Data processing method, device and equipment of power source end and storage medium

Abstract

The invention discloses a data processing method and device for a power source end, equipment and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: collecting initial data of the power source end, and carrying out the data preprocessing of the initial data to obtain target data; the types of the target data comprise environmental parameters, equipment state parameters and electrical parameters; generating a target irradiance change rate based on the target data, and determining a target time window based on the target irradiance change rate; determining a target statistical feature and a target fluctuation feature corresponding to the target data based on the target time window, and generating a plurality of initial association features corresponding to the target data based on the target statistical feature and the target fluctuation feature; and determining a maximum information coefficient between each initial association feature and the target power task, and screening out a target association feature from each initial association feature based on the maximum information coefficient, so as to process the target power task based on the target association feature. The data utilization value of the power source end can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data processing method, apparatus, equipment, and storage medium at a power source. Background Technology

[0002] The power source refers to the "source of power production / injection," which is the starting point of the power system's link from "energy generation" to "transmission and distribution." It corely corresponds to various generation-side equipment, energy plants, and related access links, serving as the core carrier of "where does electricity come from." The operation of the power source generates massive amounts of multi-dimensional time-series data. Data processing at the power source is crucial for ensuring the stable operation of the power system and improving the accuracy of power prediction and the efficiency of equipment fault early warning.

[0003] Feature extraction, as a core step in data processing at the power source, affects the effectiveness of subsequent tasks. However, current mainstream processing methods mostly use fixed time windows to extract statistical features, which has several drawbacks: First, it cannot adapt to scenarios with drastic changes in irradiance, such as when the weather changes from cloudy to sunny, the fixed window will smooth out key fluctuation information; second, it processes environmental, equipment, and electrical data in isolation, ignoring cross-dimensional correlations, such as failing to explore the chain relationship of "sudden drop in irradiance - inverter overcurrent - power drop"; and third, it has high feature redundancy, which weakens its support for subsequent power prediction and fault early warning.

[0004] In summary, how to construct an efficient and accurate data extraction process at the power source is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a data processing method, apparatus, device, and storage medium at the power source end, which can construct an efficient and accurate data extraction process at the power source end, thereby improving the utilization value of the data at the power source end. The specific solution is as follows:

[0006] Firstly, this application provides a data processing method at the power source end, including:

[0007] Initial data is collected from the power source, and the initial data is preprocessed to obtain target data; the types of target data include environmental parameters, equipment status parameters, and electrical parameters.

[0008] A target irradiance change rate is generated based on the target data, and a target time window is determined based on the target irradiance change rate.

[0009] determine a target statistical feature and a target fluctuation feature corresponding to the target data based on the target time window, and generate a plurality of initial correlation features corresponding to the target data based on the target statistical feature and the target fluctuation feature; the initial correlation features are features used to represent the correlation between different types of target data;

[0010] determine a maximum information coefficient between each of the initial correlation features and the target power task, and select a target correlation feature from each of the initial correlation features based on the maximum information coefficient, so as to process the target power task based on the target correlation feature.

[0011] Optionally, the initial data of the power source end is collected, and the initial data is preprocessed to obtain target data, including:

[0012] The initial data of the power source end is synchronously collected based on a preset collection frequency by using a target sensor;

[0013] determine abnormal data in the initial data, and mark the abnormal data in the initial data to obtain marked data;

[0014] determine missing data in the marked data, and fill the missing data to obtain filled data;

[0015] The filled data is time-aligned based on an inverter clock to obtain the target data;

[0016] The target data includes target environmental data corresponding to environmental parameters, target device state data corresponding to device state parameters, and target electrical data corresponding to electrical parameters.

[0017] Optionally, the process of determining the abnormal data in the initial data includes:

[0018] The initial electrical data is divided into first initial electrical data corresponding to a preset strong light scene and second initial electrical data corresponding to a preset weak light scene;

[0019] determine a first preset abnormal range corresponding to the preset strong light scene based on a first mean value and a first standard deviation corresponding to the first initial electrical data, and determine a second preset abnormal range corresponding to the preset weak light scene based on a second mean value and a second standard deviation corresponding to the second initial electrical data;

[0020] determine first abnormal electrical data in the first initial electrical data based on the first preset abnormal range, and determine second abnormal electrical data in the second initial electrical data based on the second preset abnormal range.

[0021] Optionally, the determining the target statistical feature and the target fluctuation feature corresponding to the target data comprises:

[0022] determining a first target statistical feature and a first target fluctuation feature corresponding to the target environment data, determining a second target statistical feature and a second target fluctuation feature corresponding to the target device state data, and determining a third target statistical feature and a third target fluctuation feature corresponding to the target electrical data;

[0023] wherein the first target statistical feature, the second target statistical feature and the third target statistical feature each include a target mean value, a target median value, a target standard deviation, a target maximum value, a target minimum value, a target kurtosis and a target skewness, the first target fluctuation feature includes the target irradiance change rate, and the second target fluctuation feature includes the device temperature change rate;

[0024] Correspondingly, in the determining the third target fluctuation feature corresponding to the target electrical data, the method comprises:

[0025] determining a ratio between a standard deviation of active power in the target electrical data and a mean value of the active power, and determining a target power fluctuation coefficient based on the ratio;

[0026] determining a maximum change amount of power per unit time in the target electrical data, and determining a target maximum ramp rate based on the maximum change amount; the power includes the active power and the reactive power;

[0027] determining a number of times that the active power exceeds a target power threshold interval, and determining a target fluctuation frequency based on the number of times.

[0028] Optionally, before the generating a plurality of initial correlation features corresponding to the target data based on the target statistical feature and the target fluctuation feature, the method further comprises:

[0029] generating a first feature vector corresponding to the target environment data based on the first target statistical feature and the first target fluctuation feature by using a one-dimensional convolutional neural network;

[0030] generating a second feature vector corresponding to the target device state data based on the second target statistical feature and the second target fluctuation feature by using a gated recurrent unit;

[0031] generating a third feature vector corresponding to the target electrical data based on the third target statistical feature and the third target fluctuation feature by using a wavelet transform analysis method, so as to generate a plurality of the initial correlation features based on the first feature vector, the second feature vector and the third feature vector.

