Microgrid regulation method based on data feature analysis, program product and device

Abstract

The application provides a micro-grid regulation method based on data feature analysis, a program product and equipment, and relates to the field of power supply and distribution systems. The micro-grid regulation method comprises: obtaining original time sequence data of electric energy of a micro-grid; using a discrete wavelet transform algorithm to decompose the original time sequence data of electric energy, to obtain low-frequency components of the original time sequence data of electric energy; performing clustering processing on the low-frequency components, to determine a data analysis model corresponding to a type to which the low-frequency components belong, as a target analysis model; using the target analysis model to predict the original time sequence data, to obtain electric energy prediction data of the micro-grid; generating a regulation scheme of the micro-grid according to the electric energy prediction data, and controlling the micro-grid to execute the regulation scheme. The scheme of the application, the analysis model conforms to the overall operation law of the micro-grid, the prediction accuracy is high, the electric energy utilization efficiency, the economy and the safety of the micro-grid can be improved, and the micro-grid can adapt to a complex energy environment.

Description

Technical Field

[0001] This invention relates to the field of power supply and distribution systems, and in particular to a microgrid control method, program product, and equipment based on data feature analysis. Background Technology

[0002] A microgrid is a self-sufficient small-scale power network composed of multiple energy generation systems. It uses renewable energy as its power source and works in conjunction with energy storage systems and electrical appliances to improve energy efficiency. With the rapid development of renewable energy, microgrids, as a flexible, efficient, and environmentally friendly power supply solution, are finding increasingly wider applications.

[0003] Microgrid systems achieve coordinated optimization of power generation, energy storage, and load through intelligent scheduling of renewable energy generation, energy storage battery charging and discharging, and load consumption. With the integration of digitalization and artificial intelligence technologies, microgrids are developing towards greater intelligence and autonomy. Microgrid control has become a cutting-edge area in the field of power supply and distribution control.

[0004] The optimal dispatch of microgrids heavily relies on accurate forecasting of renewable energy generation and load demand. Significant differences exist between different microgrids in terms of equipment types and operational objectives, resulting in substantial variations in their operational characteristics. This makes accurate forecasting of microgrids particularly challenging.

[0005] Currently, artificial intelligence technologies, such as LSTM (Long Short-Term Memory) networks, have been applied to microgrid forecasting. However, the data accumulation of a single microgrid is limited, especially in the early stages of microgrid operation, resulting in poor forecast accuracy. Furthermore, due to the differences between various microgrids as described above, models trained using general-purpose data do not accurately reflect the actual conditions of microgrids, leading to decreased forecast accuracy. Summary of the Invention

[0006] One object of the present invention is to provide a microgrid control method based on data feature analysis that at least solves any of the above-mentioned technical problems.

[0007] A further objective of this invention is to make the target analysis model conform to the operating characteristics of microgrids and improve the accuracy of power prediction data.

[0008] Another further objective of this invention is to achieve continuous optimization of microgrid operation objectives.

[0009] Specifically, this invention provides a microgrid control method based on data feature analysis. The method includes:

[0010] Acquire raw time-series power data of the microgrid;

[0011] The discrete wavelet transform algorithm is used to decompose the original time series data of electrical energy to obtain the low-frequency components of the original time series data of electrical energy.

[0012] Clustering is performed on low-frequency components to determine the data analysis model corresponding to the type of low-frequency components, which serves as the target analysis model.

[0013] The target analysis model is used to predict the original time series data to obtain the power prediction data of the microgrid;

[0014] A microgrid control scheme is generated based on power forecast data, and the microgrid is controlled to execute the control scheme.

[0015] Optionally, the steps for obtaining the raw time-series power data of the microgrid include:

[0016] Raw electrical energy data is obtained from smart meter terminals and / or power acquisition terminals in the microgrid;

[0017] The raw electrical energy data is acquired and / or data completion processed according to a preset sampling period;

[0018] The data after numerical acquisition and / or data completion are organized according to time scale to obtain the raw time series data of electrical energy.

