Optimal Scheduling System and Method for Photovoltaic-Storage-Charge Collaboration Based on Multi-Source Data Fusion

Abstract

The optical storage charge collaborative optimization scheduling system and method based on multi-source data fusion of the present application relate to the technical field of photovoltaic energy storage optimization. By obtaining historical photovoltaic power generation data, load power consumption data, and meteorological data, a photovoltaic power generation scenario tree model and a load prediction model are constructed; considering the operation revenue and energy storage aging cost, a first optimization target and a second optimization target are constructed. Based on the photovoltaic power generation scenario tree model, considering the uncertainty of photovoltaic power, combined with the load demand predicted by the load prediction model, an adaptive optimization scheduling model is established; the photovoltaic power generation scenario tree model and the load prediction model are embedded into the adaptive optimization scheduling model to solve the optimal energy storage charge and discharge strategy for the current period, achieving the improvement of the robustness and reliability of energy storage scheduling while ensuring the economy of photovoltaic power generation.

Description

Technical Field

[0001] This application relates to the field of photovoltaic energy storage optimization, and particularly to a coordinated optimization scheduling system and method for photovoltaic energy storage and load based on multi-source data fusion. Background Art

[0002] Coordinated optimization scheduling of photovoltaic energy storage and load refers to an advanced scheduling method in the power system that improves the utilization efficiency of renewable energy, smooths the load curve, and reduces the operating cost of the power system by coordinating and optimizing photovoltaic power generation, energy storage systems, and load demands. This research field involves multiple disciplines such as power systems, renewable energy, energy storage technologies, and optimization algorithms, and has broad application prospects.

[0003] Photovoltaic power generation has developed rapidly globally due to its advantages such as cleanliness and sustainability. However, photovoltaic power generation also faces problems such as intermittency and large fluctuations, posing challenges to grid dispatching and operation. Energy storage systems can, to a certain extent, smooth the fluctuations of photovoltaic power generation and improve its utilization efficiency through reasonable charging and discharging. At the same time, with the development of demand-side management technologies, some load demands have a certain degree of flexibility and can respond to the power supply and demand balance through optimized scheduling. Coordinated optimization scheduling of photovoltaic energy storage and load is to formulate the optimal scheduling strategies for each subsystem based on comprehensive consideration of photovoltaic power generation prediction, energy storage system characteristics, and load demand response capabilities, so as to maximize the benefits of the entire system.

[0004] Chinese Patent with application publication number CN119602358A discloses a multi-energy virtual power plant optimization scheduling method considering source-load coordination, including data collection, data preprocessing, establishing models for wind turbines, photovoltaic units, gas turbines, distributed energy storage, and demand-side response, establishing a two-layer optimization scheduling model, and solving the two-layer optimization scheduling model to obtain the optimized scheduling results.

[0005] Most of the existing technologies ignore the aging and lifespan problems of energy storage devices, resulting in deviations between the optimization results and actual operation. How to optimize the charging and discharging strategies of energy storage power stations while considering the aging and lifespan loss of energy storage devices to maximize their economic benefits and operation reliability has become an urgent problem to be solved. Summary of the Invention

[0006] This application aims to solve at least one of the technical problems in the related technologies to some extent. For this reason, one objective of this application is to propose a coordinated optimization scheduling system and method for photovoltaic energy storage and load based on multi-source data fusion, which realizes improving the robustness and reliability of energy storage scheduling while ensuring the economy of photovoltaic power generation.

[0007] One aspect of this application provides a coordinated optimization scheduling method for photovoltaic energy storage and load based on multi-source data fusion, including:

[0008] Step S100: Obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and construct a photovoltaic power generation scenario tree model and a load forecasting model;

[0009] Step S200: Consider the operating revenue and energy storage aging cost to construct the first optimization objective and the second optimization objective. Based on the photovoltaic power generation scenario tree model, consider the uncertainty of photovoltaic power generation, and combine with the load demand predicted by the load forecasting model to establish an adaptive optimal scheduling model;

[0010] Step S300: Embed the photovoltaic power generation scenario tree model and the load forecasting model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current period;

[0011] The specific method for obtaining historical photovoltaic power generation data, load power consumption data, and meteorological data, and constructing a photovoltaic power generation scenario tree model and a load forecasting model is as follows:

[0012] Step S110: Obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and perform preprocessing;

[0013] The historical photovoltaic power generation data includes power generation power, inverter temperature, and power generation amount data;

[0014] Step S120: Extract features from the historical photovoltaic power generation data and meteorological data to obtain input features. According to the input features and historical photovoltaic power generation data, construct a photovoltaic power generation scenario tree model. The photovoltaic power generation scenario tree model divides the input features into N sub-regions, and each sub-region corresponds to a photovoltaic power generation scenario;

[0015] Step S130: Define a deviation metric index and set a deviation threshold. When the deviation metric index MA is greater than the deviation threshold, it is considered that a significant deviation has occurred, and update the parameters of the photovoltaic power generation scenario tree model based on the current photovoltaic power generation data and meteorological data;

[0016] Step S140: Construct a load forecasting model based on the load power consumption data to predict the future load demand;

[0017] The specific method of the preprocessing includes: triggering data cleaning according to the power station alarm status, identifying and removing the power generation power, inverter temperature, and power generation amount data under abnormal power generation states; triggering data cleaning according to the electrical maintenance records, marking or removing the power generation power, inverter temperature, and power generation amount data during the power station maintenance period; detecting outliers in the load power consumption data and removing the load power consumption data in abnormal power consumption states;

[0018] The specific method for constructing a photovoltaic power generation scenario tree model according to the input features and historical photovoltaic power generation data is as follows:

[0019] Step S121: Select the long short - term memory network as the initial network, construct training samples based on the sliding window method, preset the time window size, and extract input features based on historical photovoltaic power generation data and meteorological data within each time window;

[0020] Step S122: Preset the prediction step length T. For each time window, extract the power generation power within the prediction step length of the future time as the output label, and combine the input features and output labels extracted from each time window into a training sample;

[0021] Step S123: Slide the time window to obtain a training sample set, train the photovoltaic power generation scenario tree model based on the training sample set, and divide the input features into N sub - regions in a recursive partitioning manner, with each sub - region corresponding to a photovoltaic power generation scenario;

[0022] The training process of the photovoltaic power generation scenario tree model is as follows:

[0023] Step S123.1: Starting from the root node, recursively partition the training sample set of the current node, select a partitioning feature from the input features, and divide the training sample set into a high - feature set and a low - feature set to obtain a photovoltaic power generation scenario tree containing N sub - regions;

[0024] Step S123.2: For each sub - region, count the number of training samples falling into this node, and calculate its proportion in the total number of training samples as the occurrence probability of this photovoltaic power generation scenario;

[0025] Step S123.3: For each sub - region, count the distribution parameters of the power generation power under each photovoltaic power generation scenario;

[0026] Step S123.4: Use the training sample set to optimize the parameters of the photovoltaic power generation scenario tree model by minimizing the loss function; when the value of the loss function converges, the training is completed;

[0027] Step S123.5: Use the trained photovoltaic power generation scenario tree model to predict the occurrence probability and the power generation power distribution parameters (the distribution parameters include the mean and variance) of the photovoltaic power generation scenario to which the new input features belong;

[0028] The update method for the parameters of the photovoltaic power generation scenario tree model is as follows:

[0029] Calculate the adaptive learning rate according to the gradient of the model parameters at each update;

[0030] Introduce a forgetting factor , when a significant deviation occurs, based on the adaptive learning rate of the previous update , the gradient of the model parameters and the forgetting factor , calculate the value of the model parameters at the current update ;

[0031] The construction and training method of the load forecasting model is as follows:

[0032] Step S141: Extract load characteristics based on historical load power consumption data, construct a load characteristic clustering model, divide the load characteristics into different load patterns, and each clustering category in the load curve clustering model represents a load pattern;

[0033] Step S142: Train the corresponding load forecasting sub-model for each load pattern;

[0034] Step S143: Combine each load forecasting sub-model into a load forecasting model for predicting future load demands ;

[0035] The load forecasting sub-model uses the historical load power consumption data as training data, forms a time series in chronological order, uses the LSTM model as the initial model, and uses the sliding window method to construct training samples on the time series. In each training sample, the load power consumption data of the current window is used as input data, and the load power consumption data on the future time series of the current window is predicted as the load demand.