[0032] Optionally, the generating the initial correlation features based on the first feature vector, the second feature vector and the third feature vector comprises:

[0033] generating a first mutual information matrix between the first feature vector and the second feature vector, a second mutual information matrix between the first feature vector and the third feature vector, and a third mutual information matrix between the second feature vector and the third feature vector;

[0034] generating a first initial weight corresponding to the first feature vector based on the first mutual information matrix and the second mutual information matrix, a second initial weight corresponding to the second feature vector based on the first mutual information matrix and the third mutual information matrix, and a third initial weight corresponding to the third feature vector based on the second mutual information matrix and the third mutual information matrix;

[0035] normalizing the first initial weight, the second initial weight and the third initial weight to obtain a first target weight corresponding to the first initial weight, a second target weight corresponding to the second initial weight, and a third target weight corresponding to the third initial weight;

[0036] weighting and splicing the first feature vector, the second feature vector and the third feature vector based on the first target weight, the second target weight and the third target weight to obtain a target correlation feature vector, and determining the initial correlation features based on the target correlation feature vector.

[0037] Optionally, the determining the maximum information coefficient between each of the initial correlation features and the target power task, and screening a target correlation feature from each of the initial correlation features based on the maximum information coefficient comprises:

[0038] determining a target evaluation index corresponding to the target power task and a target maximum information coefficient threshold;

[0039] determining the maximum information coefficient between each of the initial correlation features and the target evaluation index, and determining whether the maximum information coefficient is greater than or equal to the target maximum information coefficient threshold, if yes, determining the initial correlation feature as the target correlation feature;

[0040] The target evaluation index comprises a prediction error corresponding to a target power prediction task and a warning accuracy corresponding to a target device fault early warning task, and the target maximum information coefficient threshold comprises a first maximum information coefficient threshold corresponding to the target power prediction task and a second maximum information coefficient threshold corresponding to the target device fault early warning task.

[0041] Secondly, this application provides a data processing device at a power source, comprising:

[0042] The data acquisition module is used to acquire initial data from the power source and preprocess the initial data to obtain target data; the types of target data include environmental parameters, equipment status parameters, and electrical parameters.

[0043] The time window determination module is used to generate a target irradiance change rate based on the target data, and to determine a target time window based on the target irradiance change rate.

[0044] The correlation feature generation module is used to determine the target statistical features and target fluctuation features corresponding to the target data based on the target time window, and to generate a number of initial correlation features corresponding to the target data based on the target statistical features and the target fluctuation features; the initial correlation features are features used to represent the correlation between different types of target data;

[0045] The association feature filtering module is used to determine the maximum information coefficient between each of the initial association features and the target power task, and to filter out the target association features from each of the initial association features based on the maximum information coefficient, so as to process the target power task based on the target association features.

[0046] Thirdly, this application provides an electronic device, comprising:

[0047] Memory, used to store computer programs;

[0048] A processor is used to execute the computer program to implement the aforementioned data processing method at the power source.

[0049] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned data processing method at the power source end.

[0050] In this application, initial data from the power source is first collected, and the initial data is preprocessed to obtain target data. The types of target data include environmental parameters, equipment status parameters, and electrical parameters. Then, a target irradiance change rate is generated based on the target data, and a target time window is determined based on the target irradiance change rate. Subsequently, target statistical features and target fluctuation features corresponding to the target data are determined based on the target time window, and several initial correlation features corresponding to the target data are generated based on the target statistical features and the target fluctuation features. The initial correlation features are features used to represent the correlation between different types of target data. Finally, the maximum information coefficient between each initial correlation feature and the target power task is determined, and target correlation features are selected from each initial correlation feature based on the maximum information coefficient, so as to process the target power task based on the target correlation features. As can be seen from the above, this application first collects and preprocesses initial data from the power source to obtain target data in three categories: environment, equipment status, and electrical data. Then, it determines the target irradiance change rate to establish a suitable target time window. Based on this time window, it extracts target statistical features and target fluctuation features from the target data, thereby generating initial correlation features reflecting the correlation between different types of data. Finally, by calculating the maximum information coefficient between each initial correlation feature and the target power task, it selects target correlation features for processing the target power task. In this way, this application, through dynamic window adaptation to data fluctuations, can retain key information; simultaneously, it can uncover correlations between cross-type power data. By constructing an efficient and accurate data extraction process at the power source, it reduces feature redundancy and provides precise feature support for power tasks, thereby improving the efficiency and utilization value of data processing at the power source. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 A flowchart of a data processing method at the power source end provided in this application;

[0053] Figure 2 This application provides a specific flowchart for data feature extraction from a power source.

[0054] Figure 3 A schematic diagram of a specific data feature extraction process at the power source end provided in this application;

[0055] Figure 4 A schematic diagram of a data processing device at the power source end provided in this application;

[0056] Figure 5 This application provides a structural diagram of an electronic device. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The operation of power sources generates massive amounts of multi-dimensional time-series data. Data processing at the power source end is crucial for ensuring stable power system operation and improving power prediction accuracy and equipment fault early warning efficiency. Feature extraction, as a core step in power source data processing, significantly impacts the effectiveness of subsequent tasks. However, current mainstream methods often employ fixed time windows to extract statistical features, which has several drawbacks: firstly, it cannot adapt to scenarios with drastic changes in irradiance, such as when cloudy skies turn sunny, as the fixed window smooths out key fluctuation information; secondly, it processes environmental, equipment, and electrical data in isolation, ignoring cross-dimensional correlations, such as failing to uncover the chain relationship of "sudden irradiance drop - inverter overcurrent - power drop"; and thirdly, it suffers from high feature redundancy, offering weak support for subsequent power prediction and fault early warning. Therefore, this application provides a power source data processing scheme that constructs an efficient and accurate power source data extraction process, thereby enhancing the utilization value of power source data.

[0059] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a data processing method at the power source end, which may include:

[0060] Step S11: Collect initial data from the power source and preprocess the initial data to obtain target data; the types of target data include environmental parameters, equipment status parameters, and electrical parameters.

