New energy power generation equipment adaptive control system based on multi-scale data fusion

Abstract

The invention discloses a new energy power generation equipment adaptive control system based on multi-scale data fusion, and relates to the technical field of power generation control, and the system comprises the steps: obtaining a multi-source power data set corresponding to each composite fuel type; performing multi-scale feature convolution on the multi-source power data set, and outputting a multi-source high-dimensional data set; fusing the multi-source high-dimensional data set to generate a fused high-dimensional data set; receiving equipment fault information, and generating equipment control parameter items; performing feature dimension reduction on the fused high-dimensional data set, and outputting a fused low-dimensional data set; and training the equipment adaptive control model, and performing adaptive control on the new energy power generation equipment. The technical problems that in the prior art, new energy power generation equipment is poor in adaptability and unstable in power output are solved, and the technical effects that the adaptability of the equipment is remarkably improved, and unstable power output is effectively reduced are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation control, and in particular to an adaptive control system for new energy power generation equipment based on multi-scale data fusion. Background Art

[0002] In the field of renewable energy power generation, with the rapid development of various renewable energy power generation technologies, new energy sources such as wind energy, solar energy, and biomass energy are gradually being widely used. However, since these renewable energy power generation equipment usually work under different environmental conditions and load conditions, their operating status has greater uncertainty and complexity, especially when the equipment fails or switches working modes, it is more likely to cause instability in power output. Traditional control systems often find it difficult to cope with such a changing operating environment, lack sufficient adaptability and accuracy, and affect the stability of the power system.

[0003] The existing technology has technical problems such as poor adaptability of new energy power generation equipment and unstable power output. Summary of the invention

[0004] The present application provides an adaptive control system for new energy power generation equipment based on multi-scale data fusion, which is used to solve the technical problems of poor adaptability and unstable power output of new energy power generation equipment in the prior art.

[0005] In view of the above problems, the present application provides an adaptive control system for new energy power generation equipment based on multi-scale data fusion, the system comprising: A multi-source power data set acquisition module is used to obtain new energy fuel information of new energy power generation equipment, and determine whether the new energy power generation equipment is a composite fuel power generation mode according to the new energy fuel information. If the new energy power generation equipment is a composite fuel power generation mode, a multi-source power data set corresponding to each composite fuel type is obtained; a multi-source high-dimensional data set output module is used to perform multi-scale feature convolution on the multi-source power data set through a multi-scale convolutional network layer, and output a multi-source high-dimensional data set, wherein the multi-scale feature convolution includes time feature convolution, space feature convolution and type feature convolution; a fused high-dimensional data set generation module is used to fuse the multi-source high-dimensional data sets to generate a fused high-dimensional data set; a device control parameter item generation module is used for the new energy power generation equipment to receive device fault information and generate device control parameter items according to the device fault information; a fused low-dimensional data set generation module is used to perform feature dimensionality reduction on the fused high-dimensional data set using the t-SNE algorithm, and output a fused low-dimensional data set; an adaptive control module is used to train an equipment adaptive control model with the fused low-dimensional data set and the equipment control parameter items, and adaptively control the new energy power generation equipment based on the equipment adaptive control model.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: A multi-source power data set acquisition module is used to obtain a multi-source power data set corresponding to each composite fuel type; a multi-source high-dimensional data set output module is used to perform multi-scale feature convolution on the multi-source power data set through a multi-scale convolution network layer, and output a multi-source high-dimensional data set; a fusion high-dimensional data set generation module is used to fuse the multi-source high-dimensional data set to generate a fusion high-dimensional data set; an equipment control parameter item generation module is used for the new energy power generation equipment to receive equipment fault information and generate equipment control parameter items; a fusion low-dimensional data set generation module is used to perform feature dimensionality reduction on the fusion high-dimensional data set using the t-SNE algorithm, and output a fusion low-dimensional data set; an adaptive control module is used to train the equipment adaptive control model with the fusion low-dimensional data set and the equipment control parameter items, and perform adaptive control. The technical effect of significantly improving the adaptability of the equipment and effectively reducing the instability of power output is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of the structure of an adaptive control system for new energy power generation equipment based on multi-scale data fusion provided in an embodiment of the present application; Figure 2 A schematic diagram of the process of outputting a fused low-dimensional data set by a multi-source high-dimensional data set output module of an adaptive control system for new energy power generation equipment based on multi-scale data fusion provided in an embodiment of the present application.

[0009] Explanation of the reference numerals: multi-source power data set acquisition module 10 , multi-source high-dimensional data set output module 20 , fused high-dimensional data set generation module 30 , equipment control parameter item generation module 40 , fused low-dimensional data set generation module 50 , adaptive control module 60 . DETAILED DESCRIPTION

[0010] The present application provides an adaptive control system for new energy power generation equipment based on multi-scale data fusion, which is used to solve the technical problems of poor adaptability and unstable power output of new energy power generation equipment in the prior art.

[0011] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0012] Examples, such as Figure 1 As shown, the present application provides an adaptive control system for new energy power generation equipment based on multi-scale data fusion, the system comprising: The multi-source power data set acquisition module 10 is used to obtain the new energy fuel information of the new energy power generation equipment, and determine whether the new energy power generation equipment is a composite fuel power generation mode based on the new energy fuel information. If the new energy power generation equipment is a composite fuel power generation mode, obtain the multi-source power data set corresponding to each composite fuel type.

[0013] Specifically, first, obtain the new energy fuel information of the new energy power generation equipment. New energy power generation equipment can use a variety of different types of new energy fuels, such as solar energy, wind energy, hydropower, biomass energy, etc., or a combination of these energy sources. Through various sensors and monitoring systems, detailed information about the fuel used by the new energy power generation equipment can be obtained. This information includes the type, source, current reserve, supply stability, etc. of the fuel. For example, if it is a solar power generation equipment, it will obtain information such as the current light intensity and the working status of the solar panel; if it is a wind power generation equipment, it will obtain information such as the current wind speed and wind direction; if it is a biomass power generation equipment, it will obtain information such as the supply of biomass fuel and combustion efficiency.