[0019] Optionally, the steps of decomposing the raw time-series electrical energy data using the discrete wavelet transform algorithm include:

[0020] The original time-series data of electrical energy is convolved with a low-pass filter preset in the discrete wavelet transform algorithm;

[0021] The result of the convolution is downsampled to obtain a downsampled low-pass output.

[0022] The low-pass output after downsampling is iteratively subjected to low-pass filter convolution calculation and convolution result downsampling.

[0023] The low-pass output after downsampling, obtained after setting the number of iterations, is used as the low-frequency component of the original power timing data.

[0024] Optionally, the number of iterations can be set to 3.

[0025] The downsampling rate is 50%.

[0026] Optionally, the step of clustering low-frequency components includes:

[0027] Calculate the distance from low-frequency components to multiple preset cluster centers, with each cluster center's data cluster serving as a training sample for a data analysis model;

[0028] The data analysis model corresponding to the nearest cluster center is used as the target analysis model.

[0029] Optionally, after obtaining the power prediction data for the microgrid, the method further includes:

[0030] Obtain measured power data of the microgrid;

[0031] The measured power data is used to evaluate the predicted power data, and the raw time series power data is labeled according to the evaluation results.

[0032] The labeled data is used as the iterative training data for the target analysis model.

[0033] Optionally, the steps for evaluating the power forecast data using measured power data include:

[0034] Calculate the similarity between measured power energy data and predicted power energy data;

[0035] The similarity is compared with a preset threshold, and the comparison result is used as the evaluation result.

[0036] Optionally, the steps for using a target analysis model to predict the original time series data include:

[0037] The convolutional neural network module of the target analysis model is invoked to extract data features from the original time series data;

[0038] The long short-term memory network module of the target analysis model is invoked to predict data characteristics, thereby obtaining power prediction data for the microgrid.

[0039] Optionally, the steps of generating a microgrid control scheme based on power forecast data include:

[0040] Obtain the power regulation constraints and objective function of the microgrid;

[0041] Develop a microgrid control scheme under the constraint of power regulation to ensure the optimal objective function.

[0042] According to another aspect of the present invention, a computer program product is also provided, comprising a computer program that, when executed by a processor, implements the steps of any of the above-described microgrid control methods based on data feature analysis.

[0043] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of any of the above-described microgrid control methods based on data feature analysis.

[0044] According to another aspect of the present invention, a computer device is also provided, which includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the microgrid control method based on data feature analysis described above.

[0045] This invention presents a microgrid control method based on data feature analysis. It acquires raw time-series power data from the microgrid and uses discrete wavelet transform to obtain low-frequency components, thus eliminating high-frequency components that cause random fluctuations in the microgrid. These low-frequency components reflect the overall trend and periodicity of the power data (e.g., daily, weekly, monthly cycles). Clustering is performed on these low-frequency components to obtain data analysis models corresponding to their respective types. These models conform to the overall operating patterns of the microgrid, and their accuracy in predicting raw time-series data is high. The resulting microgrid control scheme improves power utilization efficiency, enhances the economy and security of microgrid operation, and enables the microgrid to adapt to complex energy environments.

[0046] Furthermore, the microgrid control method based on data feature analysis of the present invention, after controlling the microgrid to execute the control scheme, acquires measured data to evaluate the predicted data, and labels the original time series data as iterative training data according to the evaluation results, forming a feedback optimization mechanism that can continuously optimize the target analysis model, improve the accuracy of the model in predicting the power of the microgrid, and thus continuously improve the control effect of the microgrid.

[0047] Furthermore, the microgrid control method based on data feature analysis of the present invention can include a target analysis model that includes a convolutional neural network module and a long short-term memory network module. The convolutional neural network module is used to extract features from the original time series data, and the long short-term memory network module is used to predict the features. This fully leverages the advantages of the two network modules, enabling more effective mining of complex time series features from the original time series data, thereby further improving the accuracy of power prediction data.