[0061] In this embodiment, initial data from the power source needs to be collected first. During the data collection process, a target sensor can be used to synchronously collect the initial data from the power source based on a preset collection frequency. Specifically, the collection parameters and sensor configuration need to be determined first, as shown below:

[0062] When collecting initial environmental data corresponding to environmental parameters, irradiance is collected using an irradiance sensor, ambient temperature is collected using a platinum resistance temperature sensor, and wind speed is collected using a cup-type anemometer. The sky imager outputs a cloud cover level of 0-5. The irradiance sensor's measurement range is... The accuracy is The measurement range of the platinum resistance temperature sensor is The accuracy is The measurement range of the cup-type anemometer is The accuracy is .

[0063] When collecting initial equipment status data corresponding to equipment status parameters, the inverter's built-in thermocouple sensor collects the IGBT (Insulated Gate Bipolar Transistor) temperature, the combiner box voltage sensor collects the DC voltage, and the infrared thermal imager outputs... The system monitors hot spot values ​​for the primary components and collects the cooling system flow rate using an electromagnetic flowmeter; the thermocouple sensor's measurement range is... The accuracy is The measurement range of the combiner box voltage sensor is: The accuracy is FS (Full Scale), the measurement range of an electromagnetic flowmeter is... The accuracy is FS.

[0064] When collecting initial electrical data corresponding to electrical parameters, the AC side voltage is collected through the transformer substation voltage transformer, the AC side current is collected through the current transformer, and the active power, reactive power, and power factor are collected through the power transmitter; among them, the measurement range of the transformer substation voltage transformer is... The accuracy is The measurement range of the current transformer is The accuracy is The measurement range of the power transmitter is The accuracy is .

[0065] It should be noted that in this embodiment, all sensors collect data at a frequency of 1Hz during the acquisition process, and the data can be collected via a GPS (Global Positioning System) module with synchronization error. The edge computing gateway receives data and encapsulates it into second-level multidimensional time-series initial data in the format of "timestamp + power station ID + type", where the timestamp is accurate to milliseconds. In this way, by deploying multiple types of sensors and data acquisition terminals, it is possible to achieve second-level synchronous acquisition of environmental, equipment status and electrical parameters, and build a complete multidimensional time-series raw dataset.

[0066] It should be noted that, in order to preprocess the initial data to obtain the target data, the specific process may include: first, identifying abnormal data in the initial data and marking the abnormal data to obtain marked data; then, identifying missing data in the marked data and filling in the missing data to obtain filled data; finally, aligning the filled data with the inverter clock as a reference to obtain the target data; wherein, the target data includes target environmental data corresponding to environmental parameters, target equipment status data corresponding to equipment status parameters, and target electrical data corresponding to electrical parameters. Furthermore, in the process of determining abnormal data in the initial data, the specific procedure for determining abnormal electrical data may include: firstly, dividing the initial electrical data into first initial electrical data corresponding to a preset strong light scenario and second initial electrical data corresponding to a preset weak light scenario; then, determining a first preset abnormal range corresponding to the preset strong light scenario based on the first mean and first standard deviation of the first initial electrical data, and determining a second preset abnormal range corresponding to the preset weak light scenario based on the second mean and second standard deviation of the second initial electrical data; finally, determining the first abnormal electrical data in the first initial electrical data based on the first preset abnormal range, and determining the second abnormal electrical data in the second initial electrical data based on the second preset abnormal range.

[0067] Specifically, in one implementation, to determine abnormal electrical data, a preset strong light scene and a preset weak light scene can first be divided according to irradiance to determine... This is a preset strong light scene. The scenario is a preset low-light scene, and the initial electrical data is divided into first initial electrical data under a preset high-light scene and second initial electrical data under a preset low-light scene. Then, the first mean value corresponding to the first initial electrical data is calculated for each. and the first standard deviation The mean value corresponding to the second initial electrical data and standard deviation And determine the first preset anomaly range under the preset strong light scene as The data beyond the preset low-light scene corresponds to the second preset anomaly range. In addition to the data, abnormal electrical data can be obtained. Simultaneously, to determine the abnormal equipment status data, the IGBT temperature exceeding... DC voltage exceeding [0V, 1000V], low light Any hotspot value > 10 is considered an outlier and marked. Then, missing data can be filled in based on the missing time, as shown below:

[0068] (1) When consecutive deletions For seconds, the weighted time decay interpolation method can be used for filling, as shown in the following formula:

[0069] ;

[0070] Where x(ti) represents the valid data i seconds before time t.

[0071] (2) When consecutive deletions At a given time, historical data matching the current environment data within the past 7 days can be retrieved and filled with random sampling after fitting the probability distribution to maintain consistent statistical characteristics.

[0072] Finally, the padded data can be time-aligned: based on the inverter clock, synchronization error... For deviation The initial data is linearly interpolated and resampled to ensure that the same second timestamp corresponds to unique environmental data, equipment data, and electrical data, and the standardized data is output to obtain the target data.

[0073] Step S12: Generate the target irradiance change rate based on the target data, and determine the target time window based on the target irradiance change rate.

[0074] In this embodiment, it is necessary to use the irradiance sequence in the target environment data. Calculate the rate of change of target irradiance between adjacent time points The formula is shown below:

[0075] ;

[0076] Where t is a timestamp in seconds, when G(t-1)=0, . Used to quantify the intensity of irradiance fluctuation at time t relative to time t-1, expressed as a percentage.

[0077] Then, the target time window corresponding to the current moment can be determined based on the target irradiance change rate, as shown below:

[0078] when Since the lighting scene is stable, the window length of the target time window is set to 5 minutes, or 300 seconds, and the step size is set to 1 minute, or 60 seconds, in order to preserve the global statistical characteristics of the stable segment.

[0079] when Since the scene is of moderate fluctuation, the window length of the target time window is shortened to 2 minutes, or 120 seconds, and the step size is set to 30 seconds to focus on the local features of the illumination transition phase.

[0080] when In scenarios with violent fluctuations, such as rapid cloud cover, the target time window length was set to 30 seconds and the step size to 10 seconds to capture the detailed features of transient fluctuations.

[0081] In this way, the time window division can be linked with the intensity of irradiance fluctuations, thereby solving the problem of feature loss or redundancy when extracting features with a fixed window in abrupt change scenarios.

[0082] Step S13: Determine the target statistical features and target fluctuation features corresponding to the target data based on the target time window, and generate several initial correlation features corresponding to the target data based on the target statistical features and the target fluctuation features; the initial correlation features are features used to represent the correlation between different types of target data.