[0036] The specific method for constructing the first optimization objective and the second optimization objective considering the operation revenue and the energy storage aging cost, considering the photovoltaic uncertainty based on the photovoltaic power generation scenario tree model, and establishing the adaptive optimal scheduling model in combination with the load demand predicted by the load forecasting model is as follows:

[0037] Step S210: Input the energy storage device parameters, the energy storage aging model parameters, the electricity price information, and the load demand;

[0038] Step S220: Establish a dynamic model of the energy storage device between the state variable SOC and the output variable charge-discharge power based on the energy storage device parameters ;

[0039] Step S230: Construct a decay model of the maximum energy storage capacity based on the energy storage aging model parameters and an internal resistance increase model ;

[0040] Step S240: Define the operation revenue and the energy storage aging cost based on the energy storage device parameters, the energy storage aging model parameters, the electricity price information, and the load demand, and construct the first optimization objective and the second optimization objective;

[0041] Step S250: Construct the uncertainty constraint of the photovoltaic power generation power based on the photovoltaic power generation scenario tree model;

[0042] Step S260: Introduce a risk measurement index to quantify the risk loss function of the uncertainty constraint of photovoltaic power generation ;

[0043] Step S270: Construct constraint conditions based on the dynamic model, attenuation model, internal resistance increase model of the energy storage device, load demand, and risk loss function, and construct an adaptive optimal scheduling model based on the first optimization goal and the second optimization goal;

[0044] Specifically, the construction methods of the first optimization goal and the second optimization goal are as follows:

[0045] Step S241: Based on the electricity price information, which includes the electricity selling price, electricity purchasing price, energy storage charging and discharging prices, take maximizing the operating revenue as the first optimization goal ;

[0046] Step S242: Based on the internal resistance increase model, attenuation model, energy storage aging model parameters, and load demand, take minimizing the energy storage aging cost as the second optimization goal ;

[0047] Specifically, the method for constructing the uncertainty constraint of photovoltaic power generation based on the photovoltaic power generation scenario tree includes:

[0048] Step S251: For each photovoltaic power generation scenario and time period in the photovoltaic power generation scenario tree model, obtain the corresponding power generation power distribution parameters;

[0049] Step S252: Describe the uncertainty constraint of photovoltaic power generation based on the power generation power distribution parameters. The specific method is: for the th time period, define the uncertainty parameter vector of photovoltaic power generation; describe the uncertainty constraint of the uncertainty parameter vector based on the ellipsoid set;

[0050] The adaptive optimal scheduling model specifically includes: using the energy storage device dynamic model as the energy storage dynamic constraint, the attenuation model and the internal resistance increase model as the energy storage aging cost constraint, the linear constraint for calculating the risk measurement index based on the risk loss function, using the risk loss function as the risk constraint, using the load demand as the load balance constraint, and constructing an adaptive optimal scheduling model based on the first optimization goal, the second optimization goal, and the risk measurement index;

[0051] The specific method for embedding the photovoltaic power generation scenario tree model and the load forecasting model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current time period is:

[0052] Step S310: Embed the photovoltaic power generation scenario tree model into the adaptive optimal scheduling model, and generate an optimization sub-problem for the predicted value and uncertainty constraint of the power generation power corresponding to each photovoltaic power generation scenario;

[0053] Step S320: Embed the load prediction model into the adaptive optimization scheduling model, and introduce the predicted load demand as a known parameter into the optimization sub-problem;

[0054] Step S330: Perform weighted summation on each optimization sub-problem to obtain the expected objective function of the overall optimization problem;

[0055] Step S340: Solve the expected objective function of the overall optimization problem to generate the optimal energy storage charge and discharge strategy for the current time period, where the optimal energy storage charge and discharge strategy includes the optimal charging and discharging power;

[0056] Step S350: Adopt a rolling optimization mechanism to regularly roll the optimization time domain forward; repeat steps S310 - S340 on the new optimization time domain, and update the optimal energy storage charge and discharge strategy according to the latest predicted value of the photovoltaic power generation;

[0057] Step S360: Output the optimal energy storage charge and discharge strategy for each optimization time domain to form a complete energy storage optimization scheduling plan.

[0058] One aspect of the present application provides a photovoltaic - energy storage - load collaborative optimization scheduling system based on multi - source data fusion, including:

[0059] A scenario tree model construction module, configured to obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and construct a photovoltaic power generation scenario tree model and a load prediction model;

[0060] An adaptive model construction module, configured to construct a first optimization target and a second optimization target considering operating revenue and energy storage aging cost, and establish an adaptive optimization scheduling model based on the photovoltaic power generation scenario tree model considering photovoltaic uncertainty and combining the load demand predicted by the load prediction model;

[0061] A charge - discharge strategy output module, configured to embed the photovoltaic power generation scenario tree model and the load prediction model into the adaptive optimization scheduling model, and solve the optimal energy storage charge and discharge strategy for the current time period.

[0062] One aspect of the present application provides a readable storage medium, which stores a computer program, and the computer program is suitable for being loaded by a processor to execute the steps in the photovoltaic - energy storage - load collaborative optimization scheduling method based on multi - source data fusion.

[0063] The photovoltaic - energy storage - load collaborative optimization scheduling system and method proposed by the present application have the following advantages compared with the prior art:

[0064] This application innovatively introduces a scenario tree model. By recursively partitioning, the photovoltaic output is divided into multiple discrete scenarios, each scenario corresponding to a power generation power distribution and an occurrence probability, which better captures the uncertainty and volatility of photovoltaic output in time and space, and improves the fineness and comprehensiveness of modeling.

[0065] Aiming at the problem of large differences in load curves for different users and different periods, this application proposes a load forecasting method with different modes. By clustering to divide different load patterns, corresponding prediction sub-models are trained for each pattern, and finally the sub-models are combined to form an overall load forecasting model. Compared with a single model, the multi-mode model can more accurately capture the load characteristics under different modes.

[0066] This application constructs a first optimization objective and a second optimization objective including energy storage aging cost and operation revenue, balances the operation revenue and the energy storage life through a weighting coefficient, and at the same time introduces a risk measurement index to quantify the impact of photovoltaic uncertainty, realizing the coordinated optimization of energy storage aging and photovoltaic uncertainty, making the scheduling scheme more comprehensive and practical.

[0067] This application introduces a risk measurement index. Through a risk loss function and risk constraints, the risk preference of the optimization result can be flexibly controlled, avoiding extreme risks while pursuing revenue, and realizing robust optimization based on risk measurement.

