Distributed photovoltaic and energy storage load collaborative optimization operation method and system

Abstract

The invention provides a distributed photovoltaic and energy storage load collaborative optimization operation method and system. The method comprises the following steps: acquiring multi-source data and processing the multi-source data by adopting an improved wave sorting algorithm; constructing a CNN and LSTM hybrid model to predict photovoltaic output; applying a single-index time FPT algorithm and combining with a cactus shrinkage principle to predict load demand and electricity price fluctuation; establishing a collaborative optimization model taking economy and photovoltaic consumption rate as double targets; carrying out distributed iterative optimization by adopting a single-index time FPT algorithm based on a cactus shrinkage principle; and generating a multi-time-scale control instruction and issuing the multi-time-scale control instruction to each device to realize cooperative control. Through the innovative data processing method and optimization algorithm, the efficient cooperative operation of the distributed energy system is realized, the photovoltaic consumption rate and the economic benefit are improved, and the practical value is relatively high.

Description

Technical Field

[0001] This invention relates to the fields of renewable energy and smart grid technology, and in particular to a method and system for the coordinated and optimized operation of distributed photovoltaic and energy storage loads. Background Technology

[0002] With increasing environmental awareness and the development of renewable energy technologies, distributed photovoltaic (PV) power generation systems have been widely adopted globally. Distributed PV systems are typically installed at the user end, such as on building rooftops or walls, enabling local power generation and consumption, reducing transmission losses, and improving energy efficiency.

[0003] Currently, there are two main application modes for distributed photovoltaic (PV) systems: grid-connected systems prioritize supplying the generated electricity to the users themselves, with surplus electricity fed back to the grid; off-grid systems are typically equipped with energy storage devices to store the generated electricity for users to use when needed. In grid-connected applications, a "self-consumption with surplus electricity fed back to the grid" model is often adopted, allowing users to reduce their electricity purchase expenses from the grid by generating a portion of their own power, thus achieving economic benefits.

[0004] In recent years, systems integrating photovoltaic power generation, energy storage devices, and intelligent load management have become a research hotspot. These systems typically employ a predictive-optimization-control framework: first, they predict photovoltaic output and load demand; then, they construct optimization models to determine the operating strategies of each device; and finally, the controller executes optimization commands. However, existing technologies have limitations in multi-entity collaboration, especially in cases of high-proportion distributed photovoltaic grid integration. A single optimization objective struggles to balance the contradiction between system economics and photovoltaic grid integration efficiency, and the optimization algorithms suffer from low execution efficiency, making it difficult to meet the demands of real-time control.

[0005] Furthermore, existing collaborative optimization methods often neglect distribution network safety constraints, potentially leading to safety issues such as voltage exceeding limits and line overloads. Simultaneously, the lack of an effective hierarchical or distributed computing framework makes it difficult to handle collaborative optimization problems involving large-scale distributed energy resources, limiting the system's effectiveness in complex distribution networks. Summary of the Invention

[0006] The purpose of this invention is to solve the problems existing in the prior art and provide a method and system for the coordinated and optimized operation of distributed photovoltaic and energy storage loads. Through innovative data processing methods and optimization algorithms, this invention achieves efficient coordinated operation of distributed energy systems, thereby improving photovoltaic absorption rate and economic benefits.

[0007] To achieve the above objectives, the present invention provides a method for coordinated and optimized operation of distributed photovoltaic and energy storage loads, comprising the following steps: Meteorological data, photovoltaic equipment parameters, and historical power generation data are acquired. An improved wave sorting algorithm is used to perform wave analysis and dynamic segmentation on the multi-source heterogeneous data, extracting time-series features, periodic features, and environmentally relevant features to form a feature vector set. Based on the feature vector set, a deep learning hybrid model combining CNN and LSTM is constructed to predict the output of distributed photovoltaic power and obtain the photovoltaic power prediction result. Collect users' historical electricity consumption data and electricity price information, and apply the improved single exponential time FPT algorithm combined with the cactus contraction principle to perform fast Fourier transform on the load time series data to obtain load demand forecast results and electricity price fluctuation forecast results. Based on the photovoltaic power prediction results, load demand prediction results, and electricity price fluctuation prediction results, a multi-agent collaborative optimization model is established with the joint optimization objectives of minimizing the comprehensive energy cost of users and maximizing the local photovoltaic absorption rate, resulting in a dual-objective optimization function. The bi-objective optimization function is optimized by a distributed iterative algorithm based on the cactus contraction principle using a single exponential time FPT algorithm. Problem decomposition is achieved through core parameter identification and decision space compression, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources. Based on the aforementioned collaborative optimization strategy, multi-timescale control commands are generated, including day-ahead planning, hourly scheduling, and minute-level real-time control. These control commands are then transmitted in real-time to photovoltaic inverters, energy storage devices, and smart loads via a communication network, enabling dynamic matching and collaborative control of source, storage, and load resources.

[0008] Preferably, an improved wave sorting algorithm is used to perform wave analysis and dynamic segmentation on the multi-source heterogeneous data, extracting time-series features, periodic features, and environmentally relevant features to form a feature vector set, including: Anomaly detection is performed on the meteorological data, photovoltaic equipment parameters, and historical power generation data. Outliers are identified using the Z-score method, and missing data is filled in using cubic spline interpolation to obtain a preprocessed dataset. Based on the preprocessed dataset, the time series data is divided into several segments, each segment being a wave unit. The amplitude, period, and energy characteristic parameters of each wave unit are calculated to obtain a set of wave characteristic parameters. Based on the wave feature parameter set, an improved wave sorting algorithm is used to sort and classify the data, extract time-series features, periodic features, and environmentally relevant features to form a feature vector set.

[0009] Preferably, a deep learning hybrid model combining CNN and LSTM is constructed to predict distributed photovoltaic power output, obtaining photovoltaic power prediction results, including: Based on the feature vector set, local feature patterns are extracted through a multi-layer CNN, convolutional kernels are used to perform convolution operations on the features, and dimensionality reduction is performed through pooling layers to obtain a spatial feature map. The spatial feature map is reshaped to form temporal input data, which is then input into the LSTM layer for long-term temporal dependency learning to obtain a temporal feature representation. The time-series features are processed through an attention mechanism, integrated through a fully connected layer, and the prediction results are output to obtain the photovoltaic power prediction results.

[0010] Preferably, an improved single-exponential-time FPT algorithm combined with the cactus contraction principle is applied to perform a fast Fourier transform on the load time-series data to obtain load demand forecasting results and electricity price fluctuation forecasting results, including: The user's historical electricity consumption data and electricity price information are subjected to a fast Fourier transform to convert the time-domain signal into a frequency-domain representation, and the main frequency component and secondary frequency component are separated to obtain the spectrum data. The spectral data is processed using the cactus contraction function. The main frequency features are preserved and the secondary frequency information is compressed by S(f) = f × [1 - exp(-α|f| / fmax)], resulting in the compressed spectral features. Based on the compressed spectral characteristics, combined with historical data and recent trends, a segmented single-exponential time complexity calculation method is used for prediction modeling to obtain load demand prediction results and electricity price fluctuation prediction results.

[0011] Preferably, a multi-agent collaborative optimization model is established with the joint optimization objectives of minimizing the user's comprehensive energy cost and maximizing the local photovoltaic absorption rate, resulting in a dual-objective optimization function, including: Based on the photovoltaic power forecast results, load demand forecast results, and electricity price fluctuation forecast results, a comprehensive energy cost objective function for users is constructed, including grid electricity purchase cost, equipment maintenance cost, and carbon emission cost, thus obtaining the cost objective function. Based on the ratio of actual photovoltaic power generation to theoretically available photovoltaic power generation, a photovoltaic grid connection rate objective function is constructed, and the grid connection rate objective function is obtained. Based on the aforementioned cost objective function and absorption rate objective function, a fuzzy logic controller is used to dynamically adjust the weight coefficients of the two objectives according to the system operating status and user preferences, thereby obtaining a dual-objective optimization function.

[0012] Preferably, a single exponential time FPT algorithm based on the cactus contraction principle is used for distributed iterative optimization. Problem decomposition is achieved through core parameter identification and decision space compression, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources, including: The bi-objective optimization function is represented as a parameterized problem, and the set of key parameters that have the greatest impact on the objective function is identified through sensitivity analysis, thus obtaining the core parameter set; The decision space is nonlinearly transformed using a cactus shrinkage function. The original resolution is maintained in the core parameter direction, while compression is performed in the non-core parameter direction to obtain the compressed decision space. Based on the compressed decision space, the global optimization problem is decomposed into photovoltaic control sub-problems, energy storage control sub-problems, and load control sub-problems according to physical properties, resulting in a combination of independently solvable sub-problems; The ADMM algorithm is used to solve the combination of subproblems in a distributed iterative manner. By alternately updating the original and dual variables, the coordinated optimization of each subproblem is achieved, and a local optimal solution set is obtained. Based on the aforementioned local optimal solution set, the global optimal solution is gradually approximated through information exchange between coordination variables and Lagrange multipliers, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources.

[0013] Preferably, multi-timescale control commands are generated, including day-ahead planning, hourly scheduling, and minute-level real-time control. These control commands are then transmitted in real-time to photovoltaic inverters, energy storage devices, and smart loads via a communication network to achieve dynamic matching and coordinated control of source, storage, and load resources. Based on the aforementioned collaborative optimization strategy, and using 24-hour photovoltaic output forecast, load demand forecast, and electricity price forecast data, an energy storage charging and discharging plan and a day-ahead optimization plan for adjustable load operating periods are generated, thereby obtaining day-ahead control commands. Based on real-time updated forecast data and system status, rolling optimization is performed every hour to update the equipment operation plan for the next 4 to 6 hours and obtain hourly scheduling instructions; Based on the actual system status and short-term prediction deviations, tracking control, correction control and emergency control strategies are adopted to generate minute-level real-time control commands for photovoltaic inverter power regulation, energy storage charging and discharging power and load response intensity, thus obtaining minute-level control commands; The daytime control commands, hourly scheduling commands, and minute-level control commands are converted into protocols using Modbus, SunSpec, or BACnet protocol conversion technologies to obtain standardized control commands. The standardized control commands are sent to photovoltaic inverters, energy storage devices and smart loads in real time through the communication network, so as to realize dynamic matching and coordinated control of source, storage and load resources.