[0048] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0049] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0050] Figure 1 This is a schematic diagram of device connections for a microgrid according to an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of a microgrid control method based on data feature analysis according to an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram illustrating the steps of decomposing raw time-series electrical energy data in a microgrid control method based on data feature analysis according to an embodiment of the present invention.

[0053] Figure 4 This is a schematic diagram illustrating the steps of clustering low-frequency components in a microgrid control method based on data feature analysis according to an embodiment of the present invention.

[0054] Figure 5 This is a schematic diagram illustrating the steps of data prediction by a target analysis model in a microgrid control method based on data feature analysis according to an embodiment of the present invention.

[0055] Figure 6 This is a schematic diagram illustrating the steps of generating a microgrid control scheme in a microgrid control method based on data feature analysis according to an embodiment of the present invention;

[0056] Figure 7 This is a schematic diagram illustrating the steps of iteratively training a data analysis model in a microgrid control method based on data feature analysis according to an embodiment of the present invention;

[0057] Figure 8 This is a schematic diagram of a computer program product according to an embodiment of the present invention;

[0058] Figure 9 This is a schematic diagram of a computer-readable storage medium according to an embodiment of the present invention;

[0059] Figure 10 This is a schematic block diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0060] Those skilled in the art should understand that the embodiments described below are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. These partial embodiments are intended to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention. Based on the embodiments provided by the present invention, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of the present invention.

[0061] Figure 1This is a schematic diagram of the device connections of a microgrid according to an embodiment of the present invention. A microgrid generally includes: a smart controller 101, a cloud management device 102, a photovoltaic power generation system 11, an energy storage subsystem 12, electrical loads 13, and a power grid subsystem 14. The device types in the microgrid are merely illustrative; those skilled in the art can configure the necessary equipment according to requirements and power planning, such as adding wind power generation equipment, backup generators, and various environmental monitoring equipment (e.g., solar irradiance monitoring equipment, wind power monitoring equipment, etc.).

[0062] The photovoltaic power generation subsystem 11 includes a grid-connected inverter 111 and photovoltaic modules 112; the energy storage subsystem 12 includes an energy storage converter 121 and an energy storage battery 122; the electrical load 13 may include various electrical devices (such as DC charging piles, AC charging piles, lighting equipment, control power, heat pump equipment such as air conditioning and heating, etc.); the power grid subsystem 14 includes power grid interaction equipment 142 (such as transformers and distribution boxes, etc.) and various equipment of the external public power grid 21. In addition, the microgrid may also be equipped with electrical signal detection equipment, electricity meters, various sensors, security equipment, etc., used in conjunction with the above-mentioned equipment.

[0063] The intelligent controller 101 executes localized control strategies, processes data locally, and interacts with the cloud management device 102, uploading data to and receiving scheduling instructions from the cloud management device 102. Based on data feature analysis, the intelligent controller 101 automatically generates control schemes and accordingly regulates the controlled equipment (e.g., grid-connected inverter 111, energy storage converter 121, electrical equipment, and grid interaction equipment 142) within its control range. During data processing, the intelligent controller 101 can obtain the computing power and data resources required for model training and data analysis by interacting with the cloud management device 102.

[0064] The photovoltaic power generation system 11 uses photovoltaic panels to convert solar energy into electrical energy in a controlled manner, and performs preliminary processing on the direct current generated by the photovoltaic panels. For example, the grid-connected inverter 111 uses maximum power point tracking (MPPT) technology to ensure that the photovoltaic module 112 always outputs electrical energy at maximum power, thereby improving the utilization efficiency of solar energy.