[0014] Based on the acquired new energy fuel information, it is determined whether the new energy power generation equipment is in the composite fuel power generation mode. The composite fuel power generation mode means that the equipment uses two or more different types of new energy fuels to generate electricity at the same time. For example, a power generation equipment uses solar energy and wind power to generate electricity at the same time, or combines biomass energy with hydropower. By analyzing the fuel information, it can be determined whether the equipment is in this composite fuel power generation mode.

[0015] If the new energy power generation equipment is judged as a composite fuel power generation mode, then it is necessary to obtain the multi-source power data set corresponding to each composite fuel type, because different fuel types will have different effects on the operation of the equipment, so it is necessary to obtain data related to each fuel type separately. These multi-source power data sets may include electrical parameters, environmental data, equipment status information, etc. related to a specific fuel type. For example, if the equipment uses solar energy and wind power generation at the same time, then it is necessary to obtain multi-source power data sets for the solar power generation part and the wind power generation part respectively. These data sets will provide an important basis for subsequent analysis and control, so as to better manage and optimize the operation of composite fuel power generation equipment.

[0016] By obtaining the fuel information of new energy power generation equipment, it is determined whether it is a composite fuel power generation mode and obtaining the corresponding multi-source power data set, providing a data basis for optimizing the operation and management of composite fuel power generation equipment and achieving more efficient and stable power generation control.

[0017] The multi-source high-dimensional data set output module 20 is used to perform multi-scale feature convolution on the multi-source power data set through a multi-scale convolutional network layer, and output a multi-source high-dimensional data set, wherein the multi-scale feature convolution includes time feature convolution, space feature convolution and type feature convolution.

[0018] Specifically, first of all, the multi-scale convolutional network layer is a neural network structure that can process features of different scales. It can simultaneously analyze multiple dimensions such as time, space, and data type, thereby extracting richer and more comprehensive feature information.

[0019] Time feature convolution mainly operates on time series data in multi-source power data sets. Time series data includes changes in the output power of the equipment over time, fluctuations in voltage and current, etc. Through time feature convolution, periodic features and trend features are extracted. For example, it is found that the output power of the equipment shows a periodic change pattern within a day, or shows an upward or downward trend over a period of time. These time features are very important for predicting the future state of the equipment and performing adaptive control.

[0020] Spatial feature convolution focuses on spatial data in multi-source power data sets. In a renewable energy power plant, different devices are distributed in different locations and they interact with each other. Spatial feature convolution can capture this spatial relationship, such as power transmission between different devices, interaction between electric and magnetic fields, etc. In addition, spatial feature convolution can also consider differences in the operating status of devices in different geographical locations, such as changes in environmental factors such as light intensity and wind speed.

[0021] Type feature convolution performs convolution operations on different types of data in multi-source power data sets. Multi-source power data sets for new energy power generation equipment contain multiple types of data, such as electrical parameter data, environmental factor data, equipment status data, etc. Type feature convolution can extract correlation features between different types of data. For example, it may be possible to find the relationship between light intensity and the output power of solar power generation equipment, or the relationship between wind speed and the output power of wind power generation equipment. These correlation features can help better understand the operating mechanism and performance characteristics of the equipment.

[0022] By performing time feature convolution, spatial feature convolution, and type feature convolution on the multi-source power data set, a multi-source high-dimensional data set can be obtained. This data set contains richer and more abstract feature information, providing a stronger foundation for subsequent data analysis and adaptive control.

[0023] The fused high-dimensional data set generation module 30 is used to fuse the multi-source high-dimensional data sets to generate a fused high-dimensional data set.

[0024] Specifically, the purpose of fusion is to integrate multi-source high-dimensional data obtained from feature convolutions of different scales. These data extract feature information of different dimensions through temporal feature convolution, spatial feature convolution, and type feature convolution. The fusion process is to integrate these scattered information together to form a more comprehensive and representative data set.

[0025] In order to achieve fusion, based on the weighted fusion method, a weight is assigned to each source of high-dimensional features according to the importance of different features, and then the weighted features are summed. This can highlight the contribution of important features while reducing the impact of unimportant features. In the fusion process, the consistency and compatibility of the data need to be considered. Since different scales of feature convolution use different processing systems and parameter settings, the obtained high-dimensional data are different. Before fusion, these data need to be preprocessed, such as normalization, standardization and other operations to ensure that the data are in the same scale and range. After the fusion operation, the generated fused high-dimensional data set contains comprehensive feature information in multiple dimensions such as time, space and type. This data set can better reflect the operating status and characteristics of new energy power generation equipment, and provide stronger support for subsequent equipment control and optimization. Fusion of multi-source high-dimensional data sets generates fused high-dimensional data sets, integrates feature information of different dimensions, improves the ability to reflect the operating status and characteristics of new energy power generation equipment, and provides stronger support for equipment control and optimization.

[0026] The device control parameter item generating module 40 is used for the new energy power generation device to receive device fault information and generate device control parameter items according to the device fault information.

[0027] Specifically, when a fault occurs during the operation of new energy power generation equipment, the fault detection system will detect the fault in time and generate corresponding fault information, including fault type, fault location, fault severity, etc. For example, it may be damage to a key component, circuit short circuit, sensor failure, etc.

[0028] After receiving these fault information, the new energy power generation equipment will start the fault response mechanism. First, the control system of the equipment will analyze and process the fault information, and adopt different response strategies according to the different fault types and severity. For example, if it is a minor fault, the equipment can continue to operate by adjusting the operating parameters, and arrange maintenance personnel to inspect and repair it; if it is a serious fault, the equipment needs to be stopped immediately to prevent further damage.