[0068] This application adopts an adaptive data-driven optimization framework. By real-time updating the photovoltaic power generation scenario tree model and the rolling optimization scheduling strategy, it can adapt to the changes in photovoltaic output and improve the robustness of the optimal scheduling. At the same time, through methods such as data preprocessing, feature extraction, and deviation detection, the accuracy and applicability of the scenario tree model are ensured.

[0069] All links of this application cooperate with each other to form a closed-loop adaptive scheduling system. Photovoltaic power generation prediction provides the power upper limit of multiple scenarios for energy storage optimal scheduling, reducing uncertainty. Load demand prediction provides accurate demand-side information for energy storage optimal scheduling, realizing source-load interaction. Under the two-way drive of source-side information and load-side information, energy storage optimal scheduling weighs multiple objectives such as revenue, cost, and risk, generates an optimal strategy that takes into account both the current and the long term, and the rolling optimization mechanism ensures that the scheduling scheme can be updated in real time with new information to achieve continuous improvement. Description of the Drawings

[0070] Figure 1 It is the method flow chart of the photovoltaic-storage-load coordinated optimal scheduling method based on multi-source data fusion provided by this application;

[0071] Figure 2 It is the method flow chart for constructing the photovoltaic power generation scenario tree model provided by this application;

[0072] Figure 3 Flowchart of the optimal energy storage charge and discharge strategy solving method provided by this application;

[0073] Figure 4 Software interface diagram of the historical total power generation curve of the photovoltaic operation software provided by this application;

[0074] Figure 5 Software interface diagram of the historical total power generation curve of the photovoltaic operation software provided by this application;

[0075] Figure 6 Functional module diagram of the optical storage charge collaborative optimization scheduling system based on multi-source data fusion provided by this application. Detailed implementation manners

[0076] To better understand this application, various aspects of this application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of the exemplary embodiments of this application, and do not limit the scope of this application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0077] In the drawings, for ease of illustration, the size, dimensions and shape of the elements have been slightly adjusted. The drawings are only examples and are not drawn to an exact scale. As used herein, terms such as "substantially", "about" and similar terms are used as terms indicating approximation, rather than terms indicating degree, and are intended to account for the inherent deviations in measured or calculated values that would be recognized by a person of ordinary skill in the art. Additionally, in this application, the order of description of the various steps does not necessarily represent the order in which these processes occur in actual operation, unless otherwise clearly defined or derivable from the context.

[0078] It should also be understood that expressions such as "including", "having", "comprising", "containing" and / or "containing" are open-ended rather than closed-ended expressions in this specification, which means that there are the stated features, elements and / or components, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when an expression such as "at least one of..." appears after a list of listed features, it modifies the entire list of features, rather than just an individual element in the list. Further, when describing the embodiments of this application, the use of "may" means "one or more embodiments of this application". And the term "exemplary" is intended to refer to an example or illustration.

[0079] Unless otherwise defined, all terms used herein (including engineering and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that, unless clearly stated in this application, words defined in a commonly used dictionary should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and should not be interpreted in an idealized or overly formal sense.

[0080] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0081] Embodiment 1

[0082] As Figure 1 shown, the present application provides a method for collaborative optimization scheduling of optical storage and charge based on multi-source data fusion, including:

[0083] Step S100: Obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and construct a photovoltaic power generation scenario tree model and a load prediction model;

[0084] The specific method for obtaining historical photovoltaic power generation data, load power consumption data, and meteorological data and constructing a photovoltaic power generation scenario tree model and a load prediction model is as follows:

[0085] Step S110: Obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and perform preprocessing;

[0086] The historical photovoltaic power generation data includes power generation power, inverter temperature, and power generation amount data;

[0087] The power generation power reflects the real-time power generation power of the photovoltaic power station at a certain moment, and is one of the core input data for training the scenario tree model.

[0088] The inverter temperature reflects the operating state and health level of the inverter device, and is closely related to the photovoltaic power generation efficiency and reliability. The inverter temperature is used as an auxiliary input feature of the scenario tree model to describe the influence of the working condition of the power generation equipment on the power generation power. By analyzing the relationship between the inverter temperature and the power generation power, problems such as equipment abnormalities and efficiency decline can be found, providing a basis for optimizing operation and maintenance.

[0089] The generated electricity data includes total generated electricity, monthly generated electricity, and daily generated electricity, which reflect the cumulative generated electricity of the PV power station at different time scales and incorporate the time cumulative effect of the power generation power. The generated electricity data is used as an auxiliary input feature of the scenario tree model to characterize the long-term power generation capacity and efficiency level of the PV power station. By analyzing the generated electricity data in different periods, the changing trend of the power generation capacity of the PV power station can be found, providing a reference for optimizing production operations.

[0090] The generated electricity data is extracted based on the characteristics of the historical total generated electricity curve.

[0091] The load electricity consumption data includes the electricity consumption load data on the user side and the electricity consumption characteristics of different types of loads.

[0092] As Figure 4 shown, it is the software interface diagram of the historical total generated electricity curve of the PV operation software provided by this application. The historical total generated electricity, monthly generated electricity, and daily generated electricity are visualized in the form of bar charts and curves. In addition, Figure 4 the information on the inverter temperature is also displayed.

[0093] The specific methods of the preprocessing include: triggering data cleaning according to the alarm status of the power station, identifying and eliminating the power generation power, inverter temperature, and generated electricity data under abnormal power generation states; triggering data cleaning according to the electrical maintenance records, marking or eliminating the power generation power, inverter temperature, and generated electricity data during the power station maintenance period; detecting outliers in the load electricity consumption data and eliminating the load electricity consumption data in abnormal electricity consumption states.

[0094] The judgment formula for the abnormal power generation state is:

[0095] ;

[0096] where is the abnormal power generation state indication variable, is the alarm status of the power station at time , is the predefined set of abnormal power generation states, represents a fault, represents an alarm.

[0097] When , it indicates that the power station is in an abnormal power generation state, and the corresponding PV power generation data should be eliminated.

[0098] The judgment formula for the maintenance period is:

[0099] ;

[0100] where is the maintenance period indication variable, is the current moment, and are the start time and end time in the electrical maintenance record respectively.

[0101] When it indicates that the power station is under maintenance, and the corresponding photovoltaic power generation data should be excluded.

[0102] The judgment formula for the abnormal power consumption state is:

[0103]

[0104] where is the abnormal power consumption state indication variable, is the power consumption alarm state of the user side at time and is the predefined set of abnormal power consumption states, represents a power consumption fault, represents a power consumption repair.

[0105] As Figure 5 shown, it is the operation and maintenance interface diagram of the photovoltaic operation software provided by this application. The power station, alarm state and electrical maintenance record are displayed in a visual way. Specifically, when abnormal states such as faults and alarms occur in the power station, or when abnormal states such as overheating and faults occur in the inverter, abnormal values or invalid values may appear in the power generation power, inverter temperature and power generation amount data during the corresponding time period, which need to be identified and excluded; during the planned maintenance or fault repair period of the power station, or during the operations such as maintenance, repair or replacement of the inverter, the power generation power, inverter temperature and power generation amount data may be affected and cannot be collected normally or reflect the actual power generation situation, and the data during the corresponding time period needs to be marked or excluded according to the maintenance record.