[0065] The energy storage subsystem 12 utilizes the energy storage battery 122 in a controlled manner to store and release electrical energy, addressing the mismatch between electricity demand and the power generation period of the photovoltaic power generation system 11. The energy storage converter 121 (Power Conversion System, PCS) is used to convert the electrical energy from the energy storage battery 122. During peak electricity demand periods, the energy storage subsystem 12 releases the stored electrical energy to alleviate the pressure on the power grid; during off-peak electricity demand periods, it absorbs electrical energy from the grid for storage, reducing electricity costs and achieving optimal allocation of power resources.

[0066] The power grid subsystem 14 enables controlled power exchange between the microgrid and the external public power grid 21. The power grid subsystem 14 can control the interaction mode (grid-connected operation mode / off-grid operation mode / anti-reverse operation mode, etc.) between the microgrid and the public power grid 21 according to the dispatch instructions and operating constraints of the public power grid 21.

[0067] The electrical load 13 can controllably adjust its power consumption status, such as reducing load power, increasing load power, or shutting down the load.

[0068] This embodiment provides a microgrid control method based on data feature analysis. By analyzing the data of the microgrid, a control scheme is generated to improve the power utilization efficiency of the microgrid and optimize the operation target. Figure 2 This is a schematic diagram of a microgrid control method based on data feature analysis according to an embodiment of the present invention. The microgrid control method based on data feature analysis in this embodiment generally includes:

[0069] Step S201: Obtain the raw time-series power data of the microgrid. The raw time-series power data may include: the total power generation and consumption within the microgrid, the power / energy exchanged between the microgrid and the public grid, and the electricity cost of the microgrid, etc.

[0070] One optional process for obtaining raw time-series power data from a microgrid includes: acquiring raw power data from smart meter terminals and / or power acquisition terminals in the microgrid; performing numerical acquisition processing and / or data completion processing on the raw power data according to a preset sampling period; and organizing the data after numerical acquisition processing and / or data completion processing according to a time scale to obtain the raw time-series power data. Smart meter terminals may include smart meters installed at the connection point between the microgrid and the public power grid, or smart meters installed on various lines within the microgrid. Power acquisition terminals can be various power management terminals, such as sensors or measuring devices with power metering or electrical signal measurement functions.

[0071] The raw time-series data of electrical energy is a time series organized according to the time of the acquisition cycle. The specific time interval can be set as needed, such as 15 minutes, 30 minutes, 1 hour, etc.

[0072] Data completion processing can include missing value handling and outlier handling. Missing value handling addresses situations with a small number of missing values ​​(such as missing data points at certain times), and can employ linear interpolation, nearest neighbor imputation, or imputation based on the mean of historical data from the same period. For example, using average smoothing, the average of several previous and subsequent sampling points can be taken as the missing value. Outlier handling identifies numerical anomalies, such as sudden increases in load by a preset multiple or charging / discharging power significantly exceeding limits. Outliers can be discarded and imputed using methods similar to those used for missing values.

[0073] The aforementioned raw power time-series data can be normalized to the [0,1] interval using a min-max method to avoid interference from magnitude differences in model training. This involves treating the maximum power value as 1 and the minimum as 0, then converting the data between the minimum and maximum values ​​to the [0,1] interval. Alternatively, standardization can be used, subtracting the mean from the value and dividing by the standard deviation. Standardization or normalization of the raw power time-series data eliminates the influence of dimensions, preventing numerical differences from interfering with subsequent discrete wavelet transform algorithms.

[0074] Step S202: The discrete wavelet transform algorithm is used to decompose the original time-series power data to obtain the low-frequency components of the original time-series power data. The low-frequency components reflect the overall trend and periodicity of the power data (such as daily, weekly, and monthly cycles).

[0075] Step S203: Cluster the low-frequency components to determine the data analysis model corresponding to the type of the low-frequency components, which will be used as the target analysis model.

[0076] In this embodiment, multiple data analysis models can be pre-established, and each data analysis model is trained from microgrid data with the same or similar overall characteristics.

