evolution gate is introduced to perform adaptive splitting and merging update processing on the operating condition components, generating a set of operating condition components, where: Extract the distribution feature data of each industrial operation sample feature vector at each scheduling time step, specifically as follows: The industrial operation sample feature vectors of the corresponding time step in the industrial sample set are read sequentially. All industrial operation sample feature vectors under the same scheduling time step are combined into a sample feature matrix. In the sample feature matrix, the value distribution of each feature component is statistically analyzed according to the feature dimension. For each feature component, the mean, variance and sample number under the scheduling time step are calculated. The mean, variance and sample number of each feature component are combined to form a distribution feature description vector. This process is repeated for all scheduling time steps to form a distribution feature data set across scheduling time steps. A Gaussian mixture model is used to fit the distribution characteristics of the data, specifically: Based on the distribution feature data set, an initial value for the number of Gaussian components is set, and the mean parameter, covariance parameter and weight parameter corresponding to each Gaussian component are initialized. Using the distribution feature data as input samples, the parameters of each Gaussian component are iteratively updated. In each iteration, the membership probability of each distribution feature data under each Gaussian component is calculated based on the current parameters, and the mean parameter, covariance parameter and weight parameter are recalculated based on the membership probability. When the change amplitude of each Gaussian component parameter in two adjacent iterations is less than the convergence threshold, the iteration calculation ends, and the Gaussian mixture fitting result corresponding to the distribution feature data set is obtained. The component splitting and merging evolution gate is a component evolution control structure set between the component parameter set and the operating condition component set of the Gaussian mixture model. It takes the parameter state of each Gaussian component in the Gaussian mixture model as input, performs a unified constraint description on the existence form of the Gaussian component, and limits the range of component form changes that the Gaussian component can present during the model evolution process. Generate a set of operating condition components, specifically: Read all Gaussian components, treat the parameter set corresponding to each Gaussian component as a running condition component, and associate the feature vector of each industrial running sample in the industrial sample set with the corresponding running condition component based on the membership probability under each Gaussian component. Collect all the retained Gaussian components and associated industrial running samples to form the running condition component set under the current scheduling cycle. Based on the set of operating condition components, the membership probability of each industrial operating sample corresponding to each operating condition component is calculated. Then, each industrial operating sample is assigned to its corresponding operating condition component according to the magnitude of the membership probability, forming a subset of industrial samples corresponding to each operating condition component. Where: The membership probability of each industrial operation sample corresponding to each operating condition component is calculated as follows: Read the feature vector of each industrial operation sample in the industrial sample set and the parameter set corresponding to each operating condition component in the operating condition component set. Substitute the feature vector of the industrial operation sample into the probability distribution expression corresponding to each operating condition component, calculate the probability value of the industrial operation sample under each operating condition component, and normalize the probability value of the same industrial operation sample under all operating condition components to obtain the membership probability set of the industrial operation sample corresponding to each operating condition component. The industrial sample subsets corresponding to each operating condition component are formed as follows: Based on the membership probability set, the membership probability values ​​of each industrial operation sample under each operating condition component are read. The industrial operation sample is assigned to the operating condition component with the largest membership probability value. The industrial operation samples assigned to the same operating condition component are aggregated to form the corresponding industrial sample subset. The allocation and aggregation process is repeated for each operating condition component in the operating condition component set to obtain the industrial sample subset corresponding to each operating condition component. For each industrial sample subset corresponding to each operating condition component, a component boundary shell structure is introduced. Based on the operating constraint parameter data and energy storage operating boundary conditions, a boundary sample set close to the operating constraint boundary is generated from each industrial sample subset, resulting in the component boundary shell and the component main sample set, where: The component boundary shell structure refers to a sample set structure that describes the distribution of samples in the neighborhood of the operating constraint boundary within the industrial sample subset corresponding to the operating condition component. It is distinct from the sample set inside the operating condition component and represents the sample distribution part near the operating constraint boundary in the sample feature space. Energy storage operation boundary conditions refer to a set of numerical constraints consisting of the upper and lower limits of the energy storage state of charge, the upper and lower limits of the energy storage charging and discharging power, and the energy storage charging and discharging direction restrictions. These constraints limit the range of the energy storage state of charge and the range of charging and discharging power values ​​that are allowed to occur at the scheduling time step. The component boundary shell and component main sample sets are obtained as follows: In the industrial sample subsets corresponding to each operating condition component, the energy storage state of charge value, energy storage charging and discharging power value, and operating constraint parameter value corresponding to each industrial operating sample are read one by one. The values ​​are compared with the upper and lower limit values ​​in the energy storage operating boundary conditions. When the distance between the value of the industrial operating sample in any constraint dimension and the corresponding boundary value is less than the threshold, the industrial operating sample is assigned to the component boundary shell set, and the remaining industrial operating samples are assigned to the component main sample set. The industrial sample subsets corresponding to each operating condition component are filtered and divided to obtain the component boundary shell and component main sample set corresponding to each operating condition component. For each operating condition component, kernel function mapping is performed on the main sample set and the boundary shell of the component to obtain the corresponding kernel feature space representation. Ridge regularization constraints are introduced into the kernel feature space. The target quantity corresponding to the industrial energy operation data is used as the regression object. Regression modeling is performed on the available photovoltaic output, industrial load demand, and energy storage dispatchability to obtain the component-level regression results for each operating condition component, where: Kernel function mapping is performed on the component main sample set and the component boundary shell respectively, specifically as follows: For each operating condition component, the corresponding component main sample set and the industrial operating sample feature vector contained in the component boundary shell are read respectively. The industrial operating sample feature vectors are substituted into the kernel function expression one by one for numerical transformation to obtain the mapping result of each industrial operating sample under the action of the kernel function. The kernel function mapping results of all industrial operating samples in the component main sample set are arranged into a main kernel feature matrix according to the sample order, and the kernel function mapping results of all industrial operating samples in the component boundary shell are arranged into a boundary kernel feature matrix according to the sample order to obtain the kernel feature space representation result of the corresponding operating condition component. Ridge regularization constraint is a form of parameter constraint introduced in the regression modeling process. By introducing constraint terms related to the sum of squares of model parameters into the regression objective function, the range of values ​​of regression parameters is limited, forming regularization constraints in the parameter set of the regression model. Regression modeling was performed on available photovoltaic power output, industrial load demand, and energy storage dispatchability, specifically as follows: Under the corresponding operating condition component, the main kernel feature matrix and the boundary kernel feature matrix in the kernel feature space representation result are read respectively. The target quantity data corresponding to the industrial operating sample are read from the industrial energy operation data. The target quantity data are grouped according to the available photovoltaic output, industrial load demand and energy storage dispatchable capacity. Under the introduction of ridge regularity constraint, kernel regression relationship is established based on the main kernel feature matrix and the boundary kernel feature matrix respectively. The parameter solution processing is performed on each group of target quantity data to obtain the regression parameter set under the corresponding operating condition component, forming the component-level regression result corresponding to the operating condition component. Based on the component-level regression results, a component-level prediction result description set is generated for each operating condition component, including the prediction center value, the executable boundary descriptor, and the prediction stability descriptor, forming the corresponding component-level prediction result, where: The corresponding component-level prediction results are generated as follows: For each operating condition component, based on the corresponding component-level regression results, the regression output values ​​of all industrial operating samples under the operating condition component are read, and statistical calculations are performed on the regression output values ​​to obtain the prediction center value under the operating condition component. Combining the regression output values ​​with the regression output results corresponding to the component boundary shell, the upper and lower bound values ​​of the regression output values ​​within the executable constraint range are calculated to obtain the executable boundary descriptor. The change amplitude of the regression output values ​​under adjacent scheduling time steps is statistically analyzed to calculate the degree of numerical fluctuation and obtain the prediction stability descriptor. The prediction center value, the executable boundary descriptor, and the prediction stability descriptor are uniformly encapsulated to form a component-level prediction result description set.