[0014] Preferably, the amplitude, period, and energy characteristic parameters of each wave unit are calculated to obtain a set of wave characteristic parameters, including: Based on each sub-segment of the preprocessed dataset, the wave amplitude A is calculated. wave= max(x i ) - min(x i ), where x i For the data points within the sub-segment, the amplitude characteristics are obtained; For the amplitude characteristics, the dominant period T of each segment is determined by autocorrelation function analysis. wave Calculate the periodic characteristic parameters to obtain the periodic characteristics; Based on the amplitude and period characteristics, the energy characteristic E is calculated. wave = ∑(x i ²) / N, where N is the segment length, yields the energy characteristics; Based on the amplitude characteristics, periodic characteristics, and energy characteristics, a set of wave feature parameters containing time series, periodic, and energy information is constructed.

[0015] Preferably, a fuzzy logic controller is used to dynamically adjust the weight coefficients of the two objectives based on the system operating state and user preferences, resulting in a bi-objective optimization function, including: The ratio of the predicted photovoltaic output to the actual electricity demand, the current state of charge of the battery, the grid electricity price level, and the user's set preference coefficient are collected to construct the state vector for weight adjustment and obtain the input parameters for weight adjustment. Based on the weighted input parameters, a fuzzy logic rule base is established. When photovoltaic power generation is much greater than the load demand and the grid electricity price is low, the weight of photovoltaic absorption rate is increased. When photovoltaic power generation is insufficient and the electricity price is high, the weight of cost is increased, thus obtaining fuzzy logic rules. According to the fuzzy logic rules, the weight adjustment amount Δw1 is calculated, and the weight coefficients are updated by w1(t+1) = w1(t) + Δw1 and w2(t+1) = 1 - w1(t+1) to obtain the adaptive weight coefficients. Based on the adaptive weight coefficients, a bi-objective optimization function F = w1·C1 + w2·C2 is constructed, where C1 is the user's comprehensive energy cost and C2 is the negative value of the photovoltaic absorption rate, thus obtaining the bi-objective optimization function.

[0016] The present invention also provides a distributed photovoltaic and energy storage load coordinated optimization operation system, comprising: The data acquisition module is used to acquire meteorological data, photovoltaic equipment parameters and historical power generation data, as well as users' historical electricity consumption data and electricity price information, and transmit the data to the predictive analysis module. The predictive analysis module receives data from the data acquisition module, uses an improved wave sorting algorithm and a deep learning hybrid model to predict photovoltaic output, and uses a single exponential time FPT algorithm to predict load demand and electricity price, providing the prediction results to the optimization decision module. The optimization decision module receives the prediction results from the prediction analysis module, establishes a dual-objective collaborative optimization model, uses a distributed iterative optimization algorithm based on the cactus contraction principle to solve the collaborative optimization strategy, and outputs the optimization strategy to the control execution module. The control execution module receives the optimization strategy from the optimization decision module, generates multi-timescale control commands, and sends them to photovoltaic inverters, energy storage devices, and smart loads through the communication network to achieve coordinated control.

[0017] The beneficial effects of this invention are: By using an improved wave sorting algorithm to process multi-source heterogeneous data, the accuracy of photovoltaic power output prediction is improved, providing reliable data support for subsequent optimization decisions. The single exponential time FPT algorithm based on the cactus contraction principle is adopted, which significantly reduces the computational complexity of the optimization problem and makes the collaborative optimization of large-scale distributed energy systems possible. A multi-objective weight adaptive adjustment mechanism was designed, which can dynamically balance the two optimization objectives of economic efficiency and photovoltaic absorption rate according to the system operating status and user preferences, thereby improving the overall efficiency of the system. A multi-time-scale hierarchical control framework was constructed. Through the coordinated operation of daily planning, hourly scheduling and minute-level real-time control, a hierarchical response to uncertainty was achieved, thereby improving the robustness of system operation. A distributed iterative optimization strategy based on ADMM was implemented, which achieves collaborative optimization of large-scale distributed energy groups through local information exchange and coordination variable updates, thereby reducing the central computing burden. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. Those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the distributed photovoltaic and energy storage load coordinated optimization operation method of the present invention; Figure 2 This is a detailed flowchart of S1 of the present invention; Figure 3 This is a detailed flowchart of S2 of the present invention; Figure 4 This is a schematic diagram of the distributed photovoltaic and energy storage load collaborative optimization system architecture design of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0021] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0022] In this document, the term "and / or" merely describes a relationship, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0025] like Figure 1 As shown, the distributed photovoltaic and energy storage load coordinated optimization operation method provided by the present invention includes the following steps: Step S1: Obtain meteorological data, photovoltaic equipment parameters and historical power generation data, and use an improved wave sorting algorithm to perform wave analysis and dynamic segmentation on the multi-source heterogeneous data to extract time-series features, periodic features and environmentally related features to form a feature vector set; Step S2: Based on the feature vector set, construct a deep learning hybrid model combining CNN and LSTM to predict the distributed photovoltaic power output and obtain the photovoltaic power prediction result; Step S3: Collect users' historical electricity consumption data and electricity price information, and apply the improved single exponential time FPT algorithm combined with the cactus contraction principle to perform fast Fourier transform on the load time series data to obtain load demand forecast results and electricity price fluctuation forecast results. Step S4: Based on the photovoltaic power prediction results, load demand prediction results, and electricity price fluctuation prediction results, establish a multi-agent collaborative optimization model with the joint optimization objectives of minimizing the user's comprehensive energy cost and maximizing the local photovoltaic absorption rate, and obtain the dual-objective optimization function; Step S5: The bi-objective optimization function is optimized in a distributed iterative manner using the single exponential time FPT algorithm based on the cactus contraction principle. Problem decomposition is achieved through core parameter identification and decision space compression to obtain a collaborative optimization strategy for photovoltaic power generation units, energy storage units and adjustable load resources. Step S6: Based on the collaborative optimization strategy, generate multi-timescale control commands for day-ahead planning, hourly scheduling, and minute-level real-time control. Send the control commands to photovoltaic inverters, energy storage devices, and smart loads in real time through the communication network to achieve dynamic matching and collaborative control of source, storage, and load resources.

[0026] Specifically, in this embodiment of the invention, the core process of the method for coordinated optimization of distributed photovoltaic and energy storage loads includes six key steps, forming a complete prediction-optimization-control closed-loop system.

[0027] First, the system acquires meteorological data, photovoltaic equipment parameters, and historical power generation data. These data have different sampling frequencies, data formats, and physical meanings, representing typical heterogeneous data. To address this multi-source heterogeneous data, this invention employs an improved wave sorting algorithm. This algorithm treats time-series data as a wave sequence, dynamically segmenting the data into several wave units and calculating the amplitude, period, and energy characteristic parameters of each unit. Unlike traditional sorting, wave sorting allows for local disorder while maintaining global order, a characteristic that enables it to effectively handle data with random fluctuations, such as photovoltaic power generation data. In this way, the system extracts time-series features (such as trends, seasonality, and periodicity), periodic features (such as daily cycles and meteorological cycles), and environmentally relevant features (such as temperature-power relationships and irradiance-power relationships) from the original data, forming a high-dimensional feature vector set.

[0028] The second step involves constructing a hybrid deep learning model combining CNN and LSTM based on the feature vector set extracted in the previous step. This model fully leverages the advantages of different types of neural networks: CNN excels at extracting spatial features and local patterns, while LSTM specializes in capturing long-term temporal dependencies. The model architecture adopts a "CNN + LSTM + fully connected layer" structure. The input layer receives the feature vectors after wave sorting. First, multiple CNN layers extract local feature patterns. Then, the feature maps are reshaped into sequences and input into the LSTM layer to learn long-term dependencies. Finally, the features are integrated through an attention mechanism and a fully connected layer to output the prediction result. This hybrid model structure can simultaneously process the spatial and temporal features in photovoltaic power generation data, improving prediction accuracy.

[0029] The third step involves the system collecting historical electricity consumption data and electricity price information from users, and applying an improved single-exponential time FPT algorithm combined with the cactus contraction principle to predict load and electricity prices. This algorithm first performs a Fast Fourier Transform on the load time-series data, converting the time-domain signal into a frequency-domain representation and separating the dominant and secondary frequency components. Then, it processes the spectrum using the cactus contraction function S(f) = f × [1 - exp(-α|f| / fmax)]. This function preserves the main frequency characteristics while compressing secondary frequency information, significantly reducing computational complexity. During the prediction process, the algorithm employs a piecewise single-exponential time complexity calculation method, reducing its complexity to O(n·exp(c)) compared to traditional algorithms, where c is a constant independent of the problem size, enabling the algorithm to efficiently process large-scale load data.

[0030] The fourth step involves establishing a multi-agent collaborative optimization model based on the photovoltaic power forecast, load demand forecast, and electricity price fluctuation forecast obtained in the first three steps. The model's joint optimization objective is to minimize the user's overall energy cost and maximize the local photovoltaic absorption rate. The user's overall energy cost includes grid purchase cost, equipment maintenance cost, and carbon emission cost. The photovoltaic absorption rate is measured by the ratio of actual photovoltaic power generation to theoretically available photovoltaic power generation. The system innovatively designs a dual-objective optimization function F = w1·C1 + w2·C2, where w1 and w2 are the weight coefficients of the two objectives. This function is dynamically adjusted by a fuzzy logic controller based on the system's operating status and user preferences, achieving a dynamic balance between economic efficiency and environmental friendliness.