[0029] In order to generate equipment control parameter items, the control system will combine the fault information and the current operating status of the equipment, as well as the pre-set fault handling strategy for calculation. For example, if the fault is caused by excessive output power, the control system will generate a control parameter item to reduce the output power; if a sensor fails, the control algorithm that relies on the sensor will be adjusted, or a backup sensor will be enabled. The generated equipment control parameter items will be sent to the equipment's actuators, such as motors, valves, switches, etc., to adjust the equipment's operating status. At the same time, these parameter items will also be recorded for subsequent fault analysis and processing. In this way, effective control measures can be taken in a timely manner when the equipment fails, reducing the impact of the fault on the equipment operation and improving the reliability and stability of the equipment.

[0030] The fused low-dimensional data set generating module 50 is used to perform feature dimension reduction on the fused high-dimensional data set by using the t-SNE algorithm, and output a fused low-dimensional data set.

[0031] Specifically, the t-SNE algorithm is used to perform feature dimensionality reduction on the fused high-dimensional data set to output a fused low-dimensional data set. First, the fused high-dimensional data set usually contains a large amount of feature information, but high-dimensional data will face problems such as high computational complexity and data sparsity when processing and analyzing. Therefore, feature dimensionality reduction is needed to simplify the data while retaining important feature information. The t-SNE algorithm is an effective nonlinear dimensionality reduction system. Its main idea is to construct the probability distribution between data points in high-dimensional space and low-dimensional space respectively, and then find the representation in low-dimensional space by minimizing the difference between the two probability distributions. The working process of the t-SNE algorithm is as follows: In the high-dimensional space, Gaussian distribution calculation is performed on each data point in the fused high-dimensional data set, which means that for each data point, the distance between it and other data points is calculated, and a probability value is determined based on the distance to indicate the similarity between the point and other points, so that a high-dimensional probability distribution matrix is ​​obtained. In the low-dimensional space, t distribution calculation is performed on each data point in the fused low-dimensional data set. Similarly, the distance between each point and other points is calculated, and the probability value is determined to obtain a low-dimensional probability distribution matrix. Then, the KL divergence of the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix is ​​calculated. The KL divergence indicates the degree of difference between the two probability distributions. The objective function is constructed with the KL divergence. The goal is to minimize this difference so that the distribution of data points in the low-dimensional space is as close as possible to the distribution of data points in the high-dimensional space. In order to minimize the objective function, the gradient descent method is used for optimization. By continuously adjusting the position of the data points in the low-dimensional space, the objective function is gradually reduced. In each iteration, the moving direction and distance of each data point are calculated according to the gradient of the objective function. As the gradient descent proceeds, the distribution of data points in the low-dimensional space is gradually adjusted until the objective function converges. When the objective function converges, a fused low-dimensional data set is obtained. This low-dimensional data set retains the important feature information in the fused high-dimensional data set while greatly reducing the dimension of the data. This makes it easier to perform subsequent data analysis and processing, such as visualization, clustering, classification, etc. At the same time, it also reduces the consumption of computing resources and improves the efficiency of the algorithm.

[0032] The adaptive control module 60 is used to train the device adaptive control model with the fused low-dimensional data set and the device control parameter items, and to perform adaptive control on the new energy power generation equipment based on the device adaptive control model.

[0033] Specifically, the support vector machine algorithm is used to realize the training and application of the equipment adaptive control model. For the processing of the fused low-dimensional data set and the equipment control parameter items, the fused low-dimensional data set is used as the input feature vector, which contains the key information of the new energy power generation equipment in different aspects. The data after dimensionality reduction is more refined and can highlight the important features. The equipment control parameter items can be used as an additional feature or part of the label to guide the training of the model. In the training stage, the support vector machine algorithm aims to find an optimal hyperplane to separate the data points of different categories as much as possible while maximizing the classification interval. For the adaptive control problem of new energy power generation equipment, different equipment operating states or control strategies are regarded as different categories. By learning a large number of fused low-dimensional data sets and corresponding equipment control parameter items, the SVM model determines the optimal classification hyperplane, that is, determines the optimal control strategy of the new energy power generation equipment under different circumstances. The SVM algorithm determines the position and direction of the hyperplane by finding support vectors, that is, those data points closest to the hyperplane. These support vectors are crucial to the performance of the model and determine the boundaries of the classification. In the training process, the SVM algorithm determines the optimal hyperplane parameters by optimizing an objective function, maximizing the classification interval and minimizing the classification error. Once the training is completed, when a new fused low-dimensional data set is input, the SVM model quickly performs classification prediction and determines the corresponding equipment control strategy. For the adaptive control of new energy power generation equipment, the most appropriate control parameters are automatically selected based on the real-time equipment operation data, that is, the new fused low-dimensional data set, to achieve optimal control of the equipment.

[0034] In one possible implementation, Figure 2 As shown, the fused low-dimensional data set generation module also includes: Gaussian distribution calculation is performed on each data point in the fused high-dimensional data set in the high-dimensional space, and a high-dimensional probability distribution matrix is ​​output.

[0035] Perform t distribution calculation on each data point in the fused low-dimensional data set in the low-dimensional space, and output a low-dimensional probability distribution matrix.

[0036] The KL divergence of the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix is ​​calculated, and an objective function is constructed with the KL divergence, wherein the KL divergence represents the difference in probability distribution between the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix.

[0037] When the objective function converges, a fused low-dimensional data set is output.

