Zero-carbon equipment fault early warning method and system based on ESG scheduling

Abstract

The invention relates to the technical field of zero-carbon equipment fault early warning, and discloses a zero-carbon equipment fault early warning method and system based on ESG scheduling, and the method comprises the steps: collecting multi-source ESG parameter data, such as carbon footprint intensity, clean energy consumption and energy storage system charging and discharging efficiency, extracting equipment state features through employing a graph convolution network in combination with a gating circulation unit, and carrying out the early warning of the zero-carbon equipment fault; and calculating an ESG correlation degree and an inter-time-segment cross-correlation matrix, and generating a fault early warning weight through a ReLU activation function after principal component analysis dimension reduction. And dynamically adjusting the reference threshold based on the weight, correcting the deviation by means of an extended Kalman filter, and generating an early warning parameter to trigger an early warning process. Meanwhile, a multi-dimensional equipment state space is constructed to judge a fault state, a joint optimization model of ESG parameters and an early warning threshold value is solved through a non-dominated sorting genetic algorithm, collaborative optimization of ESG benefits and fault early warning is achieved, and early warning accuracy and dynamic adaptability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of zero-carbon equipment fault warning, and specifically to a zero-carbon equipment fault warning method and system based on ESG scheduling. Background Art

[0002] In the context of the global strong promotion of the carbon neutrality goal, zero-carbon equipment, as a key carrier for realizing energy transformation and sustainable development, its safe and stable operation is of crucial importance. With the wide application of zero-carbon equipment in the energy system, traditional fault warning methods are facing numerous challenges.

[0003] Most traditional fault warnings only focus on the operating parameters of the equipment itself, such as physical indicators like temperature, pressure, and vibration, while ignoring the associations of zero-carbon equipment with aspects such as the environment, society, and governance during operation. The operation of zero-carbon equipment not only involves the performance of the equipment itself but is also closely related to ESG parameters such as carbon footprint intensity, clean energy consumption, and charge-discharge efficiency of the energy storage system. These ESG parameters can reflect the performance of the equipment in terms of environmental sustainability, social responsibility fulfillment, and governance structure, and have important reference value for equipment fault warning.

[0004] In addition, zero-carbon equipment is usually in a complex energy network, and the mutual associations and influences between equipment are more significant. Traditional methods are difficult to effectively capture such associations between equipment, resulting in insufficient accuracy and timeliness of warning. At the same time, the operating conditions of zero-carbon equipment are complex and changeable, affected by various factors such as energy supply, load demand, and weather conditions. The traditional fixed-threshold warning method cannot adapt to the needs of such dynamic changes.

[0005] With the intelligent and digital development of the energy system, the operating data generated by zero-carbon equipment shows the characteristics of multi-source, massive, and high-dimensional. How to extract effective feature information from these complex data to achieve accurate warning of equipment faults has become an urgent problem to be solved. Existing data processing and analysis methods have problems such as incomplete feature extraction and low information utilization rate when dealing with high-dimensional, multi-source heterogeneous data, and are difficult to meet the needs of zero-carbon equipment fault warning.

[0006] In addition, during the operation of zero-carbon equipment, it is necessary to achieve collaborative optimization between ESG benefits and equipment fault warning. Traditional methods often separate the two and cannot maximize ESG benefits while ensuring the safe operation of the equipment. Therefore, there is an urgent need for a fault warning method that can comprehensively consider ESG parameters, equipment associations, and dynamic operating conditions to improve the safety and sustainability of zero-carbon equipment operation. Summary of the Invention

[0007] The purpose of the present invention is to provide a zero-carbon equipment fault warning method and system based on ESG scheduling to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: A zero-carbon equipment fault warning method based on ESG scheduling, the method includes: Collect multi-source ESG parameter data during the operation of zero-carbon equipment, where the ESG parameter data includes carbon footprint intensity, clean energy consumption, and charge-discharge efficiency of the energy storage system; Use a graph convolutional network to extract equipment state features from the ESG parameter data to obtain a state feature matrix, and calculate the ESG correlation degree according to the covariance change of the state feature matrix; Perform feature dimensionality reduction on the cross-correlation matrix of ESG correlation degrees at different time periods, extract core feature vectors, and perform non-linear mapping on the core feature vectors through the ReLU activation function to obtain the fault warning weights for each time period; Dynamically adjust the benchmark threshold of the state feature matrix according to the fault warning weights to generate warning parameters, and trigger the equipment fault warning process based on the warning parameters.