[0106] Step S120: Extract features from historical photovoltaic power generation data and meteorological data to obtain input features. According to the input features and historical photovoltaic power generation data, construct a photovoltaic power generation scenario tree model. The photovoltaic power generation scenario tree model divides the input features into N sub-regions, and each sub-region corresponds to a photovoltaic power generation scenario;

[0107] The photovoltaic power generation scenario tree model selects a long short-term memory network as the initial network, and predicts the photovoltaic power generation scenario in the future for a period of time according to the input features. The predicted photovoltaic power generation scenario includes the probability of the occurrence of the photovoltaic power generation scenario and the power generation power distribution under the photovoltaic power generation scenario;

[0108] The probability of the occurrence of the photovoltaic power generation scenario reflects the likelihood of the scenario under the given input conditions.

[0109] The power generation power distribution in the photovoltaic power generation scenario depicts the probabilities of different power generation power values occurring in this scenario.

[0110] The specific method for constructing a photovoltaic power generation scenario tree model based on input features and historical photovoltaic power generation data is as follows:

[0111] Step S121: Select a long short-term memory network as the initial network, construct training samples based on the sliding window method, preset the time window size, and extract input features based on historical photovoltaic power generation data and meteorological data within each time window;

[0112] The selection of the time window size is set by those skilled in the art by weighing factors such as the prediction time scale, the time granularity of the data, and the complexity of the model. Exemplarily, if it is necessary to predict the power generation power in the next 1 hour, 1 hour can be selected as the time window size.

[0113] Step S122: Preset the prediction step length T. For each time window, extract the power generation power within the prediction step length of the future time as the output label, and combine the input features and output labels extracted from each time window into a training sample;

[0114] Step S123: Slide the time window to obtain a training sample set, train the photovoltaic power generation scenario tree model based on the training sample set, and divide the input features into N sub-regions in a recursive partitioning manner, with each sub-region corresponding to a photovoltaic power generation scenario;

[0115] The training process of the photovoltaic power generation scenario tree model is as follows:

[0116] Step S123.1: Starting from the root node, recursively partition the training sample set of the current node, select a partitioning feature from the input features, and divide the training sample set into a high feature set and a low feature set to obtain a photovoltaic power generation scenario tree containing N sub-regions;

[0117] Step S123.2: For each sub-region, count the number of training samples falling into this node, and calculate the proportion of it in the total number of training samples as the occurrence probability of this photovoltaic power generation scenario;

[0118] Step S123.3: For each sub-region, count the power generation power distribution parameters under each photovoltaic power generation scenario;

[0119] Step S123.4: Use the training sample set to optimize the parameters of the photovoltaic power generation scenario tree model by minimizing the loss function; when the value of the loss function converges, the training is completed;

[0120] The scenario tree model provides fine-grained PV output scenarios for subsequent optimal scheduling, and considers extreme cases, enabling the optimal scheduling scheme to adapt to various possible situations and greatly improving the robustness of the scheduling strategy.

[0121] The loss function uses the mean square error;

[0122] Step S123.5: Use the trained PV power generation scenario tree model to predict the occurrence probability of the PV power generation scenario to which the new input features belong and its power generation power distribution parameters, where the distribution parameters include the mean and variance;

[0123] The high feature set and the low feature set refer to dividing the sample set into child nodes above the division threshold and child nodes below the division threshold according to the division feature and a preset division threshold;

[0124] Exemplarily, the input features include power generation, inverter temperature, and temperature; at the root node, power generation is selected as the division feature, and the sample set is divided into two child nodes of high power generation and low power generation. For the high power generation child node, the inverter temperature is further selected as the division feature, and the sample set is divided into two child nodes of high inverter temperature and low inverter temperature; for the low power generation child node, the temperature is further selected as the division feature, and the sample set is divided into two child nodes of high temperature and low temperature; the recursive division process continues until the stop condition is met, and a scenario tree containing multiple leaf nodes can be obtained. Each leaf node is a sub-region corresponding to a PV power generation scenario; the stop condition can be reaching the maximum tree depth or the number of node samples being less than the threshold.

[0125] Exemplarily, the scenario tree contains four leaf nodes. Scenario 1 is: high power generation, high inverter temperature. The number of training samples falling into this scenario is 100, and the total number of training samples is 500. Then the occurrence probability of this scenario is 0.2. Assuming that the training samples falling into this leaf node have a power generation power in the range of 80 - 120, estimate the Gaussian distribution parameters of the power generation power in this scenario, where the Gaussian distribution parameters include the mean and variance.

[0126] Step S130: Define a deviation metric index and set a deviation threshold. When the deviation metric index is greater than this deviation threshold, it is considered that a significant deviation has occurred, and update the parameters of the PV power generation scenario tree model based on the current PV power generation data and meteorological data;

[0127] The calculation formula of the deviation metric index is: ;

[0128] where, is the actual power generation power, is the power generation power corresponding to the predicted PV power generation scenario, n is the number of training samples, is the abnormal power generation status indicator variable corresponding to the nith training sample at a specific time;

[0129] Introducing the abnormal power generation status indicator variable into the deviation metric index appropriately reduces the weight of the deviation in the abnormal state, improving the robustness of the deviation metric.

[0130] Preferably, the deviation threshold is equal to 20%; when the actual deviation metric index exceeds this threshold, it is considered that a significant deviation has occurred, and an update of the scenario tree needs to be triggered.

[0131] The method for updating the parameters of the photovoltaic power generation scenario tree model is as follows: calculate the adaptive learning rate according to the gradient of the model parameters at each update;

[0132] Introduce a forgetting factor , when a significant deviation occurs, calculate the value of the model parameters at the current update based on the adaptive learning rate, the gradient of the model parameters, and the forgetting factor from the previous update;

[0133] The calculation formula for the adaptive learning rate is: ;

[0134] where is the learning rate of the ith update for the th model parameter, is the initial learning rate, is the th update for the th model parameter gradient, is the smoothing term used to prevent the denominator from being zero;

[0135] The calculation formula for the value of the model parameters at the current update is: ; where is the value of the ith update for the th model parameter, is the weight of the training sample at the th update;

[0136] The training sample weight is used to weight the training samples in terms of time, making the new training samples have a higher weight.

[0137] Step S140: Construct a load prediction model based on the load electricity consumption data to predict the future load demand;

[0138] The construction and training method of the load prediction model is as follows:

[0139] Step S141: Extract load characteristics based on historical time load power consumption data, construct a load characteristic clustering model, divide the load characteristics into different load patterns, and each clustering category in the load curve clustering model represents a load pattern;

[0140] The load characteristic clustering model uses the k-means clustering algorithm to cluster the load characteristics; the load characteristic clustering model divides the load characteristics into K different clustering categories, and each category corresponds to a load pattern.

[0141] Step S142: Train a corresponding load prediction sub-model for each load pattern;

[0142] The load prediction sub-model uses the load power consumption data of historical time as training data, forms a time series in chronological order, uses the LSTM model as the initial model, and uses the sliding window method to construct training samples on the time series. In each training sample, the load power consumption data of the current window is used as input data, and the load power consumption data on the future time series of the current window is predicted as the load demand.

[0143] Step S143: Combine each load prediction sub-model into a load prediction model for predicting future load demands ;

[0144] Load prediction by pattern improves the load prediction accuracy on the basis of segmenting users and grasping differences. High-quality load prediction provides reliable demand-side information for optimal scheduling, helps make scheduling decisions that match the actual load, and reduces the grid operation cost while ensuring power consumption demands.