[0031] Fifth, the system employs a distributed iterative optimization algorithm based on the cactus contraction principle for the bi-objective optimization function using a single exponential time FPT algorithm. This algorithm first identifies the set of key parameters with the greatest impact on the objective function through sensitivity analysis. Then, it uses the cactus contraction function to perform a nonlinear transformation on the decision space, maintaining the original resolution in the core parameter direction while compressing the non-core parameter direction. Based on the compressed decision space, the system decomposes the global optimization problem into three sub-problems according to physical properties: photovoltaic control, energy storage control, and load control. The ADMM algorithm is then used for distributed iterative solving. By alternately updating the original and dual variables, the sub-problems are coordinated and optimized. Finally, through information exchange between the coordination variables and Lagrange multipliers, the system gradually approaches the global optimum, obtaining a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources.

[0032] The final step involves the system generating multi-timescale control commands based on a collaborative optimization strategy. These commands include day-ahead planning (24-hour optimization cycle, 1-hour resolution), hourly scheduling (4-6 hour optimization cycle, 15-minute resolution), and minute-level real-time control (5-15 minute optimization cycle, 1-minute resolution). Day-ahead planning generates energy storage charging and discharging plans and adjustable load operation schedules based on 24-hour forecast data. Hourly scheduling performs rolling optimization every hour based on real-time updated forecast data and system status. Minute-level real-time control generates immediate control commands based on actual system status and short-term forecast deviations, employing tracking control, corrective control, and emergency control strategies. The system then distributes these multi-timescale control commands in real-time to photovoltaic inverters, energy storage devices, and smart loads via a communication network, achieving dynamic matching and collaborative control of source-storage-load resources to ensure safe and stable system operation.

[0033] By organically combining these six steps, this invention achieves efficient and optimized coordinated operation of distributed photovoltaic and energy storage loads, which not only improves the photovoltaic absorption rate but also reduces the overall energy cost for users, while ensuring the safe and stable operation of the system.

[0034] The system architecture of this invention comprises four layers: a perception layer, a communication layer, a computing layer, and an application layer. The perception layer is responsible for data acquisition, the communication layer handles data transmission, the computing layer executes prediction and optimization algorithms, and the application layer provides a user interface and management functions. The system uses a unified device interface protocol to ensure standardized communication between photovoltaic inverters, energy storage devices, and smart load controllers and the system platform, supporting plug-and-play functionality. Furthermore, the system incorporates security mechanisms including data encryption, access control, and intrusion detection to ensure system operational security and user data privacy.

[0035] like Figure 2As shown, in step S1, an improved wave sorting algorithm is used to perform wave analysis and dynamic segmentation on multi-source heterogeneous data, extracting time-series features, periodic features, and environmentally relevant features to form a feature vector set, specifically including: Step S11: Perform anomaly detection on the meteorological data, photovoltaic equipment parameters and historical power generation data, identify outliers using the Z-score method, and fill in the missing data using cubic spline interpolation technology to obtain the preprocessed dataset; Step S11 involves the preprocessing of multi-source heterogeneous data. The system first collects three types of key data: meteorological data (including solar irradiance, ambient temperature, cloud cover, wind speed, etc.), photovoltaic equipment parameters (including panel type, installation angle, inverter efficiency, etc.), and historical power generation data. These raw data often contain missing values, outliers, or inconsistent formats, requiring systematic preprocessing. In specific implementation, the system adopts an anomaly detection method based on Z-scores, calculated as Z=(x-μ) / σ, where x is the data point, μ is the mean, and σ is the standard deviation. When |Z|>3, it is determined to be an outlier. For detected outliers, the system marks and replaces them. To address the issue of missing data, the system uses cubic spline interpolation, which fits the curves between data points using piecewise cubic polynomial functions, ensuring the smoothness and continuity of the interpolation results. For example, to address short-term missing data in meteorological data, the system constructs a cubic spline function S(t) = a0 + a1t + a2t² + a3t³ using data points from adjacent time periods to calculate estimates of the missing locations. Through these preprocessing steps, the system generates a complete, continuous, and reliable dataset, laying the foundation for subsequent feature extraction.

[0036] Step S12: Based on the preprocessed dataset, the time series data is divided into several segments, each segment being a wave unit. The amplitude, period, and energy characteristic parameters of each wave unit are calculated to obtain a set of wave characteristic parameters. Step S12 implements feature parameter extraction based on wave units. The system divides the preprocessed time series data into several segments, each segment being considered a wave unit. The segmentation method employs adaptive windowing technology, dynamically adjusting the segment length according to the data change rate, using shorter segments in rapidly changing regions and longer segments in stable regions. For each wave unit, the system calculates three types of key feature parameters: amplitude feature, periodic feature, and energy feature. Amplitude feature A wave A is obtained by calculating the difference between the maximum and minimum values ​​within a sub-segment. wave = max(x i ) - min(x i This reflects the intensity of the fluctuation. Periodic characteristic T wave The autocorrelation function analysis determined that the system calculates the autocorrelation coefficient r(k) for different time lags as r(k) = ∑[(xi -μ)(x i+k -μ)] / ∑(x i -μ)², find the time lag k at which the autocorrelation coefficient first reaches a local maximum, and determine it as the dominant period. Energy characteristic E wave E is obtained by calculating the mean of the sum of squares of the data points within the sub-segment. wave = ∑(x i ²) / N reflects the energy density of the wave. Through these calculations, the system constructs a set of wave characteristic parameters containing time series, periodicity, and energy information, providing a foundation for subsequent data classification and feature extraction.

[0037] In step S12, the amplitude, period, and energy characteristic parameters of each wave unit are calculated to obtain a set of wave characteristic parameters, specifically including: Step S121: Calculate the wave amplitude A based on each sub-segment of the preprocessed dataset. wave = max(x i ) - min(x i ), where x i For the data points within the sub-segment, the amplitude characteristics are obtained; Step S122: For the amplitude characteristics, determine the dominant period T of each segment through autocorrelation function analysis. wave Calculate the periodic characteristic parameters to obtain the periodic characteristics; Step S123: Calculate the energy characteristic E based on the amplitude and period characteristics. wave = ∑(x i ²) / N, where N is the segment length, yields the energy characteristics; Step S124: Based on the amplitude characteristics, periodic characteristics and energy characteristics, construct a set of wave feature parameters containing time series, periodic and energy information.

[0038] Step S13: Based on the wave feature parameter set, the data is sorted and classified using an improved wave sorting algorithm to extract time-series features, periodic features, and environmentally relevant features to form a feature vector set.

[0039] Step S13 employs an improved wave sorting algorithm to sort and classify the data. Traditional sorting algorithms require a strict total order relation, while wave sorting allows for local disorder but maintains global order, making it more suitable for processing photovoltaic data with random fluctuation characteristics. In its implementation, the system first constructs a feature space based on the wave feature parameter set obtained in step S12. Then, it uses the improved wave sorting algorithm, whose core idea is to group data segments with similar waveforms into the same cluster. The sorting process adopts a "divide and conquer" strategy, first sorting by the dominant period T... wave Group the data, and then sort the data within each group by amplitude A. waveand energy E wave The system performs sorting. After sorting, features are extracted from three dimensions: time-series features (including trend components, seasonal components, periodic components, and autocorrelation characteristics), periodic features (including daily cycle features, meteorological cycle features, and power grid load characteristics), and environmentally relevant features (including temperature-power relationships, irradiance-power relationships, etc.). These features are further enhanced through mathematical transformations (such as wavelet transform and Fourier transform), ultimately forming a high-dimensional feature vector set, providing rich input information for deep learning prediction models.

[0040] like Figure 3 As shown, in step S2, a deep learning hybrid model combining CNN and LSTM is constructed to predict the output of distributed photovoltaic power, obtaining the photovoltaic power prediction result, specifically including: Step S21: Based on the feature vector set, local feature patterns are extracted through a multi-layer CNN, convolutional kernels are used to perform convolution operations on the features, and the spatial feature map is obtained after dimensionality reduction processing by pooling layers. Step S21 implements CNN-based local feature extraction. The system uses a multi-layer convolutional neural network to process the feature vector set and extract local feature patterns. The CNN structure contains multiple convolutional layers, each using convolutional kernels of different sizes to capture features at different scales. In the actual implementation, the system adopts a 3-layer convolutional structure: the first layer uses 32 3×3 convolutional kernels, the second layer uses 64 3×3 convolutional kernels, and the third layer uses 128 3×3 convolutional kernels. Each convolutional layer is followed by a ReLU activation function to increase non-linear expressive power. The convolution operation can be represented as O(i,j)=∑∑I(i+m,j+n)·K(m,n), where I is the input feature map, K is the convolutional kernel, and O is the output feature map. Each convolutional layer is followed by a max pooling layer, using a 2×2 pooling window for downsampling, which reduces the number of parameters while preserving key features. For example, for input data containing multi-dimensional features such as irradiance, temperature, and cloud cover, the first convolutional layer might extract edge and texture features, the second layer might identify complex patterns such as cloud change features, and the third layer might capture high-level abstract features such as weather system change patterns. Through this hierarchical feature extraction, CNNs can effectively capture spatial correlations and local patterns in photovoltaic power generation data, forming rich spatial feature maps.