[0025] In this embodiment, constructing the capacity function includes: Based on the component-level prediction results corresponding to each operating condition component and the standardized industrial operation dataset under the current scheduling time step, the component-level prediction results of energy storage dispatchable capacity corresponding to each operating condition component are mapped to the executable charging and discharging power boundary descriptor. At the same time, the component-level prediction results of industrial load demand are mapped to the load rigidity constraint urgency descriptor, and the component-level prediction results of photovoltaic available output are mapped to the available supply level descriptor, thus obtaining the component-level dispatchable capacity vector corresponding to each operating condition component. Based on historical industrial operation data and standardized industrial operation datasets within the current scheduling cycle, joint extraction processing is performed on changes in power output on the power side, changes in the state of charge of energy storage, and changes in industrial load power to obtain characteristic values ​​of operating state changes, among which: The characteristic values ​​of the changes in the operating state are obtained as follows: The system reads the power output data of the electric side, the state of charge data of the energy storage, and the real-time power data of the industrial load within the corresponding time span from historical industrial operation data. It also reads the standardized industrial operation dataset within the current scheduling cycle. The system calculates the difference between the power output data of the electric side at the current scheduling time step and the average power output data of the electric side in the corresponding historical time period, the difference between the current state of charge data of the energy storage and the average state of charge data of the energy storage in the corresponding historical time period, and the difference between the current real-time power data of the industrial load and the average real-time power data of the industrial load in the corresponding historical time period. The three differences are normalized and combined according to their weights to obtain the characteristic value of the change in the operating state. Based on the characteristic values ​​of operational status changes and the membership probabilities of each operational condition component at the current scheduling time step, the contribution relationship is updated. This updates the collaborative contribution relationship between operational condition components in historical industrial operation data, generating the capacity function at the current scheduling time step, where: Generate the capacity function for the current scheduling time step, specifically as follows: Read the characteristic values ​​of the change in operating status and the membership probability values ​​corresponding to each operating condition component. Arrange the membership probability values ​​according to the index order of the operating condition components to form a membership probability sequence. Based on the characteristic values ​​of the change in operating status, correct the collaborative contribution values ​​of the operating condition components recorded in the historical industrial operation data. Assign low update weights to the collaborative contribution values ​​corresponding to the scheduling time steps with large changes in operating status, and assign high update weights to the collaborative contribution values ​​corresponding to the scheduling time steps with small changes in operating status. After completing the update of the collaborative contribution values, aggregate the updated collaborative contribution values ​​of the operating condition components according to the combination relationship of the operating condition components to form a capacity function defined under the current scheduling time step.