[0145] The above-mentioned photovoltaic power generation scenario tree model in the steps realizes the modeling and prediction of the uncertainty of photovoltaic power generation through a series of methods such as data preprocessing, feature extraction, and machine learning algorithms. The data cleaning step based on the power station alarm status, abnormal power consumption detection, and electrical maintenance records can effectively eliminate the power generation data and load power consumption data during abnormal and maintenance periods, improve the data quality and prediction accuracy of the scenario tree model and the load prediction model. At the same time, a power station alarm status indicator variable is introduced into the deviation measurement index, which improves the robustness of the deviation measurement. Through online learning and adaptive update, the model can dynamically adapt to the changes in the actual power generation status and improve the accuracy and reliability of the prediction.

[0146] It should be noted that in practical applications, it is also necessary to make appropriate adjustments and optimizations to each link in step S100 according to the specific characteristics of the photovoltaic power station and data quality, such as feature selection, model tuning, threshold setting, etc., to improve the performance and practicality of the photovoltaic power generation scenario tree model. In addition, it is also necessary to retrain and evaluate the model regularly to adapt to long-term data distribution changes and system evolution.

[0147] Step S200: Construct the first optimization objective and the second optimization objective by considering the operation revenue and the energy storage aging cost, consider the PV uncertainty based on the PV power generation scenario tree model, and establish an adaptive optimal scheduling model in combination with the load demand predicted by the load forecasting model;

[0148] The specific method of constructing the first optimization objective and the second optimization objective by considering the operation revenue and the energy storage aging cost, considering the PV uncertainty based on the PV power generation scenario tree model, and establishing an adaptive optimal scheduling model in combination with the load demand predicted by the load forecasting model is as follows:

[0149] Step S210: Input the energy storage device parameters, the energy storage aging model parameters, the electricity price information, and the load demand;

[0150] The above data are various parameters and data required for constructing the optimal scheduling model, providing basic information for subsequent modeling and solving.

[0151] Step S220: Establish a dynamic model of the energy storage device between the state variable SOC and the output variables of charge and discharge power based on the energy storage device parameters;

[0152] The energy storage device parameters include the maximum energy storage capacity, the upper and lower limits of charge and discharge power, the charge and discharge efficiency, and the self-discharge rate at each moment;

[0153] The calculation formula of the dynamic model of the energy storage device is: ; where represents the state of charge at time t, and are the charging and discharging powers respectively, and are the charge and discharge efficiencies, is the maximum energy storage capacity at time t, is the self-discharge rate, is the time interval between two moments.

[0154] The dynamic model of the energy storage device describes the dynamic changes of the energy storage SOC during the charging, discharging, and self-discharging processes, and is an important part of the energy storage constraints in the optimal scheduling.

[0155] Step S230: Construct a decay model of the maximum energy storage capacity and an internal resistance increase model based on the energy storage aging model parameters;

[0156] The energy storage aging model parameters include the decay model parameters, the power-law equation parameters, the unit capacity decay cost, and the unit internal resistance increase cost;

[0157] The calculation formula of the decay model is: ; where is the initial energy storage capacity, is the attenuation model parameter;

[0158] The attenuation model parameter is related to the number of charge-discharge cycles and depth, and is set by those skilled in the art according to the number of charge-discharge cycles and depth.

[0159] The attenuation model assumes that the energy storage capacity consists of two exponential decay terms, which respectively reflect the short-term and long-term characteristics of energy storage aging. By fitting historical data, the attenuation model parameters can be obtained for predicting future changes in energy storage capacity.

[0160] The calculation formula of the internal resistance increase model is: ;

[0161] Wherein, is the internal resistance value at time t, is the initial internal resistance, is the cumulative number of charge-discharge cycles, is the rated cycle life, and are the power-law equation parameters.

[0162] The internal resistance increase model assumes a power-law relationship between the internal resistance and the number of cycles, and the power-law equation parameters can be identified through experimental data.

[0163] Step S240: Define the operation revenue and energy storage aging cost based on the energy storage device parameters, energy storage aging model parameters, electricity price information, and load demand, and construct the first optimization objective and the second optimization objective;

[0164] The construction method of the first optimization objective and the second optimization objective is:

[0165] Step S241: Based on the electricity price information, which includes the electricity selling price, electricity purchasing price, energy storage charging and discharging prices, take maximizing the operation revenue as the first optimization objective ;

[0166] The calculation formula of the first optimization objective is: ;

[0167] The calculation formula of the operation revenue is: ;

[0168] Wherein, is the electricity selling price, is the electricity selling power, represents taking the positive part, is the electricity purchasing price, and are the energy storage charging and discharging prices respectively.

[0169] The first optimization objective indicates that the electricity sales revenue is the electricity sales quantity multiplied by the electricity sales price. The electricity sales quantity is equal to the electricity sales power minus the curtailed power, and the curtailed power is the part where the electricity sales power exceeds the load demand. The electricity consumption cost is the electricity purchase quantity multiplied by the power supply price, and the electricity purchase quantity is the part where the load demand exceeds the electricity sales power.

[0170] Step S242: Based on the internal resistance increase model, attenuation model, energy storage aging model parameters, and load demand, minimize the energy storage aging cost as the second optimization objective ;

[0171] The calculation formula of the second optimization objective is: ;

[0172] The calculation formula of the energy storage aging cost is: ;

[0173] Among them, is the unit capacity attenuation cost, is the unit internal resistance increase cost, is the unit load scheduling cost, represents the absolute value of the deviation between the predicted value and the actual value of the load demand at time t.

[0174] The second optimization objective quantifies the energy storage aging into economic costs, where the capacity attenuation cost reflects the irrecoverable loss of the energy storage capacity, and the internal resistance increase cost reflects the decline in the energy storage efficiency.

[0175] Step S250: Based on the photovoltaic power generation scenario tree model, construct the uncertainty constraint of the photovoltaic power generation power;

[0176] The uncertainty constraint adopts an ellipsoidal set, and describes the distribution characteristics of uncertain parameters through ellipsoidal inequality constraints;

[0177] Specifically, the method for constructing the uncertainty constraint of the photovoltaic power generation power based on the photovoltaic power generation scenario tree includes:

[0178] Step S251: For each photovoltaic power generation scenario and time period in the photovoltaic power generation scenario tree model, obtain the corresponding power generation power distribution parameters;

[0179] Step S252: Describe the uncertainty constraint of the photovoltaic power generation power based on the power generation power distribution parameters;

[0180] Specifically, the description of the uncertainty constraint of the photovoltaic power generation power based on the power generation power distribution parameters is specifically: for the th time period, define the uncertain parameter vector of the photovoltaic power generation power; describe the uncertainty constraint of the uncertain parameter vector based on the ellipsoidal set;

[0181] The uncertain parameter vector can be expressed as: , where is the number of photovoltaic power generation scenarios, is the -th photovoltaic power generation scenario's power generation power at time;

[0182] The expression of the uncertainty constraint is: ;

[0183] where is the mean value in the power generation power distribution parameters at time, is the covariance matrix of the photovoltaic power generation power at time, and its -th element is , is the scale parameter of the ellipsoidal set, which controls the size of the uncertainty set.