[0041] Step S22: Reshape the spatial feature map to form temporal input data, input it into the LSTM layer for long-term temporal dependency learning, and obtain temporal feature representation; Step S22 transforms the spatial features extracted by the CNN into temporal feature representations. The system first reshapes the feature map output by the CNN, converting the three-dimensional feature map (batch_size, height, width, channels) into a two-dimensional sequence (batch_size, time_steps, features), suitable for LSTM processing. The reshaped sequence data is then input into the LSTM layer for processing. The LSTM layer is specifically designed to capture long-term temporal dependencies. In this system, the LSTM layer adopts a two-layer structure, with each layer containing 128 LSTM units. Each LSTM unit contains three gating mechanisms: an input gate, a forget gate, and an output gate, enabling it to learn long-term dependencies and filter irrelevant information. The input gate controls the degree to which new information enters the cell state, the forget gate controls the degree to which old information is retained, and the output gate controls the degree to which the cell state is output. The mathematical expression for these three gates is: i t =σ(W i ·[h t-1 ,x t ]+b i ), f t =σ(W f ·[h t-1 ,x t ]+b f ), o t =σ(W o ·[h t-1 ,x t ]+b o ), where σ is the sigmoid function, W is the weight matrix, b is the bias vector, and h t-1 x is the hidden state from the previous moment. t This is the current input. Through this gating mechanism, LSTM can effectively learn long-term temporal dependencies in photovoltaic power generation data, such as intraday power generation patterns and the lagged effects of weather changes on power generation, forming rich temporal feature representations.

[0042] The model architecture employs a "CNN + LSTM + fully connected layer" structure. The input layer receives the feature vector after wave sorting. First, it extracts local feature patterns through multiple CNN layers. Each CNN layer contains multiple convolutional kernels, capable of recognizing features at different scales. The convolutional layers use small-sized kernels (e.g., 3×3) for local feature extraction, followed by dimensionality reduction and preservation of key features through pooling layers (max pooling or average pooling). The feature maps processed by the CNN are reshaped into sequences and input into the LSTM layer. The LSTM layer consists of multiple memory units, each containing an input gate, a forget gate, and an output gate, enabling it to learn long-term dependencies and filter out irrelevant information. The output of the LSTM layer undergoes an attention mechanism to highlight the influence of important time steps. Finally, the features are integrated through a fully connected layer, and the predicted result is output.

[0043] Step S23: The time-series feature representation is processed through an attention mechanism, the features are integrated through a fully connected layer, and the prediction result is output to obtain the photovoltaic power prediction result.

[0044] Step S23 integrates features through an attention mechanism and a fully connected layer, and outputs the prediction result. The system first applies an attention mechanism to the temporal features output by the LSTM to highlight the influence of important time steps. The calculation process of the attention mechanism is as follows: first, the attention score e at each time step is calculated. t =tanh(W_a·h t +b_a), and then normalize the score to weight α using the softmax function. t =softmax(e t Finally, calculate the weighted sum c = ∑α t ·h t As a context vector, for example, in photovoltaic power generation prediction, the system might assign higher attention weights to periods around sunrise and significant cloud cover changes. The attention-processed features are integrated through fully connected layers. The system employs a three-layer fully connected network structure: the first layer contains 256 neurons, the second layer contains 128 neurons, and the number of neurons in the third output layer is determined based on the prediction task (e.g., 24 neurons for short-term prediction, corresponding to a 24-hour prediction). ReLU activation functions are used between the fully connected layers to increase non-linear expressiveness, and the final layer uses a linear activation function to output the prediction result. To prevent overfitting, dropout technology is applied between the fully connected layers with a dropout rate of 0.3. Through this structure, the system can effectively integrate the spatial features extracted by CNN and the temporal features captured by LSTM, generating accurate photovoltaic power prediction results and providing a reliable basis for subsequent optimization decisions.

[0045] Specifically, regarding the hybrid model structure of CNN+LSTM+fully connected layers, assuming the input is feature data from the past 7 days, with 24 sampling points per day, each containing 10 features (such as irradiance, temperature, cloud cover, etc.), the input shape is (7, 24, 10). First, it is processed through three CNN layers: the first layer uses 32 3×3 convolutional kernels, with an output shape of (7, 24, 32); the second layer uses 64 3×3 convolutional kernels, with an output shape of (7, 24, 64); and the third layer uses 128 3×3 convolutional kernels, with an output shape of (7, 24, 128). Then, the CNN output is reshaped into a sequence of (168, 128) and input into a two-layer LSTM, with 128 units per layer, resulting in an output shape of (168, 128). Next, an attention mechanism is applied to calculate the context vector, with a shape of (128,). Finally, a three-layer fully connected network is used: the first layer has 256 neurons with an output shape of (256,); the second layer has 128 neurons with an output shape of (128,); and the third layer has 24 neurons with an output shape of (24,), corresponding to the photovoltaic power prediction value for the next 24 hours. This hybrid model structure fully utilizes the advantages of CNN in spatial feature extraction and LSTM in temporal modeling, accurately capturing the complex patterns of photovoltaic power generation data and providing high-precision prediction results.

[0046] In step S3, an improved single exponential time FPT algorithm combined with the cactus contraction principle is applied to perform a fast Fourier transform on the load time series data to obtain load demand forecast results and electricity price fluctuation forecast results, specifically including: Step S31: Perform a fast Fourier transform on the user's historical electricity consumption data and electricity price information to convert the time-domain signal into a frequency-domain representation, separate the main frequency component and the secondary frequency component, and obtain the spectrum data; Step S31 involves frequency domain transformation of user historical electricity consumption data and electricity price information. The system first collects multi-source electricity data, including historical electricity consumption data from smart meters, user electrical equipment information, basic household or building information, and historical electricity price data. This data typically exhibits clear temporal characteristics and periodic patterns, such as intraday fluctuations, intraweekly differences, and seasonal variations. To effectively capture these characteristics, the system performs a Fast Fourier Transform (FFT) on the collected time-series data, converting the time-domain signal into a frequency-domain representation. This transformation process allows the system to separate the dominant and secondary frequency components in the data. The dominant frequency component usually corresponds to the main periodic changes in the data, such as daily, weekly, and seasonal cycles; the secondary frequency component corresponds to random fluctuations and noise. Through frequency domain analysis, the system can more clearly identify key periodic patterns in load and electricity price data, such as weekday and weekend electricity consumption differences and seasonal electricity price changes. This frequency domain representation facilitates subsequent data compression and feature extraction, enabling the system to focus on the most informative parts of the data, improving prediction efficiency and accuracy.

[0047] Step S32: The spectral data is processed using the cactus contraction function. The main frequency features are preserved and the secondary frequency information is compressed by S(f) = f × [1 -exp(-α|f| / fmax)] to obtain the compressed spectral features. Step S32 implements spectral data processing based on the cactus contraction function. Traditional spectral analysis methods often require processing all frequency components, resulting in high computational complexity and susceptibility to noise. This system innovatively applies the cactus contraction function to process spectral data, inspired by the survival strategy of cacti in water-scarce environments. The cactus contraction function applies varying degrees of compression to different frequency components, preserving primary frequency characteristics while compressing secondary frequency information. The function design ensures that high-energy primary frequency components are almost unaffected, while low-energy secondary frequency components are significantly compressed, thereby greatly reducing computational complexity while maintaining the core characteristics of the data. This processing method is particularly suitable for load and electricity price data, as such data typically contains a clear dominant cycle and a large number of secondary fluctuations. For example, for residential electricity consumption data, the system retains the daily cycle component reflecting morning and evening peaks while compressing random fluctuation components; for electricity price data, the system retains the time-of-use pricing pattern reflecting peak and off-peak periods while compressing high-frequency components of market fluctuations. Through this intelligent compression, the system significantly improves computational efficiency while maintaining prediction accuracy, enabling real-time control.

[0048] The cactus contraction principle originates from biological observation: cacti contract in water-scarce environments to reduce water evaporation while maintaining their basic structure. In algorithm design, this principle is abstracted as the idea of ​​"compressing non-critical information while preserving the core characteristics of the data." In its implementation, the algorithm first performs a Fast Fourier Transform (FFT) on the load time-series data, converting the time-domain signal into a frequency-domain representation and separating the dominant and secondary frequency components. Then, it processes the spectrum using a cactus contraction function, which can be expressed as: S(f) = f × [1 - exp(-α|f| / fmax)], where f is the frequency component, fmax is the maximum frequency, and α is the contraction parameter. This processing preserves the main frequency characteristics while compressing secondary frequency information, significantly reducing computational complexity.

[0049] Step S33: Based on the compressed spectrum characteristics, combined with historical data and recent trends, a segmented single-exponential time complexity calculation method is used for prediction modeling to obtain load demand prediction results and electricity price fluctuation prediction results.

[0050] Step S33 performs predictive modeling based on the compressed spectral features. The system employs a piecewise single-exponential time complexity calculation method, an algorithm specifically optimized for time series forecasting. Traditional time series forecasting algorithms, such as ARIMA or neural network models, typically have high computational complexity, making them difficult to meet the needs of real-time control. This system's piecewise method divides the forecast interval into multiple sub-intervals, applying a single-exponential time complexity algorithm to each sub-interval, significantly reducing the overall computational complexity. During the forecasting process, the system combines historical data from the same period, recent trends, and the influence of external factors to construct a comprehensive forecasting model. For load forecasting, the system considers user behavior patterns, weather conditions, and the impact of special events; for electricity price forecasting, the system considers market supply and demand, energy policies, and information on major events. The forecast results include short-term (within 24 hours) and medium-term (within 7 days) load demand and electricity price fluctuation forecasts, output in time series form with confidence interval information, providing reliable input for subsequent optimization decisions. This efficient forecasting method enables the system to quickly generate accurate forecast results with limited computing resources, supporting real-time optimization decisions.