[0026] In this embodiment, obtaining the comprehensive scheduling decision index includes: A working condition coordination fusion module based on Choquet integrals is constructed. This module includes a working condition scheduling representation unit, a coordination weight adjustment unit, a non-additive fusion unit, and a scheduling index generation unit, wherein: Construct a working condition coordination and fusion module based on Choquet integrals, specifically as follows: The working condition scheduling representation unit, the collaborative weight adjustment unit, the non-additive fusion unit, and the scheduling index generation unit are connected in series to form the working condition collaborative fusion module. The operating condition scheduling characterization unit, based on the component-level schedulable capability vector corresponding to each operating condition component, performs attribute cross-combination encapsulation of the photovoltaic available supply level description, the energy storage regulation capacity boundary description, and the industrial load constraint urgency description to generate the combined scheduling characterization input within the operating condition component, and arranges each operating condition component in an ordered manner to obtain the operating condition component priority sequence. The collaborative weight adjustment unit, based on the priority sequence of operating condition components, the membership probability of each operating condition component at the current scheduling time step, and the capacity function, introduces a collaborative contribution constraint mapping structure. It performs constraint mapping processing on the collaborative contribution relationships between operating condition components, and performs suppression mapping and preservation mapping on the collaborative weights, generating a collaborative weight vector corresponding to each operating condition component, where: The collaborative contribution constraint mapping structure refers to a collaborative weight mapping structure set between the priority sequence of operating condition components, membership probability data and capacity function. Based on the collaborative relationship description data between operating condition components, it constrains and limits the weight values ​​that can be used by operating condition components in the collaborative calculation process, and standardizes the mapping result range of collaborative weights under different operating states. Generate the collaborative weight vector corresponding to each operating condition component, specifically as follows: Read the sorting position of each operating condition component in the operating condition component priority sequence, read the membership probability value corresponding to each operating condition component, and read the capacity value corresponding to the combination relationship of each operating condition component in the capacity function. Perform numerical combination operation on the membership probability value and capacity value according to the index order of the operating condition component, and perform suppression mapping and preservation mapping processing on the combination operation result according to the collaborative contribution constraint mapping structure. Arrange the weight values ​​corresponding to each operating condition component in order to form a collaborative weight vector. The non-additive fusion unit, based on the combined scheduling representation input and the corresponding cooperative weight vector, and combined with the scheduling-aware Choquet fusion integral, introduces a hierarchical non-additive integration path. It performs the first-level non-additive fusion calculation within a single operating condition component to obtain the intra-component fusion value corresponding to each operating condition component. Based on the intra-component fusion value and the operating condition component priority sequence, it performs the second-level non-additive fusion calculation to obtain a sequence of non-additive fusion values, where: A hierarchical non-additive integral path refers to an integral calculation structure set up in the non-additive fusion computing process, which divides the non-additive integral calculation into multiple interrelated integral stages and limits the input data format and integral order relationship corresponding to different integral stages. The scheduling index generation unit, based on a non-additive fused value sequence, introduces a scheduling feasibility boundary triggering mapping structure. It correlates and maps the non-additive fused value sequence with the energy storage executable charge / discharge boundary, industrial load constraint boundary, and power balance boundary. This unifies the available photovoltaic output, industrial load demand, and energy storage dispatchability, generating a comprehensive scheduling decision index constrained by the scheduling feasibility boundary. Among these: The scheduling feasibility boundary triggering mapping structure refers to a boundary mapping structure set between the non-additive fusion value sequence and the comprehensive scheduling decision index. It takes the boundary data corresponding to the scheduling constraints as input and limits the range of non-additive fusion values ​​in the scheduling space. Industrial load constraint boundaries refer to a set of load value restrictions determined by real-time power data, load change rate data, and operating constraint parameter data of industrial load. These restrictions include the upper limit of power, the lower limit of power, and the load change rate limit for industrial load. Generate a comprehensive scheduling decision index constrained by the scheduling feasibility boundary, specifically: The fusion values ​​corresponding to each operating condition component in the non-additive fusion value sequence are read, and the energy storage executable charge / discharge boundary values, industrial load constraint boundary values, and power balance boundary values ​​are read separately. The fusion values ​​are compared and calculated one by one with the corresponding boundary values. The fusion values ​​that exceed any boundary range are truncated and the truncated fusion values ​​are limited to the boundary range. After the boundary limitation is completed, the fusion values ​​corresponding to each operating condition component are combined and packaged according to the index types of photovoltaic available output, industrial load demand, and energy storage dispatchability to form a comprehensive dispatch decision index.