[0184] The covariance matrix can be estimated according to the correlation and transition probability between photovoltaic power generation scenarios, and its calculation formula is: ; where is the occurrence probability of the photovoltaic power generation scenario s, is the photovoltaic power generation power vector of the photovoltaic power generation scenario s at time.

[0185] The selection of the scale parameter needs to balance robustness and conservatism. A larger means considering a larger range of uncertainties, and the optimization result is more robust but also more conservative; a smaller means considering a smaller range of uncertainties, and the optimization result is more aggressive but also more vulnerable. A commonly used selection method is based on confidence, that is, select such that the ellipsoidal set covers a certain confidence range of uncertainties, for example, the confidence is selected as 95%.

[0186] It should be noted that the above uncertainty constraint is a second-order cone constraint and can be directly embedded into the optimization model.

[0187] Optionally, the ellipsoidal set can also be combined with other forms of uncertain sets to better characterize the distribution characteristics of uncertainties.

[0188] In summary, by transforming the photovoltaic power generation scenario tree model into an ellipsoidal uncertainty set, the uncertainty of photovoltaic power generation can be directly characterized and managed in the optimization model, and the balance between robustness and conservatism can be achieved by adjusting the set parameters. This method fully utilizes the statistical information provided by the scenario tree model and is also closely combined with the robust optimization framework, reflecting the idea of multi-source data fusion optimization.

[0189] Step S260: Introduce a risk measurement index to quantify the risk loss function of the uncertainty constraint of photovoltaic power generation ;

[0190] The calculation formula of the risk measurement index is: , where is the confidence level, is the risk loss function, is the auxiliary variable, represents the maximum value of and 0, represents the minimum z value that satisfies the condition, is the expectation operator, <00005..> represents the risk measurement index;

[0191] The confidence level is a real number between 0 and 1, representing the risk confidence level of concern; preferably, is equal to 0.9; the larger the confidence level, the more concerned about the risk in extreme cases, and the larger the value of CVaR.

[0192] The auxiliary variable is a real variable, which plays an auxiliary role in the definition of CVaR, is an approximation of VaR, and is used to represent CVaR as a convex optimization problem. During the optimization solution process, z will be automatically adjusted to minimize CVaR.

[0193] The design purpose of the risk measurement index is to find a minimum z value among all possible z values to minimize the expected value.

[0194] The calculation formula of the risk loss function is: ; where represents the risk loss function under the photovoltaic power generation scenario s, is the photovoltaic curtailment cost coefficient;

[0195] Step S270: Construct constraint conditions based on the dynamic model, attenuation model, internal resistance increase model of the energy storage device, load demand, and risk loss function, and construct an adaptive optimal scheduling model based on the first optimization goal and the second optimization goal;

[0196] The specific adaptive optimization scheduling model includes: using the dynamic model of energy storage equipment as the dynamic constraint of energy storage, the attenuation model and the internal resistance increase model as the aging cost constraint of energy storage, the linear constraint of calculating the risk measurement index based on the risk loss function, using the risk loss function as the risk constraint, using the load demand as the load balance constraint, and constructing an adaptive optimization scheduling model based on the first optimization target, the second optimization target and the risk measurement index;

[0197] The constraint expression for using the load demand as the load balance constraint is: ; where represents the load demand of the m-th load unit, and M is the total number of load units;

[0198] This constraint means that at any time t, the sum of the power generation power on the power supply side and the energy storage discharge power should be equal to the sum of the power consumption power on the load side and the energy storage charging power, that is, the power supply and the load are balanced.

[0199] The calculation formula of the adaptive optimization scheduling model is: ; where are the minimum and maximum values of the first optimization target and the second optimization target, is the weighting coefficient, and .

[0200] The adaptive optimization scheduling model first normalizes the two objective functions to make their dimensions consistent, and then determines the relative importance of the two objectives through the weighting coefficient to obtain the comprehensive objective function. By adjusting the value of the weighting coefficient, the operating income and the energy storage life can be flexibly balanced.

[0201] The calculation formula of the linear constraint for calculating the risk measurement index based on the risk loss function is: , where is the part where the risk loss function exceeds the auxiliary variable z in the photovoltaic power generation scenario s.

[0202] Robust optimization based on risk measurement enables the scheduling decision to pursue the maximum benefit while also controlling the risk in extreme cases within an acceptable range. By quantitatively weighing the benefit and the risk, the generated scheduling strategy is more robust and reliable, and can ensure the economic and safe operation of the system even in adverse situations. This greatly improves the ability of the power grid to cope with uncertainties.

[0203] Step S300: Embed the photovoltaic power generation scenario tree model and the load forecasting model into the adaptive optimization scheduling model, and solve the optimal energy storage charge and discharge strategy for the current period;

[0204] The specific method of embedding the photovoltaic power generation scenario tree model and the load forecasting model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current period is as follows:

[0205] Step S310: Embed the photovoltaic power generation scenario tree model into the adaptive optimal scheduling model, and generate an optimization sub-problem for the predicted value and uncertainty constraint of the generated power corresponding to each photovoltaic power generation scenario;

[0206] Step S320: Embed the load forecasting model into the adaptive optimal scheduling model, and introduce the predicted load demand as a known parameter into the optimization sub-problem;

[0207] Step S330: Perform weighted summation on each optimization sub-problem to obtain the expected objective function of the overall optimization problem;

[0208] The calculation formula for the expected objective function of the overall optimization problem is: ; where represents the occurrence probability of the photovoltaic power generation scenario s, , , represent the first optimization objective, the second optimization objective, and the risk loss function under the photovoltaic power generation scenario s;

[0209] Step S340: Solve the expected objective function of the overall optimization problem to generate the optimal energy storage charge and discharge strategy for the current time period, and the optimal energy storage charge and discharge strategy includes the optimal charge and discharge power;

[0210] Step S350: Adopt a rolling optimization mechanism to regularly roll the optimization time domain forward; repeat steps S310 - S340 on the new optimization time domain, and update the optimal energy storage charge and discharge strategy according to the latest predicted value of the photovoltaic power generation;

[0211] Exemplarily, it rolls once every 1 hour, and the time domain length of each roll is 24 hours.

[0212] Step S360: Output the optimal energy storage charge and discharge strategy for each optimization time domain to form a complete energy storage optimal scheduling scheme.

[0213] The output of the above steps is the optimal energy storage charge and discharge strategy for each scheduling period, which is the core result of the energy storage optimal scheduling. Through rolling optimization and adaptive adjustment, this strategy can dynamically adapt to the uncertainty changes of photovoltaic power generation, improve the robustness and reliability of energy storage scheduling while ensuring the economy of the system.

[0214] This application innovatively comprehensively considers energy storage aging, photovoltaic uncertainty, and risk management, constructs an adaptive data-driven optimal scheduling framework, effectively solves the problem that traditional methods are difficult to balance economy, reliability, and adaptability, extends the energy storage life while improving the operating revenue of the photovoltaic energy storage system, has significant technical and economic benefits, provides new ideas and methods for solving the uncertainty problem in new energy optimal scheduling, and is of great significance for promoting the development of the energy Internet and smart grid.

[0215] Embodiment 2

[0216] To verify the effectiveness and creativity of the photovoltaic energy storage load collaborative optimal scheduling system based on multi-source data fusion of this application, an embodiment of this application designs a comprehensive test scenario, and through simulation experiments, compares and analyzes the performance differences between this application and the prior art.