[0051] In step S4, a multi-agent collaborative optimization model is established with the joint optimization objectives of minimizing the user's comprehensive energy cost and maximizing the local photovoltaic absorption rate, resulting in a dual-objective optimization function, which specifically includes: Step S41: Based on the photovoltaic power prediction results, load demand prediction results and electricity price fluctuation prediction results, construct the user's comprehensive energy cost objective function, including grid electricity purchase cost, equipment maintenance cost and carbon emission cost, and obtain the cost objective function; Step S41 constructs the user's comprehensive energy cost objective function. Based on the photovoltaic power forecast, load demand forecast, and electricity price fluctuation forecast obtained in the preceding steps, the system constructs a comprehensive cost objective function. This function considers three main types of costs: grid purchase cost, equipment maintenance cost, and carbon emission cost. Grid purchase cost is the direct economic expenditure of the user in purchasing electricity from the grid, obtained by the system by calculating the sum of the products of the grid purchase volume and the corresponding electricity price at different time periods. Equipment maintenance cost mainly considers the charge-discharge cycle life loss of energy storage equipment; the system estimates maintenance costs based on the number of charge-discharge cycles, depth of charge-discharge, and battery cost. Carbon emission cost is calculated by multiplying the purchased volume by the carbon emission factor and carbon price, reflecting the environmental impact of electricity consumption behavior. These three types of costs together constitute the user's comprehensive energy cost, comprehensively reflecting the user's actual expenditure in both economic and environmental aspects. Through this multi-dimensional cost construction, the system can more comprehensively evaluate the economics of different operating strategies, providing users with a truly comprehensive optimization solution, rather than simply optimizing electricity costs.

[0052] Step S42: Based on the ratio of actual photovoltaic power generation to theoretically available photovoltaic power generation, construct the photovoltaic grid connection rate objective function to obtain the grid connection rate objective function; Step S42 constructs the photovoltaic (PV) grid integration rate objective function. Besides economic efficiency, improving the local PV grid integration rate is also a crucial objective of distributed energy systems. The system constructs the PV grid integration rate objective function based on the ratio of actual PV power generation used to theoretically available PV power generation. Actual PV power generation used includes both locally directly consumed and stored power for later use, while theoretically available PV power generation is the maximum possible power generation of the PV system under predicted conditions. To maintain a minimization direction consistent with the cost objective, the system takes a negative value for the grid integration rate, transforming it into a minimization problem. This objective function directly reflects the system's efficiency in utilizing renewable energy and is significant for promoting clean energy consumption and reducing curtailment. In practical applications, improving the PV grid integration rate not only has environmental benefits but also enhances the user's return on investment and extends equipment lifespan, making it a key objective pursued by distributed energy systems. By treating the PV grid integration rate as an independent optimization objective, the system can find a balance between economic efficiency and environmental protection, achieving more comprehensive system optimization.

[0053] Step S43: Based on the cost objective function and the absorption rate objective function, a fuzzy logic controller is used to dynamically adjust the weight coefficients of the two objectives according to the system operating status and user preferences to obtain a dual-objective optimization function.

[0054] Step S43 achieves adaptive adjustment of multi-objective weights. Traditional multi-objective optimization typically uses fixed weights, which are difficult to adapt to dynamic changes in system operating status. This system innovatively employs a fuzzy logic controller to dynamically adjust the weight coefficients of the two objectives based on system operating status and user preferences. The system first collects four key indicators: the ratio of predicted photovoltaic output to actual electricity demand, the current state of charge of the battery, the grid electricity price level, and the user-set preference coefficients. Based on these indicators, the system establishes a fuzzy logic rule base to achieve intelligent weight adjustment. For example, when photovoltaic power generation far exceeds load demand and the grid electricity price is low, the system increases the weight of photovoltaic absorption rate, prioritizing environmental benefits; when photovoltaic power generation is insufficient and the electricity price is high, the system increases the weight of cost objective, prioritizing economic benefits. This dynamic adjustment mechanism enables the system to automatically balance economic and environmental benefits according to real-time conditions, achieving optimal comprehensive benefits in different scenarios. Through this adaptive weight adjustment, the system forms a true bi-objective optimization function, capable of finding the optimal balance between economic and environmental benefits under various operating conditions, providing more intelligent and flexible optimization decisions.

[0055] In step S43, a fuzzy logic controller is used to dynamically adjust the weight coefficients of the two objectives based on the system operating state and user preferences, resulting in a bi-objective optimization function, which specifically includes: Step S431: Collect the ratio of the photovoltaic power output forecast to the actual electricity demand, the current state of charge of the battery, the grid electricity price level and the user's set preference coefficient, construct the state vector for weight adjustment, and obtain the weight adjustment input parameters; Step S432: Adjust the input parameters based on the weights, establish a fuzzy logic rule base, increase the weight of photovoltaic absorption rate when photovoltaic power generation is much greater than the load demand and the grid electricity price is low, and increase the cost weight when photovoltaic power generation is insufficient and the electricity price is high, so as to obtain fuzzy logic rules. Step S433: Calculate the weight adjustment amount Δw1 according to the fuzzy logic rule, and update the weight coefficients through w1(t+1) = w1(t) +Δw1 and w2(t+1) = 1 - w1(t+1) to obtain the adaptive weight coefficients; Step S434: Based on the adaptive weight coefficients, construct a bi-objective optimization function F = w1·C1 + w2·C2, where C1 is the user's comprehensive energy cost and C2 is the negative value of the photovoltaic absorption rate, thus obtaining the bi-objective optimization function.

[0056] The bi-objective optimization function is mathematically expressed as follows: min F = w1·C1 + w2·C2, where w1 and w2 are the weight coefficients of the two sub-objectives, respectively, and satisfy w1+w2=1; C1 represents the user's comprehensive energy cost, and C2 represents the negative value of the photovoltaic absorption rate (transformed into a minimization problem). The user's comprehensive energy cost C1 is further subdivided into several components: C1 = C elek + C ma ᵢ nt + C em ᵢ ss ᵢ on C elek This represents the cost of electricity purchased from the power grid, calculated as the sum of the products of the amount of electricity purchased from the grid and the corresponding electricity price at different time periods; C ma ᵢ nt This indicates equipment maintenance costs, primarily considering the charge / discharge cycle life losses of energy storage devices; C em ᵢ ss ᵢ on The cost of carbon emissions is represented by the product of electricity purchased, the carbon emission factor, and the carbon price. The photovoltaic grid integration target C2 is defined as: C2 = -∑P p ᵥ_ᵤ se / ∑P p ᵥ_ a ᵥ a ᵢ l , where P p ᵥ_ᵤ se P represents the actual amount of photovoltaic power generated (locally consumed or stored). p ᵥ_ a ᵥ a ᵢ l This represents the theoretically available photovoltaic power generation. Negating this ratio transforms the problem into a minimization problem, aligning it with the cost objective.

[0057] In step S5, a single exponential time FPT algorithm based on the cactus contraction principle is used for distributed iterative optimization. Problem decomposition is achieved through core parameter identification and decision space compression, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources. Specifically, this includes: Step S51: The bi-objective optimization function is represented as a parameterized problem, and the set of key parameters that have the greatest impact on the objective function is identified through sensitivity analysis to obtain the core parameter set; Step S51 achieves the parameterized representation of the optimization problem and the identification of core parameters. In large-scale distributed energy systems, the number of decision variables is enormous, making it difficult for traditional optimization methods to handle efficiently. This system first transforms the bi-objective optimization function into a parameterized problem representation, which describes the original problem as a combination of problem instances and key parameters. Then, the system identifies the set of key parameters that have the greatest impact on the objective function through sensitivity analysis. Sensitivity analysis uses a local perturbation method, applying small changes to each decision variable and observing the degree of response of the objective function to quantify the importance of each parameter. Through this analysis, the system identifies a set of core parameters, which typically include the energy storage discharge power during peak electricity price periods, the adjustment range during periods of large fluctuations in photovoltaic output, and load adjustment parameters sensitive to user comfort. These core parameters have a decisive impact on the optimization results, while other parameters have a relatively smaller impact. By identifying core parameters, the system can concentrate computational resources on the most critical decision variables, significantly improving optimization efficiency and laying the foundation for subsequent decision space compression.

[0058] Step S52: The decision space is nonlinearly transformed using the cactus contraction function. The original resolution is maintained in the core parameter direction, while compression is performed in the non-core parameter direction to obtain the compressed decision space. Step S52 employs a cactus shrinkage function to perform a nonlinear transformation on the decision space. Traditional optimization methods typically require searching for the optimal solution within the complete decision space, resulting in heavy computational burden and low efficiency. This system innovatively applies a cactus shrinkage function to intelligently compress the decision space. This function maintains the original resolution in the core parameter direction while compressing it in the non-core parameter direction. This nonlinear transformation allows the decision space to maintain a fine structure in important dimensions and appropriately simplify it in secondary dimensions, ensuring the accuracy of the optimization results while significantly reducing computational complexity. For example, for energy storage charging and discharging decisions, the system maintains high resolution during peak and off-peak electricity price periods, while using lower resolution during periods of stable electricity prices; for load adjustment decisions, the system maintains fine control during peak user activity periods and uses coarse-grained control during off-peak periods. This intelligent compression strategy enables the system to efficiently explore the vast decision space with limited computational resources and find a solution close to the global optimum. Through decision space compression, the system transforms the originally difficult high-dimensional optimization problem into a simplified problem solvable in a real-time environment.