[0083] In this embodiment, target statistical features and target fluctuation features corresponding to target data can be determined within a target time window. The specific process may include: determining a first target statistical feature and a first target fluctuation feature corresponding to the target environmental data; determining a second target statistical feature and a second target fluctuation feature corresponding to the target equipment status data; and determining a third target statistical feature and a third target fluctuation feature corresponding to the target electrical data. The first, second, and third target statistical features each include a target mean, target median, target standard deviation, target maximum value, target minimum value, target kurtosis, and target skewness. The first target fluctuation feature includes the target irradiance change rate, and the second target fluctuation feature includes the equipment temperature change rate. Specifically, within each target time window, statistical features of the target environmental data, target equipment status data, and target electrical data are extracted, as shown below:

[0084] Mean: Reflects the average level of parameters within a window, such as the mean irradiance and the mean active power;

[0085] Median: Reduces outlier interference and characterizes the typical level of the parameter;

[0086] Standard deviation: reflects the degree of dispersion of parameters. For example, the larger the standard deviation of power, the more unstable the output.

[0087] Maximum / Minimum: Captures extreme values ​​within the window, such as maximum power and minimum inverter temperature;

[0088] Kurtosis: describes the steepness of the parameter distribution. For example, a kurtosis greater than 3 indicates that the power distribution is concentrated near the peak value.

[0089] Skewness: describes the symmetry of the parameter distribution. For example, a skewness greater than 0 indicates that the power distribution is right-skewed and the peak value is left-skewed.

[0090] Accordingly, the specific process for determining the third target fluctuation characteristic corresponding to the target electrical data may include: determining the ratio between the standard deviation of active power and the mean of active power in the target electrical data, and determining the target power fluctuation coefficient based on the ratio; determining the maximum change in power per unit time in the target electrical data, and determining the target maximum ramp rate based on the maximum change; the power includes active power and reactive power; determining the number of times the active power exceeds the target power threshold range, and determining the target fluctuation frequency based on the number of times. Specifically, in this embodiment, the core target fluctuation characteristics can be extracted within each target time window by combining the characteristics of the power source, as shown below:

[0091] Target power fluctuation coefficient Cp: The standard deviation of active power within the target time window. with the mean The ratio, the formula is: The larger the Cp value, the more severe the power fluctuation, which is directly related to the difficulty of power grid dispatching.

[0092] Target maximum climbing rate Rmax: The maximum change in power per unit time within the target time window, expressed by the formula: ; The change in power The sampling interval is 1 second; Rmax can be used to assess the impact on the power grid.

[0093] Target fluctuation frequency: Within a 10-minute target time window, the active power crosses the target power threshold range, i.e., the rated value. The frequency of the interval; the higher the frequency, the worse the stability.

[0094] In addition, the target fluctuation characteristics can also include irradiance-power response delay, which can be calculated using cross-correlation analysis to determine the time difference between irradiance and power changes. The formula is shown below:

[0095] ;

[0096] Under normal circumstances Second, The second indicates an abnormal inverter response.

[0097] It should be noted that the target fluctuation characteristics are directly related to the grid connection characteristics and equipment response performance of the power source. Moreover, based on dynamic window extraction, it can capture the fluctuation patterns under different lighting scenarios more accurately than a fixed window. For example, a 30-second window during severe fluctuations can identify millisecond-level power mutations.

[0098] In this embodiment, feature encoding can be performed using dedicated network branches to address the characteristics of different types of data. In one specific implementation, a one-dimensional convolutional neural network is used to generate a first feature vector corresponding to the target environmental data based on first target statistical features and first target fluctuation features; a gated recurrent unit is used to generate a second feature vector corresponding to the target equipment status data based on second target statistical features and second target fluctuation features; and a wavelet transform analysis method is used to generate a third feature vector corresponding to the target electrical data based on third target statistical features and third target fluctuation features, as shown below:

[0099] The target environmental data includes the first target statistical characteristics, such as mean irradiance and standard deviation of irradiance, as well as the first target fluctuation characteristics, such as the rate of change of irradiance. A 1D-CNN network (1D Convolutional Neural Network) with 2 convolutional layers, a kernel size of 5, and 32 output channels is used to extract local temporal features. After global average pooling, the first feature vector E corresponding to the 64-dimensional target environment data is obtained.

[0100] For the second target statistical features corresponding to the target device state data, such as the median IGBT temperature and the maximum hot spot value, as well as the second target fluctuation features corresponding to the target device state data, such as the slope of the temperature change trend, a GRU network (Gated Recurrent Unit) with one hidden layer, 64 neurons, and a dropout rate of 0.2 is used to learn long-term dependencies and output a 64-dimensional second feature vector D corresponding to the target device state data.

[0101] For the third target statistical features corresponding to the target electrical data, such as active power kurtosis and voltage standard deviation, as well as the third target fluctuation features corresponding to the target electrical data, such as target power fluctuation coefficient and target maximum ramp rate, wavelet transform analysis method is adopted, DB4 wavelet basis, 3-level decomposition, frequency domain features are extracted, and after compression by fully connected layer, a 64-dimensional third feature vector P corresponding to the target electrical data is obtained.

[0102] It should be noted that after obtaining the feature vectors corresponding to each type of data, the specific process for generating initial association features may include: firstly, generating a first mutual information matrix between the first feature vector and the second feature vector, generating a second mutual information matrix between the first feature vector and the third feature vector, and generating a third mutual information matrix between the second feature vector and the third feature vector; then, generating a first initial weight corresponding to the first feature vector based on the first mutual information matrix and the second mutual information matrix, generating a second initial weight corresponding to the second feature vector based on the first mutual information matrix and the third mutual information matrix, and generating a third initial weight corresponding to the third feature vector based on the second mutual information matrix and the third mutual information matrix; subsequently, normalizing the first initial weight, the second initial weight, and the third initial weight to obtain a first target weight corresponding to the first initial weight, a second target weight corresponding to the second initial weight, and a third target weight corresponding to the third initial weight; finally, weighted concatenation of the first feature vector, the second feature vector, and the third feature vector based on the first target weight, the second target weight, and the third target weight to obtain a target association feature vector, and determining several initial association features based on the target association feature vector.