[0009] Preferably, the equipment state feature extraction includes: Intercept the ESG parameter data at a set sampling interval, construct a state tensor and perform standardization processing; Use a multi-head graph convolutional layer to perform cross-device feature aggregation on the state tensor, and output the dot product sum of the adjacency matrix and the feature vector; Input the aggregation result into a gated recurrent unit, extract multi-time scale features through the update gate and the reset gate, and screen the feature dimensions through an attention mask to generate a state feature matrix.

[0010] Preferably, the calculation of the ESG correlation degree includes: Select the state feature matrix of any time period as the benchmark threshold, and calculate the covariance value of the feature vectors of its adjacent k time periods; Take the ratio of the mean of the covariance values to the Euclidean distance of the benchmark threshold as the in-period correlation quantity; Calculate the cross-correlation coefficient between the benchmark threshold and the features of the correlation period, and take its absolute value as the inter-period correlation quantity; Take the harmonic mean of the in-period correlation quantity and the inter-period correlation quantity as the ESG correlation degree.

[0011] Preferably, the extraction of the core feature vectors includes: Construct a cross-correlation matrix of ESG correlation degrees at different time periods, and perform principal component analysis on it to obtain orthogonal feature vectors; Select the feature vectors whose principal component variance contribution rate exceeds the set threshold to form the core feature subspace; Map the mutual correlation matrix to the core feature subspace to obtain the core feature vectors after dimensionality reduction.

[0012] Preferably, obtaining the fault warning weights for each time period includes: Normalize the core feature vectors and calculate the cosine similarity between them and the preset reference vector; Input the similarity values into a bidirectional long short-term memory network, and generate initial weights after transformation by the forward layer and the backward layer; Smooth the initial weights through moving average filtering and output the fault warning weights for each time period.

[0013] Preferably, generating the warning parameter generation includes: Perform matrix dot multiplication on the fault warning weights and the benchmark threshold to obtain the weight adjustment parameter; Calculate the deviation between the weight adjustment parameter and the benchmark threshold, and dynamically correct the deviation through an extended Kalman filter; Superimpose the corrected deviation on the benchmark threshold to generate the warning parameter.

[0014] Preferably, the system further includes: Construct a multi-dimensional equipment state space based on the warning parameters, and extract the extreme points and covariance mutation regions of the space; When the density of extreme points exceeds the set threshold or the span of the covariance mutation region is greater than the limit value, it is determined as the fault warning state, and multi-level warning instructions are generated.

[0015] Preferably, the construction of the multi-dimensional equipment state space includes: Map the warning parameters to a multi-dimensional coordinate system according to the operation time sequence to generate a state distribution point set; Use the kernel density estimation method to perform spatial reconstruction on the state distribution point set, and calculate the covariance gradient and kurtosis coefficient of the reconstructed space; Suppress noise in the gradient distribution through the Gaussian filtering algorithm to eliminate random interference.

[0016] Preferably, the system further includes: Establish a joint optimization model for ESG parameter data and equipment warning thresholds, and use the non-dominated sorting genetic algorithm to solve the optimal warning strategy; Couple the optimal warning strategy with the warning parameters in real time to generate a fault warning plan that maximizes the ESG benefits; The solution of the joint optimization model includes: Define the objective function as the weighted absolute value sum of the ESG index deviation and the warning omission rate, and the constraint condition is the safe operation range of the equipment; Perform Pareto optimal transformation on the objective function and decompose it into a warning threshold sub-problem and an ESG scheduling sub-problem; Alternately iterate to solve two sub - problems until convergence, and output the optimal early - warning strategy that meets the constraint conditions.