[0059] Step S53: Based on the compressed decision space, the global optimization problem is decomposed into photovoltaic control sub-problems, energy storage control sub-problems, and load control sub-problems according to physical properties, to obtain a combination of independently solvable sub-problems; Step S53 decomposes the global optimization problem into multiple sub-problems. Even after decision space compression, the global optimization problem remains complex and difficult to solve directly. This system employs a physical attribute decomposition strategy, breaking down the global problem into three types of sub-problems: photovoltaic control, energy storage control, and load control. The photovoltaic control sub-problem optimizes the active and reactive power setpoints of the photovoltaic inverter to maximize photovoltaic utilization and support grid voltage regulation; the energy storage control sub-problem optimizes the charging and discharging power and state of energy storage devices to achieve electricity price arbitrage and peak-valley smoothing; and the load control sub-problem optimizes the power setpoint and operating state of adjustable loads to achieve load-side response while ensuring user comfort. These three types of sub-problems each focus on different types of physical devices and have relatively independent constraints and control objectives, facilitating parallel solutions. Simultaneously, the system identifies and preserves key coupling relationships between sub-problems, such as power balance coupling, voltage influence coupling, and objective function coupling, ensuring that the decomposed sub-problem combination correctly expresses the original problem. By decomposing the problem based on physical properties, the system implements a "divide and conquer" optimization strategy, transforming complex problems into multiple manageable sub-problems, thus laying the foundation for distributed solution.

[0060] Step S54: The ADMM algorithm is used to solve the combination of subproblems in a distributed iterative manner. By alternately updating the original variables and dual variables, the coordinated optimization of each subproblem is achieved, and a local optimal solution set is obtained. Step S54 employs the ADMM algorithm for distributed iterative solution of the combined subproblems. ADMM (Alternating Direction Multipliers) is an algorithmic framework suitable for distributed optimization, capable of coordinating the solution process of multiple subproblems and progressively approximating the global optimum. In this system, the ADMM algorithm first transforms the constrained optimization problem into an augmented Lagrangian form, and then achieves iterative solution by alternately updating the original and dual variables. Specifically, the system allocates a computational unit to each subproblem, and these units can perform local optimization calculations in parallel. In each iteration, each unit first solves its own subproblem based on the current Lagrange multipliers and the results from the previous round of other units; then, the coordinator collects the optimization results from each unit, updates the Lagrange multipliers, and sends the updated information back to each unit to begin the next iteration. To improve the algorithm's convergence speed, the system implements several innovative techniques, including adaptive penalty parameter adjustment, a prediction-correction mechanism, and over-relaxation techniques. These techniques significantly improve the algorithm's convergence efficiency, enabling the system to obtain high-quality optimization results within a finite number of iterations. By using the distributed iterative solution of the ADMM algorithm, the system achieves the dispersion of computational load and the improvement of optimization efficiency, overcoming the bottleneck problem of centralized optimization.

[0061] Step S55: Based on the local optimal solution set, through information exchange of coordination variables and Lagrange multipliers, gradually approach the global optimal solution to obtain the collaborative optimization strategy of photovoltaic power generation unit, energy storage unit and adjustable load resources.

[0062] Step S55 gradually approaches the global optimum through information exchange of coordination variables and Lagrange multipliers. In distributed optimization, subproblems need to exchange information to ensure overall consistency. This system designs an information exchange mechanism based on coordination variables and Lagrange multipliers to achieve efficient collaboration between subproblems. Coordination variables typically include shared information such as net injected power at nodes, voltage at key nodes, and total system power balance; Lagrange multipliers reflect the sensitivity of constraints and the optimization direction. The system adopts a "lightweight communication" strategy, transmitting only variable information that changes significantly during iteration, reducing communication burden. Simultaneously, the system implements a hierarchical communication architecture, including device-level communication, region-level communication, and global coordination communication, improving the efficiency and reliability of information exchange. To address communication delays and data loss, the system designs a prediction compensation mechanism and a fault-tolerant retry mechanism, enhancing the robustness of distributed optimization. Through multiple iterations and information exchange, the local optima of each subproblem gradually become consistent, ultimately approaching the global optimum, forming a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources. This distributed collaborative optimization method not only solves the problem of computational efficiency, but also improves the scalability and fault tolerance of the system.

[0063] The basic framework of the FPT algorithm includes three steps: parameterized problem representation, core parameter extraction, and parameterized solution. In this system, the specific implementation of the FPT algorithm includes core parameter identification, decision space compression, and rapid expansion techniques. Through sensitivity analysis, the set of key parameters K that has the greatest impact on the objective function is identified. These parameters typically include: energy storage discharge power during peak electricity price periods, adjustment range during periods of large fluctuations in photovoltaic output, and load adjustment parameters sensitive to user comfort. The decision space is nonlinearly transformed using the cactus contraction function C(x,K). This function maintains the original resolution in the direction of core parameters and compresses the space in the direction of non-core parameters. Mathematically, C(x,K) = x · [α + (1-α)·exp(-β·d(x,K))], where d(x,K) represents the "distance" from the decision variable x to the core parameter set K, and α and β are compression parameters.

[0064] In step S6, multi-timescale control commands are generated, including day-ahead planning, hourly scheduling, and minute-level real-time control. These control commands are then transmitted in real-time to photovoltaic inverters, energy storage devices, and smart loads via a communication network, enabling dynamic matching and coordinated control of source, storage, and load resources. Specifically, this includes: Step S61: Based on the collaborative optimization strategy, and using 24-hour photovoltaic output forecast, load demand forecast, and electricity price forecast data, generate an energy storage charging and discharging plan and a day-ahead optimization plan for the adjustable load operating period, and obtain day-ahead control instructions; Step S61 generates a day-ahead optimization plan, providing long-term economic guidance for system operation. Based on 24-hour photovoltaic output forecasts, load demand forecasts, and electricity price forecasts, the system generates an optimized operation plan for the following day. The day-ahead plan generation process includes several steps: First, the system integrates forecast data and generates multiple scenarios covering optimistic, standard, and pessimistic conditions; then, a complete optimization model is applied to each scenario to solve for the optimal operation plan under that scenario; next, based on the optimization results of multiple scenarios, a robust plan is generated through scenario synthesis; finally, the system performs economic benefit assessment and feasibility verification on the generated plan to ensure its economic efficiency and feasibility. The main contents of the day-ahead plan include energy storage charging and discharging plans, adjustable load operating time arrangements, and photovoltaic power plans, which are output in a standard format as a reference for hourly scheduling. The time resolution of the day-ahead plan is typically 1 hour, the optimization cycle is 24 hours, and it is executed once daily at a fixed time (e.g., 8:00 PM). Through the day-ahead optimization plan, the system can plan energy use strategies in advance, fully utilize electricity price differences and renewable energy, and provide users with long-term economic guidance.

[0065] Step S62: Based on real-time updated forecast data and system status, perform rolling optimization every hour to update the equipment operation plan for the next 4 to 6 hours and obtain hourly scheduling instructions; Step S62 implements hourly rolling scheduling, balancing economy and real-time adaptability. Due to prediction errors and changes in system state, day-ahead plans may not fully adapt to real-time conditions. This system adopts an hourly rolling scheduling strategy to dynamically optimize the operation plan for the next 4-6 hours. Rolling scheduling uses a "shift-forward time domain" method, with each optimization covering the next 4-6 hours, and a time resolution of 15 minutes. The scheduling trigger mechanism includes time triggering (automatically executed every hour on the hour) and event triggering (triggered when the system detects that the prediction deviation exceeds a threshold or a significant change in equipment state). Before each scheduling execution, the system updates short-term prediction data and calibrates the current system state, especially the SOC value of energy storage devices and the operating status of each device. The system dynamically adjusts optimization constraints based on the day-ahead plan and the current state to ensure that rolling optimization can flexibly respond to real-time changes without deviating from the overall goal of the day-ahead plan. Considering real-time requirements, the system uses a simplified optimization model and efficient solution algorithm to ensure that optimization calculations can be completed within 5 minutes. After optimization, the system generates three types of execution plans: a near-term execution plan (detailed control plan for the next hour), a reference plan (control reference for the next 1-4 hours), and a deviation analysis report. Through hourly rolling scheduling, the system achieves a dynamic balance between economy and adaptability, providing reliable guidance for minute-level control.

[0066] Step S63: Based on the actual system status and short-term prediction deviation, adopt tracking control, correction control and emergency control strategies to generate minute-level real-time control commands for photovoltaic inverter power regulation, energy storage charging and discharging power and load response intensity, and obtain minute-level control commands; Step S63 generates minute-level real-time control commands, directly targeting the device execution layer. Minute-level real-time control is the lowest level of the entire control system, requiring rapid response to system state changes and ensuring safe and stable operation. The system collects key operational data at a high frequency (typically 1-10 seconds), including the real-time status of photovoltaic inverters, energy storage devices, adjustable loads, and distribution network nodes. Based on the collected data, the system assesses the current operating status and calculates the deviation from the hourly scheduling. Depending on the degree of deviation, the system automatically selects one of three control strategies: when the deviation is within the allowable range, it executes reference trajectory tracking control based on hourly scheduling; when the deviation exceeds the threshold but remains controllable, it executes correction control to pull the system state back to the reference trajectory; when a safety risk or significant deviation occurs, it temporarily ignores economic objectives and prioritizes safety control. Based on the selected control strategy, the system calculates control commands for each device, including the power setpoint of the photovoltaic inverter, the charging and discharging power of the energy storage device, and the power adjustment amount of the adjustable load. The generated control commands must pass multiple constraint checks to ensure they meet physical constraints, safety constraints, user constraints, and lifespan constraints. With minute-level real-time control, the system can respond quickly to instantaneous changes, ensuring the safe and stable operation of the system, which is a key link in the entire control system.