[0217] First of all, a typical industrial park microgrid is selected as the test object. This microgrid consists of a 2MW photovoltaic power station, a 1MW / 4MWh energy storage system, a load center, etc. The experimental data includes photovoltaic power output, load electricity consumption, electricity price information, etc. for one continuous year, and the time granularity is 15 minutes. Among them, the photovoltaic power output data contains 5000 measured data and 5000 simulated data. The simulated data is generated by adding random perturbations to cover more power output fluctuation scenarios; the load electricity consumption data includes 8000 regular loads and 2000 dispatchable loads, and the load fluctuations are set according to factors such as season and weather; the electricity price information refers to the local actual peak-valley electricity price and interruptible load electricity price, and random fluctuations are added to simulate the changes in the electricity market.

[0218] In the test, two optimal scheduling schemes are adopted: Scheme 1 is the photovoltaic energy storage load collaborative optimal scheduling system based on multi-source data fusion of this application, which adopts the technical route of scenario tree + CVaR risk measure + energy storage aging model + rolling optimization; Scheme 2 is a traditional scheduling strategy based on robust optimization, which uses confidence intervals to describe photovoltaic uncertainty and aims to minimize the operating cost. Both schemes are run on a computer equipped with a 2.5GHz Intel Core i7 processor and 16GB of memory.

[0219] The test first preprocesses and extracts features from the photovoltaic power output and load electricity consumption data, and divides the data into a training set and a test set, and the ratio of the training set to the test set is 80%:20%. Subsequently, the model parameters of the two schemes are trained and optimized using the training set data, and optimal scheduling simulations are carried out on the test set to generate a one-year scheduling strategy. Finally, the test compares and analyzes the performance of the two schemes on key performance indicators, as shown in Table 1 for details.

[0220] Table 1 Optimal scheduling simulation results table

[0221]

[0222] The experimental results show that this application is significantly superior to the prior art in all key indicators. In terms of operating costs, the total cost of this application is 3.282 million yuan, which is 18.3% lower than the 4.015 million yuan of the prior art. This is mainly due to the adaptive optimization and CVaR risk control, which avoid the high costs in extreme situations while ensuring economy. In terms of photovoltaic utilization rate, this application reaches 92.5%, which is 9.7% higher than the prior art. The reason is that the scenario tree model can finely depict the uncertainty of photovoltaic output and generate targeted optimization strategies to maximize the utilization of photovoltaic energy. In terms of load satisfaction rate, this application reaches 98.2%, which is 2.7% higher than the prior art. This is attributed to demand response and energy storage optimal scheduling, ensuring continuous and reliable power supply for the load.

[0223] In terms of the energy storage cycle life, the expected life of this application is 8.2 years, which is 28.1% longer than the 6.4 years of the prior art. The main reason is the energy storage aging model and multi-objective optimization, which reasonably control the charge and discharge of the energy storage in the scheduling strategy and delay the attenuation of the energy storage capacity and efficiency. Finally, in terms of computing efficiency, the computing time of this application is 15.6 minutes, which is 44.7% shorter than the 28.2 minutes of the prior art. Thanks to the adaptive rolling optimization framework, the original problem is decomposed into multiple sub-problems and solved in parallel, greatly improving the computing speed.

[0224] In summary, this application has achieved a comprehensive improvement in multiple aspects such as operating economy, energy utilization rate, power supply reliability, energy storage life, and computing efficiency, innovatively solving the problem that it is difficult for the prior art to balance economic benefits and risk management. The integration of the photovoltaic power generation scenario tree and the CVaR risk metric constructs an optimization framework for accurately quantifying uncertainty; the introduction of the energy storage aging model and rolling optimization realizes coordinated scheduling on multiple time scales; the adaptive optimization mechanism and parallel computing architecture ensure the real-time processing of massive heterogeneous data. The systematic integration of these innovative technical routes provides an efficient, reliable, and economic solution for the optimal scheduling of new energy, making an important contribution to promoting the technological progress of the energy Internet and smart grid.

[0225] Example 3

[0226] As Figure 6 shown, the optical storage load collaborative optimization scheduling system based on multi-source data fusion provided by this application includes:

[0227] A scenario tree model construction module, configured to obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and construct a photovoltaic power generation scenario tree model and a load prediction model;

[0228] An adaptive model construction module, configured to construct a first optimization objective and a second optimization objective by considering operating revenue and energy storage aging cost, consider the uncertainty of photovoltaic power based on a photovoltaic power generation scenario tree model, and combine the load demand predicted by a load forecasting model to establish an adaptive optimal scheduling model;

[0229] A charge and discharge strategy output module, configured to embed a photovoltaic power generation scenario tree model and a load forecasting model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current period.

[0230] Embodiment 4

[0231] According to an embodiment of the present application, a readable storage medium is also provided. A computer-readable instruction is stored on the computer-readable storage medium. When the computer-readable instruction is run by a processor, it can execute the method of collaborative optimal scheduling of photovoltaic energy storage and load based on multi-source data fusion according to the embodiment of the present application described with reference to the above drawings. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0232] In addition, according to the embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present application, such as: obtaining historical photovoltaic power generation data, load power consumption data, and meteorological data, and constructing a photovoltaic power generation scenario tree model and a load forecasting model; constructing a first optimization objective and a second optimization objective by considering operating revenue and energy storage aging cost, considering the uncertainty of photovoltaic power based on the photovoltaic power generation scenario tree model, and combining the load demand predicted by the load forecasting model to establish an adaptive optimal scheduling model; embedding the photovoltaic power generation scenario tree model and the load forecasting model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current period. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0233] The methods, apparatuses, and devices of the present application can be implemented in many ways. For example, the methods, apparatuses, and devices of the present application can be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only. The steps of the method of the present application are not limited to the specific order described above, unless otherwise specifically stated. In addition, in some embodiments, the present application can also be implemented as a program recorded in a recording medium, and these programs include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers a recording medium storing a program for executing the method according to the present application.

[0234] In addition, parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0235] As described above, the specific embodiments further elaborate on the objectives, technical solutions, and beneficial effects of the present application. It should be understood that the above description is only the specific embodiments of the present application and is not used to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included within the protection scope of the present application.

Claims

1. A method for collaborative optimal scheduling of optical storage and charge based on multi-source data fusion, characterized in that, Including: Obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and construct a photovoltaic power generation scenario tree model and a load forecasting model; Consider the operating revenue and energy storage aging cost to construct the first optimization objective and the second optimization objective. Based on the photovoltaic power generation scenario tree model, consider the photovoltaic uncertainty, and combine with the load demand predicted by the load forecasting model to establish an adaptive optimal scheduling model; Embed the photovoltaic power generation scenario tree model and the load forecasting model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current period; Define a deviation metric index and set a deviation threshold. When the deviation metric index MA is greater than the deviation threshold, it is considered that a significant deviation has occurred, and update the parameters of the photovoltaic power generation scenario tree model based on the current photovoltaic power generation data and meteorological data; The specific method for constructing the photovoltaic power generation scenario tree model includes: Select the long short-term memory network as the initial network, construct training samples based on the sliding window method, preset the time window size, and extract input features based on historical photovoltaic power generation data and meteorological data within each time window; Preset the prediction step length T. For each time window, extract the power generation power within the prediction step length of the future time as the output label, and combine the input features and output labels extracted from each time window into a training sample; Slide the time window to obtain a training sample set, and train the photovoltaic power generation scenario tree model based on the training sample set. Use the recursive partitioning method to divide the input features into N sub-regions, and each sub-region corresponds to a photovoltaic power generation scenario; The training process of the photovoltaic power generation scenario tree model includes: Starting from the root node, recursively partition the training sample set of the current node, select a partitioning feature from the input features, and divide the training sample set into a high feature set and a low feature set to obtain a photovoltaic power generation scenario tree containing N sub-regions; For each sub-region, count the number of training samples falling into this node, and calculate the proportion of it in the total number of training samples as the occurrence probability of this photovoltaic power generation scenario; For each sub-region, count the power generation power distribution parameters under each photovoltaic power generation scenario; Use the training sample set to optimize the parameters of the photovoltaic power generation scenario tree model by minimizing the loss function; When the value of the loss function converges, the training is completed; Use the trained photovoltaic power generation scenario tree model to predict the occurrence probability and its power generation power distribution parameters of the photovoltaic power generation scenario to which the new input features belong, and the distribution parameters include the mean and variance.