[0038] Specifically, first, the fused high-dimensional data set contains high-dimensional data information obtained after multi-scale feature extraction and fusion. In high-dimensional space, for each data point in the data set, it is necessary to calculate the relationship between it and all other data points. The purpose of using Gaussian distribution calculation is to determine the probability distribution of each data point in high-dimensional space. For a specific data point, the distance between it and other data points is calculated. The distance is calculated using the common Euclidean distance. Based on these distance values, the Gaussian distribution function is used to calculate the similarity probability between the data point and other data points. The Gaussian distribution function has good mathematical properties and can effectively describe the distribution of data points in high-dimensional space. It takes the distance between data points as input and outputs a probability value between 0 and 1, indicating the similarity between two data points. The closer the distance, the higher the probability value, indicating that the two data points are more similar; the farther the distance, the lower the probability value, indicating that the two data points are less similar. For each data point in the fused high-dimensional data set, the above calculation process is repeated, that is, the distance between it and all other data points is calculated, and the similarity probability is calculated based on the distance. In this way, a high-dimensional probability distribution matrix can be obtained, and the number of rows and columns of this matrix is ​​equal to the number of data points in the data set. Each element in the matrix represents the similarity probability of two data points in the high-dimensional space.

[0039] In low-dimensional space, t distribution is calculated for each data point in the fused low-dimensional data set to output a low-dimensional probability distribution matrix. The fused low-dimensional data set is a data set that has been processed by feature dimensionality reduction. Its dimension is greatly reduced compared to the original fused high-dimensional data set. In low-dimensional space, t distribution calculation needs to be performed for each data point in the data set. T distribution is a probability distribution with a specific mathematical form and properties. For each data point, the distance between it and all other data points is calculated, which is similar to high-dimensional space, but the t distribution function is used in low-dimensional space to determine the similarity probability between data points based on these distances. T distribution has some advantages when processing low-dimensional data. It can better capture the distribution characteristics of data points in low-dimensional space, especially for those data with complex shapes or uneven distribution. Through t distribution calculation, a low-dimensional probability distribution matrix can be obtained. The number of rows and columns of this matrix is ​​also equal to the number of data points in the data set. Each element in the matrix represents the probability of similarity between two data points in low-dimensional space. By calculating the t distribution for each data point in the fused low-dimensional data set and outputting the low-dimensional probability distribution matrix, we can understand the distribution and similarity relationship of the data in the low-dimensional space. This provides an important basis for subsequent comparison of the probability distribution differences between high-dimensional space and low-dimensional space, and for further optimizing the representation of low-dimensional space.

[0040] First, KL divergence is an indicator used to measure the difference between two probability distributions. In this step, the high-dimensional probability distribution matrix represents the probability distribution of data in the high-dimensional space, while the low-dimensional probability distribution matrix reflects the probability distribution of data in the low-dimensional space after dimensionality reduction. The process of calculating KL divergence involves comparing and calculating the corresponding elements in the two probability distribution matrices. For each element in the high-dimensional probability distribution matrix and the corresponding element in the low-dimensional probability distribution matrix, the calculation is performed according to the calculation formula of KL divergence, which measures the amount of information required to convert from one probability distribution to another. By calculating the KL divergence of the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix, the degree of difference between the two probability distributions can be quantitatively determined. This difference reflects the distribution change of data in high-dimensional space and low-dimensional space. If the KL divergence is large, it means that the probability distribution in the high-dimensional space and the low-dimensional space is very different, that is, the dimensionality reduction process loses more important information; if the KL divergence is small, it means that the low-dimensional space better retains the probability distribution characteristics in the high-dimensional space. The objective function is constructed with KL divergence, and the goal is to minimize this KL divergence. The data representation in the low-dimensional space is continuously adjusted through the optimization algorithm, so that the low-dimensional probability distribution is as close as possible to the high-dimensional probability distribution, while retaining the important features of the data. Such an objective function can guide the dimensionality reduction process in a better direction, so that the fused low-dimensional data set after dimensionality reduction can reduce the data dimension while maintaining the essential characteristics of the data, which is convenient for subsequent data analysis and processing.

[0041] The objective function is constructed by calculating the KL divergence of the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix. Its purpose is to make the data distribution in the low-dimensional space as close as possible to the data distribution in the high-dimensional space, while reducing the dimension of the data. During the optimization process, the position of the data points in the low-dimensional space is continuously adjusted to reduce the value of the objective function. As the optimization proceeds, the low-dimensional probability distribution matrix gradually approaches the high-dimensional probability distribution matrix. When the objective function converges, it means that the data distribution in the low-dimensional space is close enough to the data distribution in the high-dimensional space and has reached a relatively stable state. At this time, the output fused low-dimensional data set has the following characteristics: First, it is greatly reduced in dimension, and it is easier to perform subsequent data analysis and processing, such as visualization, clustering, classification, etc., compared with the original fused high-dimensional data set. Secondly, although the dimension is reduced, due to the convergence of the objective function, this fused low-dimensional data set still retains most of the important feature information and can better reflect the essential characteristics of the original data. When the objective function converges, the fused low-dimensional data set output not only achieves data dimensionality reduction, but also retains the key information of the original data to a certain extent, providing strong support for subsequent data analysis and application.

[0042] In a possible implementation, the fused low-dimensional data set generation module further includes: The objective function is minimized by the gradient descent method to make the objective function converge. The gradient calculation formula is as follows: ; in, is the objective function, is the position of the i-th data point in the low-dimensional space. As the gradient descends, The position of the jth data point in the low-dimensional space, is a data point in high-dimensional space and The similarity between is a data point in low-dimensional space and The similarity between is a data point in low-dimensional space and The Euclidean distance between them is calculated based on gradient descent, and the distribution of data points in the low-dimensional space is gradually adjusted until the objective function converges.

[0043] Specifically, the objective function is minimized by the gradient descent method to achieve convergence of the objective function, thereby obtaining an optimized low-dimensional spatial data distribution. The objective function reflects the difference between the data distribution in high-dimensional space and low-dimensional space. In the t-SNE algorithm, the core goal is to make the point pairs in high-dimensional space similar. and the similarity of point pairs in low-dimensional space As close as possible, that is, to allow the low-dimensional space to better retain the characteristic information of the high-dimensional space.