[0103] Specifically, first, the mutual information matrix M between the eigenvectors is calculated, where M is a 3×3 matrix; and M[i][j] represents the mutual information value between the eigenvector of class i and the eigenvector of class j, which is normalized to... Specifically, the mutual information matrix is ​​as follows: Environment-Equipment M[1][2] (the first mutual information matrix between the first and second eigenvectors), Environment-Electrical M[1][3] (the second mutual information matrix between the first and third eigenvectors), and Equipment-Electrical M[2][3] (the third mutual information matrix between the second and third eigenvectors). For example, the mutual information matrix between Environment and Electrical is usually greater than or equal to 0.7, indicating a strong correlation. Afterward, the initial weights corresponding to various eigenvectors can be generated based on the mutual information matrix, as shown below:

[0104] wE=M[1][2]+M[1][3]; where wE is the first initial weight of the first feature vector of the environmental data, which is determined by the sum of the correlation strength between the second feature vector of the equipment status data and the third feature vector of the electrical data;

[0105] wD=M[1][2]+M[2][3]; where wD is the second initial weight of the second feature vector of the equipment status data, which is determined by the sum of the correlation strength between the first feature vector of the environmental data and the third feature vector of the electrical data;

[0106] wP=M[1][3]+M[2][3]; where wP is the third initial weight of the third feature vector of electrical data, which is determined by the sum of the correlation strength between the first feature vector of environmental data and the second feature vector of equipment status data.

[0107] Next, the initial weights can be normalized using the softmax function to obtain the first target weight *we*, the second target weight *wd*, and the third target weight *wp*. Finally, the feature vectors can be concatenated according to the attention weights to obtain a 128-dimensional target association feature vector *F*, as shown in the following formula:

[0108] ;

[0109] in, This represents vector concatenation. Then, based on the target associated feature vector F, several initial associated features can be obtained, as shown below:

[0110] (1) Environmental-electrical correlation: the influence coefficient of sudden change in irradiance on power (quantitative) Nonlinear relationship with Rmax, light intensity-power conversion efficiency (the ratio of average irradiance to average active power), etc.

[0111] (2) Equipment-electrical correlation: the coupling degree between inverter temperature and power factor, hot spot value and power loss rate, etc.;

[0112] (3) Environment-equipment correlation: the inhibition coefficient of cooling flow rate on IGBT temperature under high temperature environment, etc.

[0113] Step S14: Determine the maximum information coefficient between each of the initial association features and the target power task, and select the target association feature from each of the initial association features based on the maximum information coefficient, so as to process the target power task based on the target association feature.

[0114] In this embodiment, key features can be filtered and output by quantifying the correlation between features and target power tasks. First, it is necessary to determine the target evaluation indicators and the target maximum information coefficient threshold corresponding to the target power task. The target evaluation indicators include the prediction error corresponding to the target power prediction task and the early warning accuracy corresponding to the target equipment fault early warning task. The target maximum information coefficient threshold includes the first maximum information coefficient threshold corresponding to the target power prediction task and the second maximum information coefficient threshold corresponding to the target equipment fault early warning task. Specifically, for the core application scenario of centralized power sources, two types of target power tasks and their corresponding target evaluation indicators are defined:

[0115] Task 1: Power prediction task, with the evaluation metric being prediction error MAPE (Mean Absolute Percentage Error) and the target maximum information coefficient threshold. ;

[0116] Task 2: Equipment fault early warning task, with the evaluation metric being early warning accuracy, i.e., the proportion of correct early warnings among fault samples, and the target maximum information coefficient threshold. .

[0117] Next, the MIC (Maximum Information Coefficient) values ​​between each initial correlation feature f and the target evaluation index y can be determined. It should be noted that MIC is a statistic used to measure the strength of a non-linear association between two variables, and its value ranges from [value missing]. The closer the value is to 1, the stronger the correlation between the two, regardless of whether it is linear or nonlinear. For example, the MIC between "irradiance-power conversion efficiency" and power prediction error is usually greater than or equal to 0.7; a value close to 0 indicates a weaker correlation. The core idea of ​​calculating MIC is to calculate the maximum mutual information between variables through optimal grid partitioning and normalize the data to eliminate the influence of variable scaling. Specifically, in order to determine the MIC value between the initial correlation feature f and the target evaluation index y, this embodiment uses a bin-based MIC algorithm to partition the sample data of the initial correlation feature f and the evaluation index y into bins. The MIC value is determined by dividing the maximum mutual information by the maximum value of the grid entropy, as shown below:

[0118] (1) Normalize the initial correlation feature f and the evaluation index y respectively and map them to the [0,1] interval. The purpose of normalization is to eliminate the influence of the difference in the scale of different variables on the grid division and ensure that the features and evaluation index are analyzed on the same numerical scale.

[0119] (2) Normalized The axis is uniformly divided into n intervals, and binning is performed to obtain the grid boundary in the f direction: using normalized features The x-axis is the normalized evaluation index. Construct a two-dimensional coordinate system with the y-axis as the boundary. Divide the x-axis from 0 to 1 into n continuous intervals, i.e., perform binning. Each interval has a length of 1 / n, resulting in n+1 boundary points of the x-axis.

[0120] (3) Normalized The y-axis is uniformly divided into n intervals and binned to obtain the grid boundary in the y-direction: In the two-dimensional coordinate system, the y-axis is uniformly divided into n continuous intervals from 0 to 1, that is, binned, with each interval having a length of 1 / n, resulting in n+1 boundary points of the y-axis.

[0121] (4) Through these boundary points, the two-dimensional coordinate system is divided into n×n non-overlapping grids, each grid corresponding to a unique coordinate interval, such as the grid in the i-th row and j-th column corresponding to and This forms an n×n grid matrix G.

[0122] (5) Count the number of samples in each grid and calculate the total number of samples in the i-th bin in the f direction. and the total number of samples in the j-th bin in the y-direction .

[0123] (6) Mutual information (MI) is obtained by the difference between the joint distribution and the marginal distribution of two variables: Mutual information is used to quantify the difference between the joint distribution and the marginal distribution of two variables, as shown in the following formula:

[0124] ;

[0125] in, Let m be the number of samples falling into the grid in the i-th row and j-th column, and m be the total sample size, which is the total number of sample data pairs of feature f and evaluation index y. Let f be the total number of samples in the i-th bin along the associated feature f. Let y be the total number of samples in the j-th bin in the y-direction of the evaluation index.