[0077] Step S204: The target analysis model is used to predict the original time-series data to obtain the power prediction data of the microgrid. The target analysis model may include a convolutional neural network module and a long short-term memory network module. The convolutional neural network module is used to extract features from the original time-series data, and then the long short-term memory network module is used to predict the features. This fully leverages the advantages of both network modules and can more effectively extract complex time-series features from the original time-series data.

[0078] Step S205: Generate a microgrid control scheme based on power forecast data, and control the microgrid to execute the control scheme. The control scheme specifies the adjustment actions of energy storage devices and the adjustment methods of adjustable loads under different operating conditions.

[0079] This microgrid control method based on data feature analysis acquires raw time-series power data containing various power information during microgrid operation. However, this data contains high-frequency components with random fluctuations, which can interfere with data analysis and processing. To extract more valuable information, a discrete wavelet transform algorithm is used to decompose the raw time-series power data, thereby obtaining low-frequency components and successfully eliminating the interference from the aforementioned high-frequency components.

[0080] Low-frequency components contain the overall trend of power data changes and significant periodic patterns, such as daily, weekly, and monthly periods. These patterns reflect the relatively stable characteristics of microgrid operation and can be used to more accurately analyze and understand the operational characteristics of microgrids. Based on the characteristics exhibited by low-frequency components, the method in this embodiment performs clustering processing to accurately determine the data analysis model corresponding to the type of low-frequency components. Because this analysis model closely matches the overall operational patterns of the microgrid reflected by the low-frequency components, it demonstrates high accuracy when used to predict raw time-series data.

[0081] By generating microgrid control schemes based on the prediction results of the target analysis model, accurate analysis of data characteristics enables systematic and comprehensive control of the microgrid, making the control of the microgrid more suitable for its complex energy environment and improving the stability and efficiency of microgrid operation.

[0082] Figure 3 This is a schematic diagram illustrating the steps of decomposing raw electrical energy time-series data in a microgrid control method based on data feature analysis according to an embodiment of the present invention. The steps described above for decomposing raw electrical energy time-series data using the discrete wavelet transform algorithm may include:

[0083] Step S301: Convolve the original time-series data of electrical energy with the low-pass filter preset in the discrete wavelet transform algorithm.

[0084] Step S302: Downsample the result of the convolution to obtain the downsampled low-pass output.

[0085] Step S303: Iterate the low-pass output after downsampling by performing convolution calculation of the low-pass filter and downsampling of the convolution result.

[0086] Step S304: The downsampled low-pass output obtained after a set number of iterations is used as the low-frequency component of the original power timing data. In some embodiments, the set number of iterations can be 3; the downsampling sampling rate is 50%.

[0087] Discrete Wavelet Transformation (DWT) decomposes raw electricity time-series data into detail coefficients and approximation coefficients, achieving separation of data individuality and commonality, and high sensitivity and low sensitivity. Abrupt changes or rapid shifts in raw electricity time-series data are often associated with specific electricity consumption events (such as random equipment startup or shutdown, abnormal malfunctions, or temporary activities leading to increased electricity consumption), and are more susceptible to random factors. Detail coefficients represent the high-frequency components of the data, reflecting these abrupt changes and randomness. Approximation coefficients represent the low-frequency components of the data, reflecting the overall trend and periodicity of the data; this part has commonalities and lower randomness.

[0088] The discrete wavelet decomposition process is shown in the following equation: In the formula, the high-frequency component P(t) is represented by detail coefficients D1, D2, and D3, and the low-frequency component C(t) is represented by approximation coefficient A3. In this embodiment, the number of decomposition layers is selected as 3, that is, the number of iterations is 3. This achieves a good balance between capturing global trends and detailed changes. The approximation coefficient A3 represents the overall trend of the power data, that is, the long-term trend and periodic changes driven by macroscopic factors (such as work-rest changes, diurnal variations, etc.), which have common characteristics. The detail coefficients D1, D2, and D3 describe the fluctuation characteristics of the data from different resolutions, reflecting the impact of random behavior and temporary events on the power data.