[0027] In this embodiment, obtaining the non-additive fusion value sequence includes: Based on the combined scheduling representation input, the cooperative weight vector, and the capacity function, a component reliability screening channel is introduced to perform operational condition component reliability screening processing. The cooperative weights corresponding to operational condition components with membership probabilities below a threshold are reduced, resulting in the target operational condition component set and the corresponding effective cooperative weight vector, where: The component credibility filtering channel refers to a filtering mapping structure set between the membership probability data of the operating condition component and the collaborative weight vector. It uses the membership probability value corresponding to the operating condition component as the filtering basis to limit the validity of the operating condition component in participating in subsequent non-additive fusion calculations at the current scheduling time step. Based on the combined scheduling representation input, effective collaborative weight vector, and capacity function corresponding to the target operating condition component set, and combined with the scheduling-aware Choquet fusion integral, a first-level non-additive fusion calculation is performed on the photovoltaic available supply level description, energy storage regulation capacity boundary description, and industrial load constraint urgency description within a single operating condition component, obtaining the intra-component fusion value corresponding to each target operating condition component, where: The execution of the first-layer non-additive fusion computation is as follows: For each target operating condition component, the description quantities of available photovoltaic supply level, energy storage regulation capacity boundary, and industrial load constraint urgency contained in the corresponding combined scheduling representation input are read. Simultaneously, the effective collaborative weight values ​​corresponding to the operating condition component and the capacity values ​​corresponding to the internal attribute combinations of the operating condition component in the capacity function are read. The combined scheduling representation input is arranged in order, and non-additive integral calculation is performed on the arranged attribute description quantities to obtain the corresponding intra-component fusion value, where: The non-additive integral calculation of the arranged attribute descriptors is specifically as follows: Within a single operating condition component, the photovoltaic available supply level description, energy storage regulation capacity boundary description, and industrial load constraint urgency description are sorted from largest to smallest value, and the first set, the first two sets, and the first three sets are constructed in sequence. The capacity values ​​of the corresponding sets in the capacity function are read respectively. The adjacent attribute values ​​after sorting are subtracted and segmented. The difference of each segment is multiplied by the capacity of the corresponding set and then accumulated. The accumulated result is used as the component fusion value of the operating condition component. Based on the component fusion value corresponding to each target operating condition component and the standardized industrial operation dataset under the current scheduling time step, the scheduling executability pruning process is performed. The component fusion value that exceeds the energy storage executable charging and discharging boundary and the rigid constraint boundary of industrial load is pruned to obtain the executable component fusion value corresponding to each target operating condition component. Based on the fusion values ​​within executable components and the priority sequence of operational components, a multi-granularity non-additive integration path is introduced. A second-layer non-additive fusion computation is performed using a scheduling-aware Choquet fusion integration method that operates in parallel at both coarse-grained and fine-grained segment levels, yielding a sequence of non-additive fusion values, where: Multi-granularity non-additive integration pathways refer to a parallel integration computing structure set up in the non-additive fusion computing process. The computational object of non-additive integration is divided into multiple integration sub-paths according to granularity, and the parallel computing relationship of integration sub-paths of different granularities is limited under the same scheduling time step. The execution of the second-layer non-additive fusion computation is as follows: Read the fusion value within the executable component corresponding to each target operating condition component, and determine the order of each operating condition component in the integration sequence according to the priority sequence of the operating condition components. In the multi-granularity non-additive integration path, simultaneously construct coarse-grained segment-level integration subsequence and fine-grained segment-level integration subsequence. Substitute the fusion value within the executable component into the corresponding granularity integration subsequence to perform non-additive integration calculation. Numerically aggregate the coarse-grained segment-level integration results and the fine-grained segment-level integration results to obtain the non-additive fusion value sequence.