[0067] Step S64: Perform protocol conversion on the day-ahead control command, hourly scheduling command, and minute-level control command, using Modbus, SunSpec, or BACnet protocol conversion technology to obtain standardized control commands; Step S64 performs protocol conversion of control commands, ensuring that different devices can correctly identify and execute the commands. In practical applications, photovoltaic inverters, energy storage devices, and smart loads typically come from different manufacturers and support different communication protocols. This system is designed with a standardized protocol conversion mechanism to convert optimized control commands into a standard format recognizable by various devices. The system supports multiple mainstream communication protocols, including Modbus (suitable for most industrial equipment and inverters), SunSpec (a standard protocol for photovoltaic inverters), BACnet / KNX (suitable for load control in building automation systems), and custom protocols from various manufacturers. The protocol conversion process includes three stages: data format conversion, address mapping, and command encapsulation, ensuring that control commands can be correctly parsed and executed by the target device. The system also implements a command confirmation mechanism, requiring devices to return execution confirmation information to ensure that commands are correctly received and executed. Through standardized protocol conversion, the system achieves unified control of heterogeneous devices, overcomes the technical barriers of inconsistent device interfaces, and provides technical support for the coordinated control of source-storage-load resources.

[0068] Step S65: The standardized control commands are sent to the photovoltaic inverter, energy storage device and smart load in real time through the communication network to realize dynamic matching and coordinated control of source, storage and load resources.

[0069] Step S65 distributes control commands to each device via the communication network, achieving dynamic matching and coordinated control of source-storage-load resources. The system constructs a communication network based on IoT technology, supporting wired and wireless hybrid connections to ensure reliable transmission of control commands and real-time feedback of device status. The communication network adopts a layered architecture, including a field layer network (connecting terminal devices), a regional layer network (connecting regional controllers), and a central layer network (connecting to the central management system). The system supports multiple communication technologies, including RS485 / Modbus, Ethernet, WiFi, ZigBee, 4G / 5G, etc., selecting the most suitable communication method according to different scenarios. To improve communication reliability, the system implements communication redundancy design and a fault self-healing mechanism, automatically switching to a backup channel when the primary communication channel fails. Command distribution uses a priority queue mechanism to ensure that critical commands (such as security-related commands) are processed first. The system also implements a command execution monitoring mechanism, continuously monitoring device response status, including response time, execution accuracy, and abnormal behavior, and automatically executing preset processing procedures for abnormal situations. Through this complete communication and control execution mechanism, the system realizes dynamic matching and coordinated control of source-storage-load resources, ensuring that optimization strategies can be accurately executed and maximizing the system's economic and environmental benefits.

[0070] The hierarchical control framework comprises three layers: the day-ahead planning layer, the hourly scheduling layer, and the minute-level real-time control layer, forming a progressively refined control structure from top to bottom. The day-ahead planning layer has a time resolution of 1 hour, an optimization cycle of 24-hour rolling optimization, and executes once daily at a fixed time. Its controlled objects are energy storage charging and discharging plans, adjustable load start / stop plans, and operating segment arrangements. The hourly scheduling layer has a time resolution of 15 minutes, an optimization cycle of 4-6 hours rolling optimization, and executes hourly or is triggered when the system state changes significantly. Its controlled objects are energy storage charging and discharging power curves, adjustable load power curves, and photovoltaic power limitations. The minute-level real-time control layer has a time resolution of 1 minute or less, an optimization cycle of 5-15 minutes rolling optimization, executes quasi-real-time cyclically, or is triggered by system emergencies. Its controlled objects are immediate control commands for equipment.

[0071] like Figure 4 As shown, the present invention also provides a distributed photovoltaic and energy storage load collaborative optimization operation system, comprising: The data acquisition module is used to acquire meteorological data, photovoltaic equipment parameters and historical power generation data, as well as users' historical electricity consumption data and electricity price information, and transmit the data to the predictive analysis module. The predictive analysis module receives data from the data acquisition module, uses an improved wave sorting algorithm and a deep learning hybrid model to predict photovoltaic output, and uses a single exponential time FPT algorithm to predict load demand and electricity price, providing the prediction results to the optimization decision module. The optimization decision module receives the prediction results from the prediction analysis module, establishes a dual-objective collaborative optimization model, uses a distributed iterative optimization algorithm based on the cactus contraction principle to solve the collaborative optimization strategy, and outputs the optimization strategy to the control execution module. The control execution module receives the optimization strategy from the optimization decision module, generates multi-timescale control commands, and sends them to photovoltaic inverters, energy storage devices, and smart loads through the communication network to achieve coordinated control.

[0072] System integration and real-time control execution encompasses four aspects: communication network construction and data transmission, equipment control interface development, control command execution and monitoring, and system operation evaluation and adaptive optimization. The communication network utilizes IoT technology, supporting both wired and wireless hybrid connections to ensure reliable transmission of control commands and real-time feedback of equipment status. The equipment control interface supports mainstream communication protocols such as Modbus and DLT645, enabling unified control of equipment from different manufacturers. The control command execution mechanism includes command verification, priority processing, and execution status monitoring to ensure correct execution of commands by the equipment. System operation evaluation indicators include economic benefits, photovoltaic grid integration rate, and system stability. Based on the evaluation results, the optimization model parameters are adaptively adjusted to form a closed-loop control. Example

[0073] The method of this invention is applied to a distributed photovoltaic and energy storage system in an industrial park. This system includes a 500kW rooftop photovoltaic system, a 200kWh energy storage system, and a 300kW adjustable load. First, the system collected three months of historical data, including meteorological data (irradiance, temperature, cloud cover), photovoltaic power generation data, load consumption data, and electricity price information. An improved wave sorting algorithm was used to process this multi-source heterogeneous data, dividing the time-series data into several wave units. The amplitude, period, and energy characteristic parameters of each unit were calculated to form a feature vector set. Based on this feature vector set, a hybrid CNN and LSTM model was constructed to predict photovoltaic output. Simultaneously, a single exponential time FPT algorithm combined with the cactus contraction principle was applied to predict load demand and electricity price.

[0074] During the optimization phase, the system established a dual-objective optimization function, with weight coefficients dynamically adjusted by a fuzzy logic controller based on the system state. A single-exponential-time FPT algorithm based on the cactus contraction principle was employed for distributed iterative optimization, reducing the computational complexity from O(n²) to O(n·exp(c)) and shortening the optimization computation time from 15 minutes to 30 seconds, meeting real-time control requirements. The system generated multi-timescale control commands, including daily planning, hourly scheduling, and minute-level control, which were distributed to various devices for execution via the communication network.

[0075] Application results show that, compared with traditional methods, the method of this invention increases the photovoltaic absorption rate of the park by 18.7%, reduces electricity costs by 15.3%, significantly improves system operation stability, and effectively solves the collaborative optimization problem under the condition of high proportion of distributed photovoltaic access. Example

[0076] The method of this invention was applied in a smart microgrid in a residential community. This microgrid includes 200 residential users, a total of 300kW of distributed photovoltaic power, 150kWh of community energy storage, and various types of adjustable household loads. The system first collected various types of data, including historical meteorological data, photovoltaic power generation data, time-of-use electricity consumption data, and electricity price information. An improved wave sorting algorithm was used to process the data, extracting time-series features, periodic features, and environmentally relevant features to form a feature vector set. Based on this feature vector set, the system constructed a hybrid deep learning model combining CNN and LSTM to predict photovoltaic output, achieving a prediction accuracy (MAPE) of 8.5%, which is 27% higher than traditional methods. Simultaneously, the system applied a single exponential time FPT algorithm combined with the cactus contraction principle to predict user load demand and electricity price fluctuations.

[0077] During the optimization decision-making phase, the system establishes a collaborative optimization model with economic efficiency and photovoltaic (PV) grid integration rate as dual objectives. A fuzzy logic controller dynamically adjusts the weight coefficients of the two objectives. When PV power generation is sufficient and electricity prices are low, the system automatically increases the weight of the PV grid integration rate; when PV power generation is insufficient and electricity prices are high, the system increases the weight of the economic efficiency objective. A single exponential time (FPT) algorithm based on the cactus contraction principle is used for distributed iterative optimization. The system decomposes the global optimization problem into three sub-problems: PV control, energy storage control, and load control, and achieves distributed solution through the ADMM algorithm.

[0078] The system generates control commands at multiple time scales based on the optimization results, including a 24-hour daily plan, a 4-6 hour rolling scheduling plan, and minute-level real-time control commands. These commands are distributed to various devices for execution via the Internet of Things (IoT) communication network, enabling dynamic matching and coordinated control of source, storage, and load resources.

[0079] Application results show that, compared with traditional methods, the method of this invention increases the self-consumption rate of photovoltaic power in the community by 21.5%, reduces residents' electricity expenses by 12.8%, and ensures both system operational stability and user comfort. Especially under weather conditions with large fluctuations in photovoltaic output, the system still maintains good coordinated control, demonstrating strong robustness.

[0080] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0081] It should be noted that those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention. If such modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include such modifications and variations.

[0082] This disclosure also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of the distributed photovoltaic and energy storage load collaborative optimization operation method described in the above-described method embodiments. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0083] Furthermore, this disclosure also provides a computer program product storing a computer program. When the computer program is run by a processor, it executes the steps of any of the above-described distributed photovoltaic and energy storage load coordinated optimization operation methods. For details, please refer to the above-described method embodiments, which will not be repeated here.