[0089] The number of decomposition layers determines the number of scales of data that can be separated. Too few layers may fail to separate key periodic components, while too many layers may cause the coarsest approximation coefficients to lose useful information.

[0090] The three-step iterative decomposition process can include: performing an initial convolution calculation between the raw time-series power data and a pre-defined low-pass filter in the discrete wavelet transform algorithm; downsampling the result of the initial convolution to obtain the downsampled low-pass output, i.e., the first-layer approximation coefficients A1; performing a second convolution calculation between the first-layer approximation coefficients A1 and the pre-defined low-pass filter in the discrete wavelet transform algorithm; downsampling the result of the second convolution to obtain the downsampled low-pass output, i.e., the second-layer approximation coefficients A2; and performing a third convolution calculation between the second-layer approximation coefficients A2 and the pre-defined low-pass filter in the discrete wavelet transform algorithm; downsampling the result of the third convolution to obtain the downsampled low-pass output, i.e., the third-layer approximation coefficients A3. Each decomposition only further breaks down the low-frequency components; the high-frequency components (detail coefficients) are not decomposed, ultimately resulting in one set of low-frequency approximation coefficients A3 and three sets of high-frequency detail coefficients D1, D2, and D3.

[0091] Figure 4This is a schematic diagram illustrating the steps of clustering low-frequency components in a microgrid control method based on data feature analysis according to an embodiment of the present invention. The steps of clustering low-frequency components described above may include:

[0092] Step S401: Calculate the distance from the low-frequency component to multiple preset cluster centers. The data cluster of each cluster center is a training sample for a data analysis model.

[0093] Multiple cluster centers can be obtained from the low-frequency components of pre-collected electricity consumption data using clustering estimation (e.g., the K-Means clustering algorithm). Each cluster center represents a representative location in the data space, corresponding to a typical overall electricity consumption trend. The original training data is divided into multiple data clusters based on the clustering results. Each data cluster contains samples that are closest to the same cluster center. Each data cluster is trained independently with a data analysis model. Finally, a mapping relationship is established between the cluster centers and the corresponding data analysis models.

[0094] Step S402: The data analysis model corresponding to the nearest cluster center is taken as the target analysis model.

[0095] For each newly input low-frequency component sample, calculate its distance to all preset cluster centers (distances can be Euclidean distance, Manhattan distance, etc.). By comparing the distances to different cluster centers, the nearest cluster center can be found. Through the above mapping relationship, the corresponding target analysis model can be determined.

[0096] The data analysis model can support hot updates. The raw time-series data of electricity, which is the object of analysis, can be labeled and used as iterative training data for the target analysis model.

[0097] Figure 5 This is a schematic diagram illustrating the steps of data prediction using a target analysis model in a microgrid control method based on data feature analysis according to an embodiment of the present invention. The steps described above for predicting raw time-series data using a target analysis model may include:

[0098] Step S501: The convolutional neural network module of the target analysis model is invoked to extract data features from the original time-series data. The input layer of the convolutional neural network (CNN) module receives input data from the dataset, and the convolutional layer completes feature extraction by setting the feature detector size and convolutional kernel size. The pooling layer compresses the amount of data in the convolutional layer, thereby reducing overfitting. The pooled data is then used as input to the Long Short-Term Memory Network module for processing the time series data.

[0099] Step S502: The Long Short-Term Memory (LSTM) network module of the target analysis model is invoked to predict the data features, obtaining the microgrid's power prediction data. LSTM addresses the gradient vanishing and gradient exploding problems of Recurrent Neural Networks (RNNs) by storing useful information and eliminating unnecessary information through storage units. LSTM controls the flow of data through forget gates, input gates, and output gates. By learning long-term dependencies (such as periodicity and trends) in the feature sequence, LSTM ultimately outputs the microgrid's power prediction data.