[0044] For the position of the i-th data point in the low-dimensional space , which is a key variable in the algorithm. As the gradient descent proceeds, it is constantly adjusted The position of the jth data point in the low-dimensional space is moved in the direction that can reduce the objective function value. In this process, they also jointly affect the change of the objective function.

[0045] Similarity between data points in high-dimensional space It is calculated based on the distribution of data points in high-dimensional space, and it reflects the closeness of the relationship between data points in high-dimensional space. The similarity between data points in low-dimensional space It is calculated based on the positional relationship of data points in low-dimensional space.

[0046] Data points in low-dimensional space and The Euclidean distance between It measures their actual distance in space. This distance can further affect the similarity Calculation.

[0047] Based on the gradient descent system, the gradient of the objective function for each data point in the low-dimensional space is calculated each time. This gradient indicates the direction in which the objective function value rises fastest at the current position, and what needs to be done is to adjust the data point position in the direction in which the objective function value decreases. For each data point in the low-dimensional space , calculate the gradient direction and size at the current position according to the gradient calculation formula. If the objective function value is large, it means that the current low-dimensional space data distribution is very different from the high-dimensional space data distribution. At this time, the gradient will guide the data point to move in the direction that can reduce this difference.

[0048] As the gradient descent continues, the distribution of data points in the low-dimensional space is gradually adjusted, and the objective function value changes in each iteration. If the objective function value is gradually decreasing, it means that the data distribution in the low-dimensional space is getting closer and closer to the data distribution in the high-dimensional space. When the objective function converges, it means that the data distribution in the low-dimensional space has reached a stable state. At this time, the difference in data distribution between the high-dimensional space and the low-dimensional space is minimized, and an optimized fused low-dimensional data set is obtained. This data set reduces the data dimension while retaining the important features of the high-dimensional space, which facilitates subsequent data analysis and processing. By minimizing the objective function through the gradient descent method, adjusting the position of the data points in the low-dimensional space, and minimizing the difference in data distribution between the high-dimensional space and the low-dimensional space, an optimized fused low-dimensional data set is obtained, which is convenient for subsequent data analysis and processing and retains the important features of the high-dimensional space.

[0049] In a possible implementation, the fused low-dimensional data set generation module further includes: A correlation analysis is performed on the fused high-dimensional data set using the device control parameter items to determine related high-dimensional data sets in the fused high-dimensional data set.

[0050] The t-SNE algorithm is used to retain the relevant high-dimensional data sets to perform feature dimensionality reduction on the remaining fused high-dimensional data sets, and an updated fused low-dimensional data set is output.

[0051] Specifically, the fused high-dimensional data set is subjected to correlation analysis based on the equipment control parameter items. The equipment control parameter items are generated based on the fault information of the new energy power generation equipment, and are closely related to the operating status and control of the equipment. The purpose of the correlation analysis is to determine which data in the fused high-dimensional data set have a strong correlation with the equipment control parameter items. Through various correlation analysis systems, such as calculating the Pearson correlation coefficient, the Spearman correlation coefficient, etc., the linear or nonlinear relationship between each feature in the fused high-dimensional data set and the equipment control parameter items can be measured. If a feature has a high correlation coefficient with the equipment control parameter item, it means that this feature greatly affects the control decision of the equipment, and then the data point where this feature is located is considered to be part of the relevant high-dimensional data set. After the correlation analysis, the relevant high-dimensional data set in the fused high-dimensional data set is determined. This relevant high-dimensional data set contains those data points that are closely related to the equipment control parameter items, which have important reference value for the adaptive control of the equipment.

[0052] The t-SNE algorithm is used to retain the relevant high-dimensional data sets, and the remaining fused high-dimensional data sets are subjected to feature dimensionality reduction, and the updated fused low-dimensional data sets are output. The t-SNE algorithm has been used to perform feature dimensionality reduction on the fused high-dimensional data sets in the previous steps. The t-SNE algorithm is used again here to further reduce the dimension of the data while retaining important information, and to retain the relevant high-dimensional data sets separately, because these data are closely related to equipment control and their importance cannot be lost in the dimensionality reduction process. For the remaining fused high-dimensional data sets, that is, the part excluding the relevant high-dimensional data sets, the t-SNE algorithm is applied again for feature dimensionality reduction. The t-SNE algorithm finds the representation in the low-dimensional space by constructing the probability distribution between data points in the high-dimensional space and the low-dimensional space and minimizing the difference between the two probability distributions. In this process, the noise and redundant information in the data can be effectively removed while retaining the main features of the data. After being processed by the t-SNE algorithm, an updated fused low-dimensional data set is obtained. This data set contains both the retained relevant high-dimensional data sets and the dimensionality of the data is reduced by dimensionality reduction of the remaining data, making the data easier to analyze and process later. The updated fused low-dimensional dataset can better serve the adaptive control of renewable energy power generation equipment and provide more accurate information support for the efficient operation and optimization of the equipment.

[0053] In a possible implementation, the multi-source high-dimensional data set output module further includes: Among them, the time feature convolution includes performing a convolution operation on the time series data of the multi-source power data set to extract periodic features and trend features.

[0054] The spatial feature convolution includes performing a convolution operation on the spatial data of the multi-source power data set, including the association between different devices and the operating status of the devices at different locations.

[0055] The type feature convolution includes performing a convolution operation on different types of data in the multi-source power data set to extract correlation features between different types of data.

[0056] Specifically, time feature convolution mainly operates on time series data of multi-source power data sets. Time series data includes changes in the output power of equipment over time, fluctuations in voltage and current, etc. By performing convolution operations on these time series data, periodic features and trend features can be extracted. Periodic features refer to the repetitive patterns of data in time. For example, the output power of new energy power generation equipment shows periodic changes in a day, with higher power during the day and lower power at night. Through time feature convolution, this periodic change law can be captured, providing important reference information for equipment control and optimization. Trend features reflect the changing trend of data over a longer time range. For example, the output power of a device may show an upward or downward trend as the device ages or environmental factors change. Time feature convolution can extract this trend feature to help predict the future state of the equipment and take corresponding maintenance measures in advance.