[0017] Preferably, the present invention further includes a zero - carbon equipment fault early - warning system based on ESG scheduling. The system includes: A multi - source data acquisition module, which is used to obtain the carbon footprint intensity, clean energy consumption, and energy storage system charge - discharge efficiency data during the operation in real - time, and construct a multi - dimensional ESG parameter time - series matrix; A state feature extraction module, which uses a graph convolutional network to perform cross - device feature aggregation on the multi - dimensional ESG parameter time - series matrix, captures the device state features at multiple time scales through a gated recurrent unit, and outputs a state feature matrix; An ESG correlation calculation module, which generates the intra - period correlation quantity and the inter - period correlation quantity according to the covariance change of the state feature matrix, and calculates their harmonic mean as the ESG correlation degree; A cross - correlation matrix decomposition module, which constructs a cross - correlation matrix for the ESG correlation degrees in different periods, and extracts the core eigenvectors through principal component analysis; A weight mapping module, which uses the ReLU activation function to perform non - linear mapping on the core eigenvectors to generate the fault early - warning weights for each period; A parameter optimization module, which dynamically adjusts the benchmark threshold based on the fault early - warning weights, and corrects the parameter deviation amount through an extended Kalman filter to generate early - warning parameters; A real - time early - warning module, which converts the early - warning parameters into equipment fault early - warning signals, and transmits them to the monitoring platform through the industrial Internet to trigger the early - warning instruction.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This method collects multi - source ESG parameter data such as the carbon footprint intensity, clean energy consumption, and energy storage system charge - discharge efficiency during the operation of zero - carbon equipment, breaking through the limitation of traditional fault early - warning that only focuses on equipment physical parameters, incorporating environmental, social, and governance factors into the early - warning system, making the early - warning more comprehensive and scientific, and being able to more accurately reflect the actual operation status and potential risks of zero - carbon equipment.

[0019] In terms of data processing, using a graph convolutional network to extract device state features from ESG parameter data can effectively capture the correlation between devices. Through cross - device feature aggregation, more comprehensive device state information can be obtained. Combining with a gated recurrent unit to extract features at multiple time scales, and using an attention mask to screen the feature dimensions to generate a state feature matrix, greatly improves the efficiency and accuracy of feature extraction, laying a solid foundation for subsequent fault early - warning.

[0020] By calculating the ESG correlation degree and comprehensively considering the correlation quantity within a time period and the correlation quantity between time periods, the internal relationship between ESG parameters in different time periods can be deeply analyzed, and the changing trend of the equipment operation state can be better grasped. Feature dimensionality reduction is performed on the cross-correlation matrix, the core eigenvectors are extracted, and a non-linear mapping is performed through the ReLU activation function to obtain the fault warning weight, realizing the dynamic adjustment of the warning weight and making the warning more in line with the actual operation of the equipment.

[0021] The reference threshold is dynamically adjusted based on the fault warning weight, and the extended Kalman filter is used to correct the deviation quantity to generate warning parameters, which can adapt to the dynamic changes of the zero-carbon equipment operation conditions and improve the accuracy and timeliness of the warning. A multi-dimensional equipment state space is constructed, and the fault warning state is determined by analyzing the extreme point density and the covariance mutation region, which can more comprehensively capture the abnormal state of the equipment and reduce the phenomena of missed alarms and false alarms.

[0022] A joint optimization model of ESG parameter data and equipment warning threshold is established, and the non-dominated sorting genetic algorithm is used to solve the optimal warning strategy, realizing the collaborative optimization of ESG benefits and equipment fault warning. While ensuring the safe operation of the equipment, the ESG benefits are maximized, promoting the sustainable development of zero-carbon equipment.