[0100] Figure 6 This is a schematic diagram illustrating the steps of generating a microgrid control scheme in a microgrid control method based on data feature analysis according to an embodiment of the present invention. The step of generating a microgrid control scheme based on power forecast data may include:

[0101] Step S601: Obtain the power regulation constraints and objective function of the microgrid. The regulation objectives of the microgrid may include economy, stability, and security. The objective function can be a single-objective optimization function for the above regulation objectives, or a multi-objective optimization function integrating multiple regulation objectives. Power regulation constraints may include: power supply and demand balance conditions (e.g., the total power generation of all energy generation systems within the microgrid must be equal to or greater than the sum of the microgrid's power load plus line losses and other power consumption, and the balance requirements for energy storage charging and discharging), power generation equipment operation constraints (e.g., minimum and maximum power limits for power generation equipment, and limits on the rate of change of power generation), power demand constraints (e.g., the size of different levels of power load, and the adjustable range of load), electrical parameter constraints (e.g., power quality requirements, voltage deviation limits, and frequency deviation limits), economic cost constraints (e.g., power generation costs and electricity purchase costs), and security constraints (e.g., reserve capacity reservation limits, failure rate limits, and maintenance time).

[0102] Step S602: Develop a microgrid control scheme under power regulation constraints to ensure the objective function is optimal. Optimization is performed under these constraints using Particle Swarm Optimization (PSO) or Genetic Algorithm (GA). An optional implementation process is as follows: Define the objective function (e.g., minimize operating cost); list all constraints (power balance, equipment output, SOC, etc.); determine decision variables (e.g., output of each generator, energy storage charging and discharging, purchased electricity, adjustable load, etc.). Select a suitable optimization algorithm (e.g., PSO) and iteratively optimize under the constraints to obtain the optimal values ​​of the decision variables. If the optimal value satisfies all constraints, a control scheme is generated based on the optimal value; otherwise, the algorithm parameters need to be adjusted.

[0103] Figure 7This is a schematic diagram illustrating the steps of iteratively training a data analysis model in a microgrid control method based on data feature analysis according to an embodiment of the present invention. After the above-described step of obtaining the microgrid's power prediction data, the method may further include:

[0104] Step S701: Obtain the measured power data of the microgrid.

[0105] Step S702: Use measured power data to evaluate the predicted power data, and label the original time series data of power based on the evaluation results.

[0106] One possible evaluation method is to calculate the similarity between measured power data and predicted power data; compare the similarity with a preset threshold, and use the comparison result as the evaluation result. If the similarity is less than the threshold, it is considered that there is a large deviation between the predicted power data and the measured power data, and it is marked as a prediction error; while if the similarity exceeds the threshold, it is considered that the predicted power data and the measured power data are relatively close, and it is marked as a prediction accuracy.

[0107] Step S703 involves using the labeled data as iterative training data for the target analysis model. Through iterative training, the target analysis model can be continuously optimized, improving the accuracy of the model's power prediction for the microgrid and thus continuously improving the microgrid's control effect.

[0108] This embodiment also provides a computer program product 810, a computer-readable storage medium 820, and a computer device 830. Figure 8 This is a schematic diagram of a computer program product 810 according to an embodiment of the present invention. Figure 9 This is a schematic diagram of a computer-readable storage medium 820 according to an embodiment of the present invention. Figure 10 This is a schematic block diagram of a computer device 830 according to an embodiment of the present invention.

[0109] Computer program product 810 includes computer program 811, which, when executed by processor 831, implements the steps of the microgrid control method based on data feature analysis described above. Computer-readable storage medium 820 stores the aforementioned computer program 811, which, when executed by processor 831, implements the steps of the microgrid control method based on data feature analysis described above. Computer device 830 may include memory 832, processor 831, and computer program 811 stored in memory 832 and running on processor 831.

[0110] The computer program 811 used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, status setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​and procedural programming languages.

[0111] Computer program 811 may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, to perform aspects of the invention, electronic circuits including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may execute computer-readable program instructions to personalize the electronic circuits by utilizing state information of computer-readable program instructions.