[0126] (7) Calculate grid entropy: Grid entropy reflects the fineness of grid division and is used to normalize mutual information to eliminate the influence of the number of grids on the results. The formula is as follows:

[0127] ;

[0128] Where H is the grid entropy, and min(n,m) represents the smaller of the grid dimension n and the sample size m.

[0129] (8) Calculate the maximum information coefficient: The maximum information coefficient is the ratio of the maximum mutual information to the grid entropy, as shown in the following formula:

[0130] ;

[0131] in, This represents the maximum mutual information value under the grid division. Since the grid is fixed, the MI value under this division is directly taken here.

[0132] In this way, the minimum information coefficient (MIC) values ​​between each initial correlation feature f and the target evaluation index y can be calculated. Then, it can be determined whether the MIC is greater than or equal to the target maximum information coefficient threshold; if so, the initial correlation feature is determined as the target correlation feature. Specifically, for the power prediction task, the MIC is retained. The associated characteristics, such as the influence coefficient of sudden changes in irradiance and power conversion efficiency, are retained for equipment fault early warning tasks. The correlation characteristics include temperature-power factor coupling degree and hot spot-power loss rate. It should be noted that if the correlation characteristics simultaneously meet the maximum information coefficient threshold of two types of power tasks, they will be retained first, such as the cooling flow suppression coefficient.

[0133] Next, the selected target correlation features can be packaged into an optimal feature set in the following format:

[0134] Feature name, such as "Impact coefficient of irradiance mutation on power", feature vector value, such as 0.85, timestamp (aligned with window time), associated task, such as "power prediction" / "equipment fault warning" / "both", MIC value, such as 0.72.

[0135] Finally, the output can be sent to the task application module through the RESTful interface. When connecting to the power prediction system, only the features related to the power prediction task are output; when connecting to the fault early warning platform, only the features related to the equipment fault early warning task are output, and queries by time range and feature type are supported.

[0136] In this embodiment, the fusion design of dynamic window feature extraction and multi-type data association modeling overcomes the limitations of traditional fixed windows in adapting to highly fluctuating power source data. It accurately captures fluctuation characteristics under different lighting scenarios, such as transient power changes during drastic irradiance fluctuations. Simultaneously, it uncovers cross-dimensional coupling patterns between environment, equipment, and electrical data, such as the chain effect of sudden irradiance changes on inverter temperature and power output. This increases the correlation strength between extracted features and target power tasks such as power prediction and fault early warning by more than 30%, providing high-value feature support for intelligent data analysis at the power source. Furthermore, this embodiment effectively reduces data noise interference through layered preprocessing for dynamic anomaly identification and missing value imputation, combined with a feature selection mechanism based on the maximum information coefficient. The anomaly identification accuracy is ≥95%, and redundant features are reduced, decreasing the number of retained features by 40%~50%. This improves the robustness of feature extraction and reduces subsequent computational complexity, lowering power prediction errors by 15%~20% and improving equipment fault early warning accuracy by more than 25%, significantly enhancing the utilization efficiency and application effect of power source data.

[0137] In summary, in one specific implementation, see [link to relevant documentation]. Figure 2 and Figure 3 As shown, the specific process for data feature extraction at the power source end is as follows:

[0138] (1) Acquisition module, used to collect environmental parameters, equipment status parameters and electrical parameters at the power source end, forming second-level multi-dimensional time series raw data, i.e. initial data;

[0139] (2) The processing module is used to perform dynamic outlier identification, scenario-based filling of missing values, and time series alignment on the second-level multidimensional time series raw data to obtain standardized data, i.e. target data;

[0140] (3) Extraction module, used to obtain the irradiance change rate based on standardized data, adaptively divide time windows based on the irradiance change rate, and extract basic statistical features and fluctuation features in different time windows;

[0141] (4) Construction module, used to construct a multimodal network that integrates and improves the attention mechanism, and input basic statistical features and fluctuation features, and output correlation features, i.e. initial correlation features;

[0142] (5) The filtering module is used to calculate the maximum information coefficient between the associated features and the target task, and to filter the key features, i.e. the target associated features, through the maximum information coefficient, to form the optimal feature set and output it to the application module.

[0143] As can be seen from the above, in this embodiment, initial data from the power source is first collected, and the initial data is preprocessed to obtain target data. The types of target data include environmental parameters, equipment status parameters, and electrical parameters. Then, a target irradiance change rate is generated based on the target data, and a target time window is determined based on the target irradiance change rate. Subsequently, target statistical features and target fluctuation features corresponding to the target data are determined based on the target time window, and several initial correlation features corresponding to the target data are generated based on the target statistical features and the target fluctuation features. The initial correlation features are features used to represent the correlation between different types of target data. Finally, the maximum information coefficient between each initial correlation feature and the target power task is determined, and target correlation features are selected from each initial correlation feature based on the maximum information coefficient, so as to process the target power task based on the target correlation features. As can be seen from the above, this embodiment first collects initial data from the power source and preprocesses it to obtain target data including three categories: environment, equipment status, and electrical data. Then, it determines the target irradiance change rate to establish a suitable target time window. Based on the target time window, it extracts target statistical features and target fluctuation features of the target data, thereby generating initial correlation features reflecting the correlation between different types of data. Finally, by calculating the maximum information coefficient between each initial correlation feature and the target power task, it selects target correlation features for processing the target power task. In this way, this embodiment, by dynamically adapting to data fluctuations through a window, can retain key information; simultaneously, it can explore the correlation between cross-type power data. By constructing an efficient and accurate data extraction process at the power source, it reduces feature redundancy and provides precise feature support for power tasks, thereby improving the data processing efficiency and utilization value at the power source.

[0144] Accordingly, see Figure 4 As shown in the illustration, this application also provides a data processing device at the power source end, which may include:

[0145] The data acquisition module 11 is used to acquire initial data from the power source and preprocess the initial data to obtain target data; the types of target data include environmental parameters, equipment status parameters, and electrical parameters.

[0146] The time window determination module 12 is used to generate a target irradiance change rate based on the target data, and to determine a target time window based on the target irradiance change rate.