[0023] The multi-source data acquisition module of the system can obtain the multi-dimensional ESG parameter time series matrix in real time, providing rich data support for the entire warning system; each functional module such as the state feature extraction module and the ESG correlation calculation module cooperates with each other, realizing the whole process automation processing from data acquisition, feature extraction, correlation calculation to warning generation, improving the efficiency and reliability of the warning. The warning signal is transmitted to the monitoring platform through the industrial Internet, realizing the real-time warning and rapid response of faults, providing a strong guarantee for the safe and stable operation of zero-carbon equipment. Description of the Drawings

[0024] Figure 1 It is the working principle diagram of the zero-carbon equipment fault warning method based on ESG scheduling described in the present invention; Figure 2 It is the process diagram of equipment state feature extraction; Figure 3 It is the flow chart of ESG correlation degree calculation; Figure 4 It is the flow chart of core eigenvector extraction. Detailed Embodiments

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] Please refer to Figures 1-4 , a zero-carbon device fault warning method based on ESG scheduling involved in the present invention, and the specific implementation steps are as follows: Collect multi-source ESG parameter data during the operation of zero-carbon devices, where the ESG parameter data includes carbon footprint intensity, clean energy consumption, and charge-discharge efficiency of energy storage systems.

[0027] Use a graph convolutional network to extract device state features from ESG parameter data to obtain a state feature matrix, and calculate the ESG correlation degree according to the covariance change of the state feature matrix.

[0028] Perform feature dimensionality reduction on the cross-correlation matrix of ESG correlation degrees at different time periods, extract core feature vectors, and perform non-linear mapping on the core feature vectors through the ReLU activation function to obtain the fault warning weights for each time period.

[0029] Dynamically adjust the benchmark threshold of the state feature matrix according to the fault warning weights to generate warning parameters, and trigger the device fault warning process based on the warning parameters.

[0030] Embodiment 1: When extracting device state features, intercept ESG parameter data at a set sampling interval. The set sampling interval here needs to comprehensively consider factors such as the type of device, operating conditions, and actual application scenarios. For example, for zero-carbon devices such as wind turbines that are greatly affected by environmental factors and have obvious fluctuations in operating states, in order to be able to capture the subtle changes in their operating states in a timely manner, the sampling interval can be set to a relatively short time, such as once every minute or every five minutes; for some relatively stable energy storage system devices in operation, the sampling interval can be appropriately extended, such as once every ten minutes or fifteen minutes, which can not only ensure obtaining enough data to reflect the device state but also reduce the data acquisition volume and the burden of subsequent processing. After intercepting the data, a state tensor needs to be constructed. The construction method of the state tensor is related to the dimension of the collected ESG parameter data. The three types of parameter data, namely carbon footprint intensity, clean energy consumption, and charge-discharge efficiency of energy storage systems, will be organized into a tensor structure with a specific dimension to facilitate subsequent processing and analysis.

[0031] After constructing the state tensor, it needs to be normalized. The purpose of normalization is to eliminate the differences in dimension and numerical range between different parameters, so that each parameter has the same importance in the subsequent feature extraction process. Specifically, normalization usually involves subtracting the mean of each parameter data and then dividing it by its standard deviation. After such processing, the mean of the data is 0 and the standard deviation is 1, thus achieving data normalization. For example, the unit of carbon footprint intensity may be kilograms of carbon dioxide per kilowatt-hour, while the unit of clean energy consumption may be kilowatt-hours. The dimension and numerical range of the two are quite different, and through normalization, they can be converted into comparable values.