[0084] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium, which can be a volatile or non-volatile computer-readable storage medium. In another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and apparatuses described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this disclosure, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0086] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] In addition, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0088] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for coordinated and optimized operation of distributed photovoltaic and energy storage loads, characterized in that, Includes the following steps: Meteorological data, photovoltaic equipment parameters, and historical power generation data are acquired. An improved wave sorting algorithm is used to perform wave analysis and dynamic segmentation on the multi-source heterogeneous data, extracting time-series features, periodic features, and environmentally relevant features to form a feature vector set. Based on the feature vector set, a deep learning hybrid model combining CNN and LSTM is constructed to predict the output of distributed photovoltaic power and obtain the photovoltaic power prediction result. Collect users' historical electricity consumption data and electricity price information, and apply the improved single exponential time FPT algorithm combined with the cactus contraction principle to perform fast Fourier transform on the load time series data to obtain load demand forecast results and electricity price fluctuation forecast results. Based on the photovoltaic power prediction results, load demand prediction results, and electricity price fluctuation prediction results, a multi-entity collaborative optimization model is established with the joint optimization objectives of minimizing the comprehensive energy cost of users and maximizing the local photovoltaic absorption rate, resulting in a dual-objective optimization function. The bi-objective optimization function is optimized by a distributed iterative algorithm based on the cactus contraction principle using a single exponential time FPT algorithm. Problem decomposition is achieved through core parameter identification and decision space compression, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources. Based on the aforementioned collaborative optimization strategy, multi-timescale control commands are generated, including day-ahead planning, hourly scheduling, and minute-level real-time control. These control commands are then transmitted in real-time to photovoltaic inverters, energy storage devices, and smart loads via a communication network, enabling dynamic matching and collaborative control of source, storage, and load resources.

2. The method according to claim 1, characterized in that, An improved wave sorting algorithm is used to perform wave analysis and dynamic segmentation on the multi-source heterogeneous data, extracting time-series features, periodic features, and environment-related features to form a feature vector set, including: Anomaly detection is performed on the meteorological data, photovoltaic equipment parameters, and historical power generation data. Outliers are identified using the Z-score method, and missing data is filled in using cubic spline interpolation to obtain a preprocessed dataset. Based on the preprocessed dataset, the time series data is divided into several segments, each segment being a wave unit. The amplitude, period, and energy characteristic parameters of each wave unit are calculated to obtain a set of wave characteristic parameters. Based on the wave feature parameter set, an improved wave sorting algorithm is used to sort and classify the data, extract time-series features, periodic features, and environmentally relevant features to form a feature vector set.

3. The method according to claim 1, characterized in that, A hybrid deep learning model combining CNN and LSTM is constructed to predict distributed photovoltaic power output, yielding photovoltaic power prediction results, including: Based on the feature vector set, local feature patterns are extracted through a multi-layer CNN, convolutional kernels are used to perform convolution operations on the features, and dimensionality reduction is performed through pooling layers to obtain a spatial feature map. The spatial feature map is reshaped to form temporal input data, which is then input into the LSTM layer for long-term temporal dependency learning to obtain a temporal feature representation. The time-series features are processed through an attention mechanism, integrated through a fully connected layer, and the prediction results are output to obtain the photovoltaic power prediction results.

4. The method according to claim 1, characterized in that, An improved single-exponential-time FPT algorithm, combined with the cactus contraction principle, is applied to perform a Fast Fourier Transform on load time-series data to obtain load demand forecasting results and electricity price fluctuation forecasting results, including: The user's historical electricity consumption data and electricity price information are subjected to a fast Fourier transform to convert the time-domain signal into a frequency-domain representation, and the main frequency component and secondary frequency component are separated to obtain the spectrum data. The spectral data is processed using a cactus shrinkage function. The main frequency features are preserved and the secondary frequency information is compressed by S(f) = f × [1 - exp(-α|f| / fmax)], resulting in compressed spectral features, where f is the frequency component, fmax is the maximum frequency, and α is the shrinkage parameter. Based on the compressed spectral characteristics, combined with historical data and recent trends, a segmented single-exponential time complexity calculation method is used for prediction modeling to obtain load demand prediction results and electricity price fluctuation prediction results.

5. The method according to claim 1, characterized in that, A multi-agent collaborative optimization model is established with the joint optimization objectives of minimizing the comprehensive energy cost for users and maximizing the local photovoltaic grid integration rate. This yields a dual-objective optimization function, including: Based on the photovoltaic power forecast results, load demand forecast results, and electricity price fluctuation forecast results, a comprehensive energy cost objective function for users is constructed, including grid electricity purchase cost, equipment maintenance cost, and carbon emission cost, thus obtaining the cost objective function. Based on the ratio of actual photovoltaic power generation to theoretically available photovoltaic power generation, a photovoltaic grid connection rate objective function is constructed, and the grid connection rate objective function is obtained. Based on the aforementioned cost objective function and absorption rate objective function, a fuzzy logic controller is used to dynamically adjust the weight coefficients of the two objectives according to the system operating status and user preferences, thereby obtaining a dual-objective optimization function.

6. The method according to claim 1, characterized in that, A single exponential time FPT algorithm based on the cactus contraction principle is used for distributed iterative optimization. Problem decomposition is achieved through core parameter identification and decision space compression, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources, including: The bi-objective optimization function is represented as a parameterized problem, and the set of key parameters that have the greatest impact on the objective function is identified through sensitivity analysis, thus obtaining the core parameter set; The decision space is nonlinearly transformed using a cactus shrinkage function. The original resolution is maintained in the core parameter direction, while compression is performed in the non-core parameter direction to obtain the compressed decision space. Based on the compressed decision space, the global optimization problem is decomposed into photovoltaic control sub-problems, energy storage control sub-problems, and load control sub-problems according to physical properties, resulting in a combination of independently solvable sub-problems; The ADMM algorithm is used to solve the combination of subproblems in a distributed iterative manner. By alternately updating the original and dual variables, the coordinated optimization of each subproblem is achieved, and a local optimal solution set is obtained. Based on the aforementioned local optimal solution set, the global optimal solution is gradually approximated through information exchange between coordination variables and Lagrange multipliers, resulting in a collaborative optimization strategy for photovoltaic power generation units, energy storage units, and adjustable load resources.

7. The method according to claim 1, characterized in that, The system generates multi-timescale control commands, including day-ahead planning, hourly scheduling, and minute-level real-time control. These commands are then transmitted in real-time to photovoltaic inverters, energy storage devices, and smart loads via a communication network, enabling dynamic matching and coordinated control of source, storage, and load resources. This includes: Based on the aforementioned collaborative optimization strategy, and using 24-hour photovoltaic output forecast, load demand forecast, and electricity price forecast data, an energy storage charging and discharging plan and a day-ahead optimization plan for adjustable load operating periods are generated, thereby obtaining day-ahead control commands. Based on real-time updated forecast data and system status, rolling optimization is performed every hour to update the equipment operation plan for the next 4 to 6 hours and obtain hourly scheduling instructions; Based on the actual system status and short-term prediction deviations, tracking control, correction control and emergency control strategies are adopted to generate minute-level real-time control commands for photovoltaic inverter power regulation, energy storage charging and discharging power and load response intensity, thus obtaining minute-level control commands; The daytime control commands, hourly scheduling commands, and minute-level control commands are converted into protocols using Modbus, SunSpec, or BACnet protocol conversion technologies to obtain standardized control commands. The standardized control commands are sent to photovoltaic inverters, energy storage devices and smart loads in real time through the communication network, so as to realize dynamic matching and coordinated control of source, storage and load resources.

8. The method according to claim 2, characterized in that, Calculate the amplitude, period, and energy characteristic parameters of each wave unit to obtain the wave characteristic parameter set, including: Based on each sub-segment of the preprocessed dataset, the wave amplitude A is calculated. wave = max(x i ) - min(x i ), where x i For the data points within the sub-segment, the amplitude characteristics are obtained; For the amplitude characteristics, the dominant period T of each segment is determined by autocorrelation function analysis. wave Calculate the periodic characteristic parameters to obtain the periodic characteristics; Based on the amplitude and period characteristics, the energy characteristic E is calculated. wave = ∑(x i ²) / N, where N is the segment length, yields the energy characteristics; Based on the amplitude characteristics, periodic characteristics, and energy characteristics, a set of wave feature parameters containing time series, periodic, and energy information is constructed.

9. The method according to claim 5, characterized in that, A fuzzy logic controller is used to dynamically adjust the weight coefficients of the two objectives based on the system operating state and user preferences, resulting in a bi-objective optimization function, including: The ratio of the predicted photovoltaic output to the actual electricity demand, the current state of charge of the battery, the grid electricity price level, and the user's set preference coefficient are collected to construct the state vector for weight adjustment and obtain the input parameters for weight adjustment. Based on the weighted input parameters, a fuzzy logic rule base is established. When photovoltaic power generation is much greater than the load demand and the grid electricity price is low, the weight of photovoltaic absorption rate is increased. When photovoltaic power generation is insufficient and the electricity price is high, the weight of cost is increased, thus obtaining fuzzy logic rules. According to the fuzzy logic rules, the weight adjustment amount Δw1 is calculated, and the weight coefficients are updated by w1(t+1) = w1(t) + Δw1 and w2(t+1) = 1 - w1(t+1) to obtain the adaptive weight coefficients. Based on the adaptive weight coefficients, a bi-objective optimization function F = w1·C1 + w2·C2 is constructed, where C1 is the user's comprehensive energy cost and C2 is the negative value of the photovoltaic absorption rate, thus obtaining the bi-objective optimization function.

10. A distributed photovoltaic and energy storage load collaborative optimization operation system, characterized in that, include: The data acquisition module is used to acquire meteorological data, photovoltaic equipment parameters and historical power generation data, as well as users' historical electricity consumption data and electricity price information, and transmit the data to the predictive analysis module. The predictive analysis module receives data from the data acquisition module, uses an improved wave sorting algorithm and a deep learning hybrid model to predict photovoltaic output, and uses a single exponential time FPT algorithm to predict load demand and electricity price, providing the prediction results to the optimization decision module. The optimization decision module receives the prediction results from the prediction analysis module, establishes a dual-objective collaborative optimization model, uses a distributed iterative optimization algorithm based on the cactus contraction principle to solve the collaborative optimization strategy, and outputs the optimization strategy to the control execution module. The control execution module receives the optimization strategy from the optimization decision module, generates multi-timescale control commands, and sends them to photovoltaic inverters, energy storage devices, and smart loads through the communication network to achieve coordinated control.

Images