[0147] The correlation feature generation module 13 is used to determine the target statistical features and target fluctuation features corresponding to the target data based on the target time window, and to generate a number of initial correlation features corresponding to the target data based on the target statistical features and the target fluctuation features; the initial correlation features are features used to represent the correlation between different types of target data;

[0148] The association feature filtering module 14 is used to determine the maximum information coefficient between each of the initial association features and the target power task, and to filter out the target association features from each of the initial association features based on the maximum information coefficient, so as to process the target power task based on the target association features.

[0149] In some specific embodiments, the data acquisition module 11 may include:

[0150] The initial data acquisition submodule is used to synchronously acquire the initial data from the power source using the target sensor based on a preset acquisition frequency;

[0151] An abnormal data determination submodule is used to determine abnormal data in the initial data and mark the abnormal data in the initial data to obtain marked data;

[0152] The missing data filling submodule is used to determine the missing data in the marked data and fill the missing data to obtain the filled data;

[0153] The target data determination submodule is used to perform timing alignment on the filled data based on the inverter clock to obtain the target data; wherein, the target data includes target environmental data corresponding to environmental parameters, target equipment status data corresponding to equipment status parameters, and target electrical data corresponding to electrical parameters.

[0154] In some specific implementations, the abnormal data determination submodule may include:

[0155] An electrical data partitioning unit is used to partition the initial electrical data into a first initial electrical data corresponding to a preset strong light scenario and a second initial electrical data corresponding to a preset weak light scenario.

[0156] The preset anomaly range determination unit is used to determine the first preset anomaly range corresponding to the preset strong light scene based on the first mean and the first standard deviation corresponding to the first initial electrical data, and to determine the second preset anomaly range corresponding to the preset weak light scene based on the second mean and the second standard deviation corresponding to the second initial electrical data.

[0157] An abnormal electrical data determination unit is used to determine a first abnormal electrical data in the first initial electrical data based on the first preset abnormal range, and to determine a second abnormal electrical data in the second initial electrical data based on the second preset abnormal range.

[0158] In some specific embodiments, the associated feature generation module 13 may include:

[0159] The feature determination submodule is used to determine the first target statistical feature and the first target fluctuation feature corresponding to the target environmental data, and to determine the second target statistical feature and the second target fluctuation feature corresponding to the target equipment status data, and to determine the third target statistical feature and the third target fluctuation feature corresponding to the target electrical data; wherein the first target statistical feature, the second target statistical feature and the third target statistical feature each include target mean, target median, target standard deviation, target maximum value, target minimum value, target kurtosis and target skewness, the first target fluctuation feature includes the target irradiance change rate, and the second target fluctuation feature includes the equipment temperature change rate;

[0160] Accordingly, the feature determination submodule may include:

[0161] The ratio unit is used to determine the ratio between the standard deviation of active power and the mean of active power in the target electrical data, and to determine the target power fluctuation coefficient based on the ratio.

[0162] A change determination unit is used to determine the maximum change in power per unit time in the target electrical data, and to determine the target maximum ramp rate based on the maximum change; the power includes active power and reactive power;

[0163] The frequency determination unit is used to determine the number of times the active power exceeds the target power threshold range, and to determine the target fluctuation frequency based on the number of times.

[0164] In some specific embodiments, the data processing device at the power source end may further include:

[0165] The first feature vector generation submodule is used to generate a first feature vector corresponding to the target environment data based on the first target statistical features and the first target fluctuation features using a one-dimensional convolutional neural network.

[0166] The second feature vector generation submodule is used to generate a second feature vector corresponding to the target device status data based on the second target statistical features and the second target fluctuation features using a gated loop unit.

[0167] The initial correlation feature generation submodule is used to generate a third feature vector corresponding to the target electrical data based on the third target statistical features and the third target fluctuation features using wavelet transform analysis method, so as to generate a number of initial correlation features based on the first feature vector, the second feature vector and the third feature vector.

[0168] In some specific implementations, the initial association feature generation submodule may include:

[0169] The mutual information matrix generation unit is used to generate a first mutual information matrix between the first feature vector and the second feature vector, generate a second mutual information matrix between the first feature vector and the third feature vector, and generate a third mutual information matrix between the second feature vector and the third feature vector.

[0170] An initial weight generation unit is configured to generate a first initial weight corresponding to the first feature vector based on the first mutual information matrix and the second mutual information matrix, generate a second initial weight corresponding to the second feature vector based on the first mutual information matrix and the third mutual information matrix, and generate a third initial weight corresponding to the third feature vector based on the second mutual information matrix and the third mutual information matrix.

[0171] The target weight generation unit is used to normalize the first initial weight, the second initial weight, and the third initial weight to obtain the first target weight corresponding to the first initial weight, the second target weight corresponding to the second initial weight, and the third target weight corresponding to the third initial weight.

[0172] An initial association feature generation unit is used to perform weighted concatenation of the first feature vector, the second feature vector, and the third feature vector based on the first target weight, the second target weight, and the third target weight to obtain a target association feature vector, and to determine a number of initial association features based on the target association feature vector.

[0173] In some specific embodiments, the association feature filtering module 14 may include:

[0174] An evaluation index determination unit is used to determine the target evaluation index and the target maximum information coefficient threshold corresponding to the target power task.

[0175] The target association feature determination unit is used to determine the maximum information coefficient between each of the initial association features and the target evaluation index, and to determine whether the maximum information coefficient is greater than or equal to the target maximum information coefficient threshold. If so, the initial association feature is determined as the target association feature. The target evaluation index includes the prediction error corresponding to the target power prediction task and the early warning accuracy corresponding to the target equipment fault early warning task. The target maximum information coefficient threshold includes the first maximum information coefficient threshold corresponding to the target power prediction task and the second maximum information coefficient threshold corresponding to the target equipment fault early warning task.

[0176] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the data processing method at the power source end disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0177] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0178] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0179] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the data processing method at the power source end executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0180] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned data processing method for the power source. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0181] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0182] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0183] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0184] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0185] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A data processing method at a power source, characterized in that, include: Initial data is collected from the power source, and the initial data is preprocessed to obtain the target data; The types of target data include environmental parameters, equipment status parameters, and electrical parameters; A target irradiance change rate is generated based on the target data, and a target time window is determined based on the target irradiance change rate. Based on the target time window, the target statistical features and target fluctuation features corresponding to the target data are determined, and several initial correlation features corresponding to the target data are generated based on the target statistical features and the target fluctuation features. The initial association feature is a feature used to represent the association between different types of target data; The maximum information coefficient between each of the initial correlation features and the target power task is determined, and the target correlation feature is selected from each of the initial correlation features based on the maximum information coefficient, so as to process the target power task based on the target correlation feature.