[0032] After completing the normalization process, a multi-head graph convolutional layer is used to perform cross-device feature aggregation on the state tensor. The design of the multi-head graph convolutional layer takes into account the possible complex connection relationships and mutual influences between zero-carbon devices. In practical applications, a zero-carbon device system may include multiple different types of devices, such as wind turbines, solar panels, energy storage batteries, etc. These devices are interconnected through the power grid or other physical connections, forming a complex network structure. Each "head" in the multi-head graph convolutional layer can be regarded as extracting the correlation features between devices from different perspectives. Through the parallel processing of multiple "heads", various feature relationships in the device network can be captured more comprehensively. When performing cross-device feature aggregation, first, an adjacency matrix needs to be constructed. The adjacency matrix is used to represent the connection relationships between devices, and the element values in the matrix represent the correlation strength between devices. Then, the state tensor is operated with the adjacency matrix, and the dot product sum of the adjacency matrix and the eigenvector is output. This dot product sum contains the features of the device itself and the feature information of the mutual influence between devices.

[0033] The aggregation result of the multi-head graph convolutional layer is input into the gated recurrent unit (GRU). The gated recurrent unit is a special type of recurrent neural network that can effectively process time series data and capture the time-dependent relationships in the data. In the gated recurrent unit, there are two important structures: the update gate and the reset gate. The role of the update gate is to control how much of the previous moment's state information can be passed to the current moment, and the reset gate is used to control the degree of ignoring the previous moment's state information. Through the coordinated action of these two gates, the extraction of multi-time scale features can be achieved. For example, for short-term device state changes, the update gate will allow more current moment information to pass through, while for long-term trend features, the update gate will retain more historical state information.

[0034] After extracting multi-time scale features through a gated recurrent unit, it is also necessary to screen the feature dimensions through an attention mask. The role of the attention mask is to focus on those feature dimensions that are more important for device status analysis and suppress unimportant feature dimensions. In actual device operation data, the contribution degrees of different feature dimensions to device fault warning are different. Some dimensions may contain a large amount of noise information, while some dimensions directly reflect the key status of the device. Through the attention mask, a weight can be assigned to each feature dimension. A dimension with a larger weight indicates a higher importance and will be retained, while a dimension with a smaller weight will be weakened or ignored. After being screened by the attention mask, a state feature matrix is finally generated. This matrix contains the device state feature information after cross-device feature aggregation, multi-time scale feature extraction, and feature dimension screening, and can more accurately reflect the actual operation state of the device, providing a reliable basis for subsequent ESG correlation calculation and fault warning.

[0035] During the entire process of device state feature extraction, each step is closely connected. The processing result of the previous step is the input of the next step. By reasonably designing the processing methods and parameters of each step, key features reflecting the device state can be effectively extracted from multi-source ESG parameter data, laying a solid foundation for subsequent fault warning analysis. For example, the reasonable setting of the sampling interval ensures the timeliness and effectiveness of the data, the normalization process guarantees the comparability of the data, the multi-head graph convolutional layer realizes the comprehensive aggregation of cross-device features, the gated recurrent unit captures multi-time scale features in the time series, and the attention mask screens out important feature dimensions. The collaborative work of these steps enables the generated state feature matrix to accurately and comprehensively describe the operation state of the device.

[0036] Embodiment 2: When calculating the ESG correlation, select the state feature matrix of any time period as the benchmark threshold. The selection of this time period needs to combine the device operation cycle and historical data rules. For example, a certain time period of continuous and stable operation under the normal operation state of the device can be selected, such as the period when the operation parameters tend to be stable after the device starts, or a typical operation period can be selected according to the device maintenance cycle. The determination of the benchmark threshold needs to be representative and can reflect the state feature distribution of the device under normal working conditions.

[0037] Calculate the covariance value of the eigenvectors of the k time periods adjacent to the benchmark threshold. The selection of the adjacent k time periods needs to consider the time correlation of the equipment state changes. The value of k can be adjusted according to the equipment type and operating characteristics. For example, for energy storage system equipment with a relatively fast response speed, the value of k can be set to 5 - 10 time periods to capture the characteristic changes within a short time; for equipment such as wind turbines that are greatly affected by the environment and have relatively slow state changes, the value of k can be set to 15 - 20 time periods to cover longer-term characteristic fluctuations. The calculation of the covariance value is for the correlation between eigenvectors, reflecting the degree of co-variation of different characteristics in adjacent time periods. For example, whether the fluctuations of the carbon footprint intensity and the charge-discharge efficiency of the energy storage system are consistent in adjacent time periods.