2. The collaborative optimization scheduling method for optical storage and charge based on multi-source data fusion according to claim 1, wherein The specific method for obtaining historical photovoltaic power generation data, load power consumption data, and meteorological data, and constructing a photovoltaic power generation scenario tree model and a load forecasting model also includes: Obtain historical photovoltaic power generation data and meteorological data, and perform preprocessing; The historical photovoltaic power generation data includes power generation power, inverter temperature, and power generation amount data; Construct a load forecasting model based on the load power consumption data to predict the future load demand.

3. The collaborative optimization scheduling method of optical storage and charge based on multi-source data fusion according to claim 2, wherein The construction and training method of the load forecasting model is: [[ID=​12]]Step S141: Extract load features based on the load power consumption data of historical time, construct a load feature clustering model, divide the load features into different load patterns, and each clustering category in the load curve clustering model represents a load pattern; Step S142: Train the corresponding load prediction sub-model for each load pattern; Step S143: Combine each load forecasting sub-model into a load forecasting model for predicting future load demand ; The load prediction sub-model uses the load power consumption data of historical time as training data, forms a time series in chronological order, uses the LSTM model as the initial model, and uses the sliding window method to construct training samples on the time series. In each training sample, the load power consumption data of the current window is used as input data, and the load power consumption data on the future time series of the current window is predicted as the load demand.

4. The method for collaborative optimal scheduling of optical storage and charge based on multi-source data fusion according to claim 3, wherein The specific method for constructing the first optimization objective and the second optimization objective considering the operating income and the energy storage aging cost, considering the photovoltaic uncertainty based on the photovoltaic power generation scenario tree model, and combining the load demand predicted by the load prediction model to establish the adaptive optimal scheduling model is as follows: Input the energy storage device parameters, energy storage aging model parameters, electricity price information, and load demand; Establish a dynamic model of the energy storage device between the state variable SOC and the output variable charge and discharge power based on the parameters of the energy storage device ; Construct a decay model of the maximum energy storage capacity and an internal resistance increase model based on the energy storage aging model parameters and an internal resistance increase model ; Define the operating income and the energy storage aging cost based on the energy storage device parameters, energy storage aging model parameters, electricity price information, and load demand, and construct the first optimization objective and the second optimization objective; Construct the uncertainty constraint of the photovoltaic power generation power based on the photovoltaic power generation scenario tree model; Introduce risk measurement indicators and quantify the risk loss function of the uncertainty constraint of photovoltaic power generation ; Construct the constraint conditions based on the energy storage device dynamic model, attenuation model, internal resistance increase model, load demand, and risk loss function, and construct the adaptive optimal scheduling model based on the first optimization objective and the second optimization objective; The adaptive optimal scheduling model specifically includes: using the energy storage device dynamic model as the energy storage dynamic constraint, the attenuation model and the internal resistance increase model as the energy storage aging cost constraint, the linear constraint of calculating the risk metric index based on the risk loss function, using the risk loss function as the risk constraint, using the load demand as the load balance constraint, and constructing the adaptive optimal scheduling model based on the first optimization objective, the second optimization objective, and the risk metric index.

5. The method for collaborative optimal scheduling of optical storage and charge based on multi-source data fusion according to claim 4, wherein The construction method of the first optimization objective and the second optimization objective is as follows: Based on electricity price information, the electricity price information includes the selling electricity price, the purchasing electricity price, and the energy storage charging and discharging price, with maximizing the operating revenue as the first optimization objective ; Based on the internal resistance increase model, attenuation model, energy storage aging model parameters and load demand, minimizing the energy storage aging cost is taken as the second optimization goal .

6. The method for collaborative optimal scheduling of optical storage and charge based on multi-source data fusion according to claim 5, wherein The method for constructing the uncertainty constraint of the photovoltaic power generation power based on the photovoltaic power generation scenario tree includes: For each photovoltaic power generation scenario and time period in the photovoltaic power generation scenario tree model, obtain the corresponding power generation power distribution parameters; Describing the uncertainty constraint of photovoltaic power generation based on the distribution parameters of power generation, the specific method is as follows: For the th time period, define the uncertain parameter vector of photovoltaic power generation; describe the uncertainty constraint of the uncertain parameter vector based on the ellipsoidal set.

7. The method for collaborative optimization scheduling of optical storage and charge based on multi-source data fusion according to claim 6, characterized in that The specific method for embedding the photovoltaic power generation scenario tree model and the load prediction model into the adaptive optimal scheduling model to solve the optimal energy storage charge and discharge strategy for the current time period is as follows: Step S310: Embed the photovoltaic power generation scenario tree model into the adaptive optimal scheduling model, and generate an optimization sub-problem for the predicted value and uncertainty constraint of the power generation power corresponding to each photovoltaic power generation scenario; Step S320: Embed the load prediction model into the adaptive optimal scheduling model, and introduce the predicted load demand as a known parameter into the optimization sub-problem; Step S330: Perform weighted summation on each optimization sub-problem to obtain the expected objective function of the overall optimization problem; Step S340: Solve the expected objective function of the overall optimization problem to generate the optimal energy storage charge and discharge strategy for the current time period, and the optimal energy storage charge and discharge strategy includes the optimal charge and discharge power; Step S350: Adopt a rolling optimization mechanism to regularly roll the optimization time domain forward; repeat Steps S310 - S340 on the new optimization time domain, and update the optimal energy storage charge - discharge strategy according to the latest predicted value of photovoltaic power generation. Step S360: Output the optimal energy storage charge - discharge strategy for each optimization time domain to form a complete energy storage optimization scheduling plan.

8. The optical storage and load collaborative optimization scheduling system based on multi-source data fusion is implemented based on the optical storage and load collaborative optimization scheduling method based on multi-source data fusion described in any one of claims 1-7, and is characterized in that Including: A scenario tree model construction module, which is used to obtain historical photovoltaic power generation data, load power consumption data, and meteorological data, and construct a photovoltaic power generation scenario tree model and a load prediction model. An adaptive model construction module, which is used to consider the operation revenue and energy storage aging cost to construct the first optimization objective and the second optimization objective, and based on the photovoltaic power generation scenario tree model, consider the photovoltaic uncertainty, and combine with the load demand predicted by the load prediction model to establish an adaptive optimization scheduling model. A charge - discharge strategy output module, which is used to embed the photovoltaic power generation scenario tree model and the load prediction model into the adaptive optimization scheduling model to solve the optimal energy storage charge - discharge strategy for the current period.

Images