2. The data processing method at the power source end according to claim 1, characterized in that, The process of collecting initial data from the power source and preprocessing the initial data to obtain target data includes: The initial data from the power source is synchronously collected using the target sensor based on a preset collection frequency; Identify the abnormal data in the initial data and mark the abnormal data in the initial data to obtain the marked data; Identify the missing data in the marked data and fill in the missing data to obtain the filled data; The padded data is time-aligned based on the inverter clock to obtain the target data; The target data includes target environmental data corresponding to environmental parameters, target equipment status data corresponding to equipment status parameters, and target electrical data corresponding to electrical parameters.

3. The data processing method at the power source end according to claim 2, characterized in that, The process of determining abnormal data in the initial data includes: The initial electrical data is divided into first initial electrical data corresponding to a preset strong light scenario and second initial electrical data corresponding to a preset weak light scenario; Based on the first mean and first standard deviation of the first initial electrical data, a first preset abnormal range corresponding to the preset strong light scene is determined, and based on the second mean and second standard deviation of the second initial electrical data, a second preset abnormal range corresponding to the preset weak light scene is determined. The first abnormal electrical data in the first initial electrical data is determined based on the first preset abnormal range, and the second abnormal electrical data in the second initial electrical data is determined based on the second preset abnormal range.

4. The data processing method at the power source end according to claim 2, characterized in that, Determining the target statistical characteristics and target fluctuation characteristics corresponding to the target data includes: The first target statistical feature and the first target fluctuation feature corresponding to the target environmental data are determined, the second target statistical feature and the second target fluctuation feature corresponding to the target equipment status data are determined, and the third target statistical feature and the third target fluctuation feature corresponding to the target electrical data are determined. The first target statistical feature, the second target statistical feature, and the third target statistical feature all include target mean, target median, target standard deviation, target maximum value, target minimum value, target kurtosis, and target skewness. The first target fluctuation feature includes the target irradiance change rate, and the second target fluctuation feature includes the equipment temperature change rate. Accordingly, the process of determining the third target fluctuation characteristic corresponding to the target electrical data includes: Determine the ratio between the standard deviation of active power and the mean of active power in the target electrical data, and determine the target power fluctuation coefficient based on the ratio; Determine the maximum change in power per unit time in the target electrical data, and determine the target maximum ramp rate based on the maximum change; the power includes active power and reactive power; The number of times the active power exceeds the target power threshold range is determined, and the target fluctuation frequency is determined based on the number of times.

5. The data processing method at the power source end according to claim 4, characterized in that, Before generating several initial correlation features corresponding to the target data based on the target statistical features and the target fluctuation features, the method further includes: Using a one-dimensional convolutional neural network, a first feature vector corresponding to the target environment data is generated based on the first target statistical features and the first target fluctuation features; Using a gated loop unit, a second feature vector corresponding to the target device status data is generated based on the second target statistical features and the second target fluctuation features; Using wavelet transform analysis, a third feature vector corresponding to the target electrical data is generated based on the third target statistical features and the third target fluctuation features, so as to generate several initial correlation features based on the first feature vector, the second feature vector and the third feature vector.

6. The data processing method at the power source end according to claim 5, characterized in that, The generation of several initial associated features based on the first feature vector, the second feature vector, and the third feature vector includes: Generate a first mutual information matrix between the first feature vector and the second feature vector, generate a second mutual information matrix between the first feature vector and the third feature vector, and generate a third mutual information matrix between the second feature vector and the third feature vector. A first initial weight corresponding to the first feature vector is generated based on the first mutual information matrix and the second mutual information matrix, and a second initial weight corresponding to the second feature vector is generated based on the first mutual information matrix and the third mutual information matrix, and a third initial weight corresponding to the third feature vector is generated based on the second mutual information matrix and the third mutual information matrix. The first initial weight, the second initial weight, and the third initial weight are normalized to obtain the first target weight corresponding to the first initial weight, the second target weight corresponding to the second initial weight, and the third target weight corresponding to the third initial weight. The first feature vector, the second feature vector, and the third feature vector are weighted and concatenated based on the first target weight, the second target weight, and the third target weight to obtain the target association feature vector, and several initial association features are determined based on the target association feature vector.

7. The data processing method at the power source end according to any one of claims 1 to 6, characterized in that, The step of determining the maximum information coefficient between each of the initial correlation features and the target power task, and filtering the target correlation features from each of the initial correlation features based on the maximum information coefficient, includes: Determine the target evaluation index and the target maximum information coefficient threshold corresponding to the target power task; The maximum information coefficient between each initial association feature and the target evaluation index is determined, and it is determined whether the maximum information coefficient is greater than or equal to the target maximum information coefficient threshold. If so, the initial association feature is determined as the target association feature. The target evaluation index includes the prediction error corresponding to the target power prediction task and the early warning accuracy corresponding to the target equipment fault early warning task. The target maximum information coefficient threshold includes the first maximum information coefficient threshold corresponding to the target power prediction task and the second maximum information coefficient threshold corresponding to the target equipment fault early warning task.

8. A data processing device at a power source, characterized in that, include: The data acquisition module is used to acquire initial data from the power source and preprocess the initial data to obtain target data. The types of target data include environmental parameters, equipment status parameters, and electrical parameters; The time window determination module is used to generate a target irradiance change rate based on the target data, and to determine a target time window based on the target irradiance change rate. The correlation feature generation module is used to determine the target statistical features and target fluctuation features corresponding to the target data based on the target time window, and generate several initial correlation features corresponding to the target data based on the target statistical features and the target fluctuation features; The initial association feature is a feature used to represent the association between different types of target data; The association feature filtering module is used to determine the maximum information coefficient between each of the initial association features and the target power task, and to filter out the target association features from each of the initial association features based on the maximum information coefficient, so as to process the target power task based on the target association features.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the data processing method at the power source end as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the data processing method at the power source end as described in any one of claims 1 to 7.

Images