[0057] Spatial feature convolution focuses on the spatial data of multi-source power data sets. In a new energy power plant, different devices are distributed in different locations and they interact with each other. Spatial feature convolution can capture this spatial relationship and extract useful feature information. On the one hand, spatial feature convolution can consider the association between different devices. For example, there is mutual influence of airflow between multiple wind power generation devices, or there is a power transmission relationship between solar power generation equipment and energy storage equipment. By performing convolution operations on spatial data, the association features between these devices can be extracted to provide a basis for the coordinated control of the equipment. On the other hand, spatial feature convolution can also consider the differences in the operating status of equipment at different locations. For example, environmental factors such as light intensity and wind speed may be different in different geographical locations, which will lead to differences in the output power and operating efficiency of new energy power generation equipment at different locations. Spatial feature convolution can capture this spatial difference and provide a reference for equipment layout optimization and operation management.

[0058] Type feature convolution performs convolution operations on different types of data in multi-source power data sets. Multi-source power data sets for new energy power generation equipment contain multiple types of data, such as electrical parameter data, environmental factor data, equipment status data, etc. By performing convolution operations on different types of data, the correlation features between different types of data can be extracted. For example, the relationship between light intensity and the output power of solar power generation equipment, or the relationship between wind speed and the output power of wind power generation equipment, is found. These correlation features can help better understand the operating mechanism and performance characteristics of the equipment, and provide more comprehensive information support for the control and optimization of the equipment. In summary, through time feature convolution, space feature convolution, and type feature convolution, rich feature information can be extracted from multi-source power data sets, providing strong support for subsequent data analysis and equipment adaptive control.

[0059] In a possible implementation, the adaptive control module further includes: A plurality of groups of multi-source power data sets of the new energy power generation equipment in a plurality of working modes are obtained, wherein the plurality of working modes include a load mode, a non-load mode, an energy storage mode and a fault mode.

[0060] A plurality of fused low-dimensional data sets corresponding to the plurality of multi-source power data sets are obtained.

[0061] A multi-mode-device adaptive control model is trained according to the multiple groups of fused low-dimensional data sets, and mode adaptive control is performed on the new energy power generation equipment using the multi-mode-device adaptive control model.

[0062] Specifically, first, multiple sets of multi-source power data sets of new energy power generation equipment under multiple working modes are obtained. New energy power generation equipment usually has different working modes, which reflect the status of the equipment under different operating conditions. The multiple working modes here include load mode, non-load mode, energy storage mode and fault mode. In load mode, the new energy power generation equipment is supplying power to the external load, and its parameters such as output power and voltage will be adjusted according to the load demand. In non-load mode, the equipment is idle or only supplies power to the internal system. The operating parameters at this time are different from those in load mode. The energy storage mode usually involves storing excess electrical energy for subsequent use. The operation mode of the equipment in this mode also has its characteristics, while the fault mode is a special state when the equipment fails. The data at this time contains fault feature information. By collecting multi-source power data sets of the equipment in these different working modes, the operating status and data characteristics of the equipment under various conditions can be fully understood. The multi-source power data set includes data sources from multiple aspects such as electrical parameters, environmental data, and equipment status information of the equipment.

[0063] Multiple groups of fused low-dimensional data sets corresponding to multiple groups of multi-source power data sets are obtained. For each group of multi-source power data sets obtained, they are processed through the previous steps, including multi-scale feature convolution, fusion, t-SNE algorithm dimensionality reduction and other operations, and finally the corresponding fused low-dimensional data sets are obtained. These fused low-dimensional data sets retain the important feature information in the original multi-source power data sets, but the dimension is greatly reduced, which is more convenient for subsequent analysis and processing. Each group of fused low-dimensional data sets reflects the operating characteristics of new energy power generation equipment under a specific working mode.

[0064] The support vector machine algorithm is used to train a multi-mode-device adaptive control model based on multiple groups of fused low-dimensional data sets, and this model is used to perform mode adaptive control on new energy power generation equipment.

[0065] Multiple groups of fused low-dimensional data sets are obtained after multi-scale feature extraction, fusion and dimensionality reduction processing of multiple groups of multi-source power data sets under different working modes of new energy power generation equipment. These data sets not only retain key feature information but also reduce data dimensions, providing an efficient data foundation for model training.

[0066] Divide multiple fused low-dimensional data sets into training sets and test sets, usually in a certain ratio, such as 70% as training sets and 30% as test sets. This ensures that the model has enough data to learn the rules during training, and can also be verified on the test set to evaluate the performance and generalization ability of the model.

[0067] Determine the kernel function of the support vector machine. Common kernel functions include linear kernel function, polynomial kernel function, radial basis kernel function (RBF), etc. The RBF kernel function performs well in dealing with nonlinear problems. It can map data to high-dimensional space, thereby better capturing complex relationships in the data. Select appropriate kernel function parameters, such as the bandwidth parameter of the RBF kernel function. This parameter is crucial to the performance of the model. The optimal parameter value can be found through systems such as cross-validation. Cross-validation can further divide the training set into multiple subsets, and train and verify on different subset combinations to find the best parameter settings.

[0068] The support vector machine model is trained using the training set data, with the samples in the fused low-dimensional data set as input and the corresponding device control parameters as output targets.

[0069] The support vector machine searches for a hyperplane so that most of the sample points in the training data set are as close to the hyperplane as possible, while ensuring that the hyperplane has a certain generalization ability. During the training process, the model will search for support vectors that play a key role in determining the hyperplane. These support vectors are a part of the sample points in the training data set that are located on or near the boundary of the hyperplane.