[0028] In this embodiment, the process of obtaining the corresponding energy storage charging and discharging power scheduling command and photovoltaic power allocation scheduling command includes: Based on the comprehensive scheduling decision indicators and the standardized industrial operation dataset under the current scheduling time step, the target range of photovoltaic adjustable power, the target range of energy storage charging and discharging, and the target range of industrial load power supply demand are extracted to form a set of collaborative scheduling target parameters. Based on the set of collaborative scheduling target parameters, joint variables are constructed for photovoltaic power allocation and energy storage charging and discharging power. A scheduling constraint set is then constructed by combining operational constraint parameter data. This scheduling constraint set includes boundary constraints for energy storage state of charge, energy storage charging and discharging power, industrial load power supply, and photovoltaic output allocation. The joint variables for photovoltaic power allocation and energy storage charging / discharging power are constructed as follows: Read the photovoltaic adjustable power target range and energy storage charging and discharging target range corresponding to the set of coordinated scheduling target parameters, define the photovoltaic power allocation and energy storage charging and discharging power as scheduling decision variables, and combine and arrange the scheduling decision variables according to the scheduling time step order to form a joint decision variable vector; repeat the variable definition and combination process for each scheduling time step to obtain the set of joint decision variables; Energy storage charge state boundary constraints refer to a set of numerical restrictions consisting of the minimum and maximum values ​​of the charge state that energy storage is allowed to occur at the scheduling time step. Energy storage charge and discharge power boundary constraints refer to a set of numerical limits consisting of the maximum charging power and the maximum discharging power that energy storage is allowed to achieve at a scheduling time step. Industrial load power supply boundary constraints refer to a set of numerical limits consisting of the minimum and maximum power supply values ​​allowed for industrial loads at a given scheduling time step. Photovoltaic power output allocation boundary constraints refer to a set of numerical restrictions consisting of the minimum and maximum power output values ​​that can be allocated by photovoltaic power at a scheduling time step. Based on the set of collaborative scheduling target parameters and the set of scheduling constraints, collaborative optimization is performed on the photovoltaic power allocation and energy storage charging and discharging power to obtain a set of candidate scheduling instructions that satisfy the constraints at each scheduling time step. Then, a target scheduling instruction set is obtained, which includes energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions, wherein: A collaborative optimization solution is performed on the photovoltaic power allocation and the energy storage charging and discharging power, specifically as follows: The system reads the set of joint decision variables, the set of collaborative scheduling target parameters, and the set of scheduling constraints. It substitutes the joint decision variables into the expression of the scheduling optimization objective and iterates through the combinations of values ​​of the joint decision variables at each scheduling time step. During each update process, it verifies whether the values ​​of the joint decision variables satisfy the boundary constraints in the set of scheduling constraints. When the combination of values ​​of the joint decision variables satisfies all scheduling constraints, it records the scheduling result corresponding to the combination of values ​​and compiles them into a set of candidate scheduling instructions that satisfy the constraints at each scheduling time step.