[0112] For the purposes of this embodiment, computer program product 810 is a related product that includes computer program 811.

[0113] For the purposes of this embodiment, a computer-readable storage medium 820 is a tangible device capable of holding and storing a computer program 811. It can be any device that includes, stores, communicates, propagates, or transmits the computer program 811 for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable storage medium 820 include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital versatile disc (DVD), memory stick, floppy disk, mechanical encoding device, and any suitable combination thereof.

[0114] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A microgrid control method based on data feature analysis, characterized in that, include: Obtain the raw time-series power data of the microgrid; The original time-series data of the power energy was decomposed using the discrete wavelet transform algorithm to obtain the low-frequency components of the original time-series data of the power energy. The low-frequency components are clustered to determine the data analysis model corresponding to the type of the low-frequency components, which is then used as the target analysis model. The target analysis model is used to predict the original time series data to obtain the power prediction data of the microgrid; A control scheme for the microgrid is generated based on the power forecast data, and the microgrid is controlled to execute the control scheme. The step of clustering the low-frequency components includes: Calculate the distance from the low-frequency component to multiple preset cluster centers, where each cluster center's data cluster is a training sample for the data analysis model; The data analysis model corresponding to the nearest cluster center is used as the target analysis model.

2. The microgrid control method based on data feature analysis according to claim 1, characterized in that, The step of acquiring the raw time-series power data of the microgrid includes: Raw electrical energy data is obtained from the smart meter terminals and / or power acquisition terminals of the microgrid; The raw electrical energy data is subjected to numerical acquisition processing and / or data completion processing according to a preset sampling period; The data obtained after numerical acquisition and / or data completion processing are organized according to time stamps to obtain the raw time-series data of electrical energy.

3. The microgrid control method based on data feature analysis according to claim 1, characterized in that, The step of decomposing the raw time-series electrical energy data using the discrete wavelet transform algorithm includes: The original time-series data of electrical energy is convolved with the low-pass filter preset in the discrete wavelet transform algorithm; The result of the convolution is downsampled to obtain a downsampled low-pass output; The low-pass output after downsampling is iteratively subjected to convolution calculation of the low-pass filter and downsampling of the convolution result; The low-pass output obtained after setting the number of iterations is used as the low-frequency component of the original power timing data.

4. The microgrid control method based on data feature analysis according to claim 3, characterized in that, The set number of iterations is 3; The downsampling rate is 50%.

5. The microgrid control method based on data feature analysis according to claim 1, characterized in that, After obtaining the power prediction data for the microgrid, the process further includes: Obtain the measured power data of the microgrid; The measured power data is used to evaluate the predicted power data, and the original time-series power data is labeled according to the evaluation results. The labeled data is used as the iterative training data for the target analysis model.

6. The microgrid control method based on data feature analysis according to claim 5, characterized in that, The step of evaluating the predicted power data using the measured power data includes: Calculate the similarity between the measured power data and the predicted power data; The similarity is compared with a preset threshold, and the comparison result is used as the evaluation result.

7. The microgrid control method based on data feature analysis according to claim 1, characterized in that, The step of using the target analysis model to predict the original time series data includes: The convolutional neural network module of the target analysis model is invoked to extract the data features of the original time series data; The long short-term memory network module of the target analysis model is invoked to predict the data features, thereby obtaining the power prediction data of the microgrid.

8. The microgrid control method based on data feature analysis according to claim 1, characterized in that, The step of generating the microgrid control scheme based on the power prediction data includes: Obtain the power regulation constraints and objective function of the microgrid; Under the aforementioned power regulation constraints, a regulation scheme for the microgrid is formulated to ensure that the objective function is optimal.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the microgrid control method based on data feature analysis as described in any one of claims 1 to 8.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the microgrid control method based on data feature analysis as described in any one of claims 1 to 8.

Images