[0070] By using the minimum optimization algorithm of the optimization algorithm sequence to solve the parameters of the support vector machine model, the parameters of the support vector machine can be solved quickly and the training efficiency can be improved.

[0071] The trained support vector machine model is evaluated using the test set data, and the mean square error, mean absolute error and other indicators of the model on the test set are calculated to evaluate the prediction accuracy and generalization ability of the model. If the model performs poorly on the test set, you can consider adjusting the kernel function parameters and increasing the training data to improve the model. When the new energy power generation equipment is running, the working mode of the equipment is monitored in real time, and the corresponding fused low-dimensional data set is selected according to the current working mode of the equipment. This fused low-dimensional data set is input into the trained multi-mode-device adaptive control model, and the model predicts the control parameters suitable for the mode based on the input data and the current working mode. These control parameters can include the adjustment values ​​of electrical parameters such as the output power, voltage, and current of the equipment, or the changes in the operating mode and control strategy of the equipment. Finally, the predicted control parameters are sent to the control system of the new energy power generation equipment, and the control system performs the corresponding control operations. In this way, the multi-mode-device adaptive control model can automatically adjust the control parameters according to the real-time working mode of the equipment, and realize the mode adaptive control of the new energy power generation equipment. The support vector machine algorithm can be used to effectively train a multi-mode-device adaptive control model, and use this model to perform mode adaptive control of renewable energy power generation equipment. This system can improve the efficiency, stability and reliability of the equipment in different working modes, and provide strong support for the intelligent control of renewable energy power generation equipment.

[0072] In a possible implementation, the adaptive control module further includes: Acquire a real-time multi-source power data set of the new energy power generation equipment.

[0073] According to the real-time multi-source power data set, a real-time fused low-dimensional data set is output.

[0074] The real-time fused low-dimensional data set is input into the device adaptive control model to obtain adaptive control parameters that meet the preset power demand, and the new energy power generation equipment is adaptively controlled with the adaptive control parameters.

[0075] Specifically, obtain the real-time multi-source power data set of new energy power generation equipment. This data set contains various information about the equipment at the current moment, such as electrical parameters (voltage, current, power, etc.), environmental data (light intensity, wind speed, temperature, etc.) and equipment status information. Through sensors and monitoring equipment, these data can be collected in real time to understand the operation status of the equipment in a timely manner.

[0076] Based on the real-time multi-source power data set, a real-time fused low-dimensional data set is output. This process is similar to the previous processing of multiple sets of multi-source power data sets, but it is for real-time data. Through multi-scale feature extraction, fusion, and dimensionality reduction operations, the high-dimensional real-time multi-source power data set is converted into a low-dimensional fused low-dimensional data set. The advantage of this is that the complexity of the data is reduced, which is convenient for subsequent analysis and processing, and the key feature information in the data can also be retained.

[0077] The real-time fused low-dimensional data set is input into the device adaptive control model. This model is obtained by training multiple sets of fused low-dimensional data sets, and can predict the adaptive control parameters that meet the preset power demand based on the input data. For example, if the preset power demand is to maintain a stable output power under certain load conditions, then the model will predict the appropriate control parameters based on the information in the real-time fused low-dimensional data set, such as adjusting the output voltage, current and other electrical parameters of the device, or changing the operation mode of the device. Once the adaptive control parameters are obtained, these parameters can be used to adaptively control the new energy power generation equipment. The control system will adjust the operating state of the equipment according to these parameters to achieve the goal of meeting the preset power demand. Through this real-time adaptive control, the efficiency, stability and reliability of the new energy power generation equipment can be improved, and it can better adapt to different operating conditions and power demands, forming a complete real-time adaptive control process. Through real-time monitoring and control of new energy power generation equipment, it is ensured that the equipment can operate efficiently and stably to meet various power demands.

[0078] In a possible implementation, the adaptive control module further includes: If the new energy power generation equipment is a composite fuel power generation mode, the data correlation of each mode in the composite fuel power generation mode is analyzed and a correlation matrix is ​​output.

[0079] Model parameters of the multi-scale convolutional network layer are learned based on the correlation matrix.

[0080] Specifically, when the new energy power generation equipment is in the composite fuel power generation mode, the Pearson correlation coefficient algorithm is used to analyze the data correlation of each mode and output the correlation matrix. The Pearson correlation coefficient is an indicator to measure the degree of linear correlation between two variables. For different modes in the composite fuel power generation mode, the multi-source power data sets of each mode are first sorted out, and for each pair of different modes, the Pearson correlation coefficients between each characteristic variable in the data set are calculated respectively. For example, for the power output time series data under the two modes, it is regarded as two sets of variables, and the linear correlation degree of the two time series is determined by calculating the Pearson correlation coefficient. If the correlation coefficient is close to 1, it indicates that the two modes have a strong positive correlation in terms of power output over time, that is, their changing trends are very similar; if the correlation coefficient is close to -1, it indicates that there is a strong negative correlation, which means that when the power output of one mode increases, the power output of the other mode tends to decrease; if it is close to 0, it means that the linear correlation is weak, and the relationship between the two modes in the power output time series is not obvious. For other characteristic variables, such as fuel consumption rate, equipment temperature, voltage and current, the Pearson correlation coefficients are also calculated one by one. By comprehensively analyzing the correlation coefficients of each characteristic variable in different modes, a correlation matrix is ​​constructed. Each element in the matrix represents the Pearson correlation coefficient value of two specific modes on a certain characteristic variable. Through the Pearson correlation coefficient algorithm, the data correlation between different composite fuel power generation modes is accurately quantified, providing an important basis for the subsequent model parameter learning of the multi-scale convolutional network layer and the adaptive control of new energy power generation equipment.