[0029] The target scheduling instruction set obtained through filtering is as follows: In the candidate scheduling instruction set, the comprehensive scheduling decision index value corresponding to each candidate scheduling instruction is read, the comprehensive scheduling decision index value is sorted by size, and the scheduling instruction whose index value meets the selection condition in the sorting result is selected as the target scheduling instruction. The selected scheduling instructions are encapsulated in the order of scheduling time steps to form the target scheduling instruction set. The target scheduling instruction set is sent to the photovoltaic-side execution equipment and the energy storage-side execution equipment for execution.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a large-scale equipment manufacturing industrial park in an eastern coastal region. This park is equipped with a rooftop distributed photovoltaic system and a centralized energy storage system to support the electricity needs of production lines, auxiliary equipment, and public facilities. The park's installed photovoltaic capacity is 5.2MW, and the rated capacity of the energy storage system is 3.6MWh. The industrial load exhibits significant fluctuations during weekdays, with peak loads reaching 6.8MW, and the rate of load change increases significantly during peak hours.

[0031] In the actual operation of the industrial park, the method of this invention was applied to 60 consecutive days of industrial energy operation data. Data such as power output from the power supply side, energy storage state of charge, energy storage charging and discharging power, real-time industrial load power, load change rate, and operating constraint parameters were collected and analyzed in 15-minute scheduling time steps. Through multi-condition modeling and non-additive collaborative fusion, the scheduling system can adjust the energy storage charging and discharging strategy in advance during the midday period when photovoltaic output fluctuates significantly, maintaining the continuity of energy storage discharge power during periods of rapid load increase. Actual operation results show that after applying the method of this invention, the number of energy storage charging and discharging switches decreased from an average of 18 times per day to 11 times, the photovoltaic curtailment rate decreased from 7.4% to 3.1%, and the power supply deviation during peak industrial load periods decreased by approximately 32%. No scheduling commands failed to execute throughout the entire implementation period, and the stability of the scheduling results was significantly improved, verifying the effectiveness and feasibility of this invention in achieving photovoltaic and energy storage collaborative optimization scheduling in complex industrial scenarios.

[0032] Table 1 Comparison of the Effects of Photovoltaic Energy Storage Collaborative Optimization Scheduling in Industrial Parks

[0033] As shown in Table 1, under the same photovoltaic (PV) installed capacity and energy storage configuration, the method of this invention demonstrates significantly better performance than the traditional dispatching method in several key operational indicators. Regarding PV operation, although both methods have an installed capacity of 5.2MW, the average daily PV power generation under the method of this invention increases to 19.8MWh, approximately 6.45% higher than the 18.6MWh of the traditional dispatching method. Simultaneously, the PV curtailment rate significantly decreases from 7.4% to 3.1%, a reduction of over 58%. This indicates that the present invention can more fully absorb PV output while meeting system constraints, reducing curtailment caused by dispatch mismatch.

[0034] At the energy storage operation level, the method of this invention effectively reduces the scheduling impact intensity of the energy storage system. The average number of daily charge-discharge switching times is reduced from 18 to 11, a reduction of nearly 40%, and the equivalent full charge-discharge times are also reduced from 1.42 to 1.21, indicating that the energy storage operation is smoother and avoids the lifespan loss caused by frequent charge-discharge.

[0035] From the perspective of overall scheduling stability and comprehensive operational effectiveness, the method of this invention did not experience any scheduling command execution failures within the scheduling cycle, significantly reduced the number of power over-constraint occurrences, and improved the comprehensive scheduling stability score from 0.73 to 0.89. The comprehensive energy efficiency improvement rate on the industrial power consumption side reached 4.6%, fully demonstrating that the method of this invention, while ensuring system safety and constraint feasibility, achieved coordinated and optimized scheduling among photovoltaic, energy storage, and industrial loads, and possesses significant practical application value and engineering promotion significance.

[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis, characterized in that, include: Collect industrial energy operation data in industrial scenarios, preprocess the industrial energy operation data, and form a standardized industrial operation dataset; Based on standardized industrial operation datasets, feature vectors of industrial operation samples are constructed, and the characteristics and coupling relationships of photovoltaic, energy storage and industrial load operation status are encapsulated in a structured manner to form an industrial sample set; An improved kernel Gaussian mixture ridge regression model is constructed. The improved Gaussian mixture model is used to fit the industrial sample set to obtain the set of operating condition components and their corresponding membership probabilities. The corresponding industrial sample features are mapped to the kernel feature space and ridge regularization constraints are introduced. Regression modeling is performed on photovoltaic available output, industrial load demand and energy storage dispatchable capacity respectively to obtain the corresponding component-level prediction results. Based on the component-level prediction results, a component-level schedulable capacity vector is constructed, and a capacity function is constructed based on historical industrial operation data and the characteristics of operation status changes within the current scheduling cycle. Based on the capacity function and component-level schedulable capability vectors, a working condition coordination fusion module based on Choquet integral is constructed to perform non-additive fusion on the component-level schedulable capability vectors of each operating condition component to obtain comprehensive scheduling decision indicators. Based on the comprehensive scheduling decision indicators, the corresponding energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions are obtained and then issued for execution.

2. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The industrial energy operation data specifically includes power output data on the electric side, energy storage state of charge data, energy storage charging and discharging power data, real-time power data of industrial load, load change rate data, and operation constraint parameter data.

3. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The preprocessing of industrial energy operation data specifically includes timestamp unification, missing value completion, outlier removal, and dimension normalization.

4. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The formation of the industrial sample set includes: At each scheduling time step, the industrial energy operation data at the corresponding time is read from the standardized industrial operation dataset and used as the basic operation data for the current time step; Based on standardized industrial energy operation data, the operation status data reflecting the photovoltaic output level and energy storage operation status are aggregated and packaged to form photovoltaic energy storage operation status characteristics, and the operation status data reflecting industrial electricity demand and its changing characteristics are aggregated and packaged to form industrial load operation status characteristics. Based on the operating status characteristics of photovoltaic energy storage and industrial load, a coupling relationship feature is constructed to characterize the mutual influence between photovoltaic output, energy storage regulation capability and industrial load constraint intensity. The photovoltaic energy storage operation status characteristics, industrial load operation status characteristics, and coupling relationship characteristics under the same scheduling time step are uniformly spliced ​​together to generate the industrial operation sample feature vector of the corresponding time step. The industrial operation sample feature vectors under each scheduling time step are then collected in chronological order to form an industrial sample set.

5. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The process of obtaining the corresponding component-level prediction result includes: An improved kernel Gaussian mixture ridge regression model is constructed to extract the distribution feature data of each industrial operation sample feature vector at each scheduling time step. The Gaussian mixture model is used to perform fitting processing on the distribution feature data. A component splitting and merging evolution gate is introduced to perform adaptive splitting and merging update processing on the operating condition components to generate a set of operating condition components. Based on the set of operating condition components, the membership probability of each industrial operating sample corresponding to each operating condition component is calculated, and each industrial operating sample is assigned to the corresponding operating condition component according to the size of the membership probability, forming a subset of industrial samples corresponding to each operating condition component. For each industrial sample subset corresponding to each operating condition component, a component boundary shell structure is introduced. Based on the operating constraint parameter data and energy storage operating boundary conditions, a set of boundary samples close to the operating constraint boundary is generated from each industrial sample subset, thus obtaining the component boundary shell and the component main sample set. For each operating condition component, kernel function mapping is performed on the main sample set and the boundary shell of the component to obtain the corresponding kernel feature space representation. Ridge regularization constraint is introduced in the kernel feature space. The target quantity corresponding to the industrial energy operation data is used as the regression object. Regression modeling is performed on the available photovoltaic output, industrial load demand and energy storage dispatchability to obtain the corresponding component-level regression results under each operating condition component. Based on the component-level regression results, a component-level prediction result description set containing the prediction center value, the executable boundary descriptor, and the prediction stability descriptor is generated for each operating condition component, forming the corresponding component-level prediction result.

6. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The capacity function is constructed as follows: Based on the component-level prediction results corresponding to each operating condition component and the standardized industrial operation dataset under the current scheduling time step, the component-level prediction results of energy storage dispatchable capacity corresponding to each operating condition component are mapped to the executable charging and discharging power boundary descriptor. At the same time, the component-level prediction results of industrial load demand are mapped to the load rigidity constraint urgency descriptor, and the component-level prediction results of photovoltaic available output are mapped to the available supply level descriptor, thus obtaining the component-level dispatchable capacity vector corresponding to each operating condition component. Based on historical industrial operation data and standardized industrial operation datasets within the current scheduling cycle, the changes in power output on the power side, the changes in energy storage state of charge, and the changes in industrial load power are jointly extracted and processed to obtain characteristic values ​​of the change in operating state. Based on the characteristic values ​​of the change in operating status and the membership probability of each operating condition component at the current scheduling time step, the contribution relationship is updated, the collaborative contribution relationship between operating condition components in historical industrial operating data is updated, and the capacity function at the current scheduling time step is generated.

7. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The obtained comprehensive scheduling decision indicators include: A working condition coordination and fusion module based on Choquet integral is constructed. The working condition coordination and fusion module includes a working condition scheduling representation unit, a coordination weight adjustment unit, a non-additive fusion unit, and a scheduling index generation unit. The operating condition scheduling characterization unit, based on the component-level schedulable capability vector corresponding to each operating condition component, performs attribute cross-combination encapsulation of the photovoltaic available supply level description, the energy storage regulation capacity boundary description, and the industrial load constraint urgency description to generate the combined scheduling characterization input within the operating condition component, and arranges each operating condition component in an ordered manner to obtain the operating condition component priority sequence. The collaborative weight adjustment unit, based on the priority sequence of operating condition components, the membership probability and capacity function of each operating condition component at the current scheduling time step, introduces a collaborative contribution constraint mapping structure, performs constraint mapping processing on the collaborative contribution relationship between operating condition components, performs suppression mapping and maintenance mapping on collaborative weights, and generates a collaborative weight vector corresponding to each operating condition component. The non-additive fusion unit is based on the combined scheduling representation input and the corresponding collaborative weight vector. Combined with the scheduling-aware Choquet fusion integral, a hierarchical non-additive integral path is introduced. The first layer of non-additive fusion calculation is performed within a single operating condition component to obtain the component fusion value corresponding to each operating condition component. Based on the component fusion value and the operating condition component priority sequence, the second layer of non-additive fusion calculation is performed to obtain the non-additive fusion value sequence. The scheduling index generation unit is based on the non-additive fused value sequence and introduces a scheduling feasibility boundary triggering mapping structure. It associates and maps the non-additive fused value sequence with the energy storage executable charge and discharge boundary, industrial load constraint boundary and power balance boundary. It encapsulates the available photovoltaic output, industrial load demand and energy storage scheduling capability in a unified way to generate a comprehensive scheduling decision index constrained by the scheduling feasibility boundary.

8. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 7, characterized in that, The obtained non-additive fusion value sequence includes: Based on the combined scheduling representation input, the collaborative weight vector and the capacity function, a component credibility screening channel is introduced to perform the operational condition component credibility screening process. The collaborative weights corresponding to operational condition components with membership probabilities below the threshold are reduced to obtain the target operational condition component set and the corresponding effective collaborative weight vector. Based on the combined scheduling representation input, effective collaborative weight vector and capacity function corresponding to the target operating condition component set, and combined with the scheduling-aware Choquet fusion integral, the first-level non-additive fusion calculation is performed on the photovoltaic available supply level description, energy storage regulation capacity boundary description and industrial load constraint urgency description within a single operating condition component to obtain the component fusion value corresponding to each target operating condition component. Based on the component fusion value corresponding to each target operating condition component and the standardized industrial operation dataset under the current scheduling time step, the scheduling executability pruning process is performed. The component fusion value that exceeds the energy storage executable charging and discharging boundary and the rigid constraint boundary of industrial load is pruned to obtain the executable component fusion value corresponding to each target operating condition component. Based on the fusion values ​​within executable components and the priority sequence of operating condition components, a multi-granularity non-additive integration path is introduced. The second-layer non-additive fusion calculation is performed using a scheduling-aware Choquet fusion integration method that is parallel at both the coarse-grained and fine-grained segment levels, resulting in a sequence of non-additive fusion values.

9. The photovoltaic energy storage collaborative optimization scheduling method based on industrial data analysis according to claim 1, characterized in that, The solution yields the corresponding energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions, including: Based on the comprehensive scheduling decision indicators and the standardized industrial operation dataset under the current scheduling time step, the target range of photovoltaic adjustable power, the target range of energy storage charging and discharging, and the target range of industrial load power supply demand are extracted to form a set of collaborative scheduling target parameters. Based on the set of collaborative scheduling target parameters, joint variables are constructed for photovoltaic power allocation and energy storage charging and discharging power, and a set of scheduling constraints is constructed by combining the operational constraint parameter data. The set of scheduling constraints includes boundary constraints for energy storage charge state, boundary constraints for energy storage charging and discharging power, boundary constraints for industrial load power supply, and boundary constraints for photovoltaic output allocation. Based on the set of collaborative scheduling target parameters and the set of scheduling constraints, collaborative optimization is performed on the photovoltaic power allocation and energy storage charging and discharging power to obtain a set of candidate scheduling instructions that meet the constraints at each scheduling time step, and then the target scheduling instruction set is obtained by screening. The target scheduling instruction set includes energy storage charging and discharging power scheduling instructions and photovoltaic power allocation scheduling instructions. The target scheduling instruction set is sent to the photovoltaic-side execution equipment and the energy storage-side execution equipment for execution.

Images