[0081] The correlation matrix provides important guidance information for the parameter learning of the multi-scale convolutional network layer. This matrix reflects the data correlation between each mode in the composite fuel power generation mode. By analyzing the values ​​in the matrix, we can understand the similarity and correlation of different modes in different features. For the learning of convolution weights, the correlation matrix is ​​used to adjust the weight distribution on different feature channels. If two modes have a high correlation on a certain feature, then the convolution weight is appropriately increased on the corresponding feature channel to better capture and utilize this correlation. The correlation matrix also provides valuable reference for learning the scale of feature convolution. Different modes may show correlation at different time, space and type feature scales. If the correlation matrix shows that some modes have a high correlation at a specific time scale, then the scale parameters of the time feature convolution in the multi-scale convolutional network layer are adjusted to make it more suitable for processing the data of these related modes. Similarly, the scale parameters of the spatial feature convolution and the type feature convolution can also be adjusted accordingly according to the correlation matrix. The specific learning process uses optimization algorithms such as gradient descent. By calculating the gradient of the loss function to the convolution weight and the feature convolution scale, the parameters are updated according to the direction and size of the gradient. When calculating the loss function, the correlation information between the modes reflected by the correlation matrix is ​​taken into account, so that the network can better adapt to the characteristics of the composite fuel power generation mode during the learning process and improve the processing ability and accuracy of different mode data.

[0082] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0084] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. Adaptive control system for renewable energy power generation equipment based on multi-scale data fusion, characterized in that: The system comprises: A multi-source power data set acquisition module is used to obtain new energy fuel information of new energy power generation equipment, and determine whether the new energy power generation equipment is a composite fuel power generation mode according to the new energy fuel information. If the new energy power generation equipment is a composite fuel power generation mode, obtain a multi-source power data set corresponding to each composite fuel type; A multi-source high-dimensional data set output module, used to perform multi-scale feature convolution on the multi-source power data set through a multi-scale convolutional network layer, and output a multi-source high-dimensional data set, wherein the multi-scale feature convolution includes time feature convolution, space feature convolution and type feature convolution; A fused high-dimensional data set generation module, used to fuse the multi-source high-dimensional data sets to generate a fused high-dimensional data set; An equipment control parameter item generating module, used for the new energy power generation equipment to receive equipment fault information and generate equipment control parameter items according to the equipment fault information; A fused low-dimensional data set generation module is used to perform feature dimensionality reduction on the fused high-dimensional data set using a t-SNE algorithm, and output a fused low-dimensional data set; An adaptive control module is used to train an equipment adaptive control model using the fused low-dimensional data set and the equipment control parameter items, and to perform adaptive control on the new energy power generation equipment based on the equipment adaptive control model.

2. The system according to claim 1, characterized in that The fused low-dimensional data set generation module also includes: Performing Gaussian distribution calculation on each data point in the fused high-dimensional data set in a high-dimensional space, and outputting a high-dimensional probability distribution matrix; Performing t-distribution calculation on each data point in the fused low-dimensional data set in a low-dimensional space, and outputting a low-dimensional probability distribution matrix; Calculating the KL divergence of the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix, and constructing an objective function with the KL divergence, wherein the KL divergence represents the difference in probability distribution between the high-dimensional probability distribution matrix and the low-dimensional probability distribution matrix; When the objective function converges, a fused low-dimensional data set is output.

3. The system according to claim 2, characterized in that The fused low-dimensional data set generation module also includes: ； in, is the objective function, is the position of the i-th data point in the low-dimensional space. As the gradient descends, The position of the jth data point in the low-dimensional space, is a data point in high-dimensional space and The similarity between is a data point in low-dimensional space and The similarity between is a data point in low-dimensional space and The Euclidean distance between them is calculated based on gradient descent, and the distribution of data points in the low-dimensional space is gradually adjusted until the objective function converges.

4. The system according to claim 1, characterized in that The fused low-dimensional data set generation module also includes: Performing correlation analysis on the fused high-dimensional data set using the device control parameter items to determine related high-dimensional data sets in the fused high-dimensional data set; The t-SNE algorithm is used to retain the relevant high-dimensional data sets to perform feature dimensionality reduction on the remaining fused high-dimensional data sets, and an updated fused low-dimensional data set is output.

5. The system according to claim 1, wherein: The multi-source high-dimensional data set output module also include: Wherein, the time feature convolution includes performing a convolution operation on the time series data of the multi-source power data set to extract periodic features and trend features; The spatial feature convolution includes performing a convolution operation on the spatial data of the multi-source power data set, including associations between different devices and operating states of the devices at different locations; The type feature convolution includes performing a convolution operation on different types of data in the multi-source power data set to extract correlation features between different types of data.

6. The system according to claim 1, wherein: The adaptive control module also includes: Acquire multiple groups of multi-source power data sets of the new energy power generation equipment under multiple working modes, wherein the multiple working modes include a load mode, a non-load mode, an energy storage mode, and a fault mode; Acquire multiple groups of fused low-dimensional data sets corresponding to the multiple groups of multi-source power data sets respectively; A multi-mode-device adaptive control model is trained according to the multiple groups of fused low-dimensional data sets, and mode adaptive control is performed on the new energy power generation equipment using the multi-mode-device adaptive control model.

7. The system according to claim 1, characterized in that The adaptive control module also includes: Acquire a real-time multi-source power data set of the new energy power generation equipment; Outputting a real-time fused low-dimensional data set according to the real-time multi-source power data set; The real-time fused low-dimensional data set is input into the device adaptive control model to obtain adaptive control parameters that meet the preset power demand, and the new energy power generation equipment is adaptively controlled with the adaptive control parameters.

8. The system of claim 1, wherein: The adaptive control module also includes: If the new energy power generation equipment is a composite fuel power generation mode, analyzing the data correlation of each mode in the composite fuel power generation mode and outputting a correlation matrix; Model parameters of the multi-scale convolutional network layer are learned based on the correlation matrix.

Images