[0038] Take the ratio of the Euclidean distance between the mean value of the covariance value and the benchmark threshold as the within-period correlation quantity. The mean value of the covariance value reflects the average level of the correlation of the eigenvectors in the adjacent k time periods, and the Euclidean distance is used to measure the difference amplitude between this mean value and the benchmark threshold. Specifically, first calculate the arithmetic mean of the k covariance values, then calculate the Euclidean distance between this mean value and the corresponding eigenvector in the benchmark threshold, that is, the square root of the sum of the squares of the differences in each dimension. Finally, divide the Euclidean distance by the covariance mean to obtain a dimensionless within-period correlation quantity. This correlation quantity reflects the degree of deviation of the eigenvector correlation in the current time period from the benchmark state. The larger the value, the more significant the difference between the characteristic changes in the time period and the benchmark threshold.

[0039] Then calculate the cross-correlation coefficient between the benchmark threshold and the characteristics of the associated time period, and take its absolute value as the between-period correlation quantity. The selection of the characteristics of the associated time period can include the eigenvectors of the previous time period, the next time period, or the eigenvectors of multiple time periods before and after, which is determined according to the time lag of the equipment state transmission. The calculation of the cross-correlation coefficient is used to measure the similarity between the benchmark threshold and the characteristics of the associated time period in the time series. Specifically, it is obtained by calculating the ratio of the covariance of two eigenvectors to the product of their respective standard deviations, and the value range is between [-1, 1]. After taking the absolute value, the direction difference of positive and negative correlations can be eliminated, and only the quantitative value of the similarity degree is retained. For example, if the change trend of a certain eigenvector of the benchmark threshold is consistent with that of the eigenvector of the next time period, the absolute value of the cross-correlation coefficient is close to 1, indicating a close between-period correlation.

[0040] Take the harmonic mean of the within-period correlation quantity and the between-period correlation quantity as the ESG correlation degree. The calculation method of the harmonic mean is 2 times the product of the two numbers divided by the sum of the two numbers. This calculation method can avoid the dominant influence of an overly large or small value of a certain correlation quantity on the result, and more evenly reflect the correlation degree within and between time periods. For example, when the within-period correlation quantity is large but the between-period correlation quantity is small, the harmonic mean will tend to the smaller value, comprehensively reflecting the overall correlation level.

[0041] In actual operation, attention should be paid to the time alignment of data and the consistency of feature dimensions. The collected ESG parameter data needs to be arranged in strict chronological order to ensure the time continuity of feature vectors in adjacent time periods and avoid calculation biases in correlation due to timestamp errors. At the same time, the benchmark threshold and the dimensions of feature vectors in each time period need to be consistent, and the feature dimensions corresponding to parameters such as carbon footprint intensity, clean energy consumption, and charge-discharge efficiency of energy storage systems need to correspond one by one to ensure the accuracy of covariance, Euclidean distance, and cross-correlation coefficient calculations.

[0042] In addition, the calculation of ESG correlation needs to consider the dynamic changes in the equipment operation scenario. When the equipment switches from the normal operation state to the variable load condition or is affected by external environmental disturbances, the benchmark threshold needs to be dynamically updated according to real-time operation data to adapt to the new operation state. For example, when the grid load demand suddenly increases and the energy storage system switches from the charging state to the discharging state, it is necessary to reselect the state feature matrix under this condition as the benchmark threshold to avoid calculation distortion of the correlation due to the mismatch between the benchmark threshold and the current operation state.

[0043] The entire calculation process of ESG correlation realizes the quantitative analysis of the correlation of equipment state features in the time series through the fusion of multi-dimensional correlation quantities. The correlation quantity within a time period captures the difference between the feature changes in the current time period and the benchmark state, and the correlation quantity between time periods reflects the transfer relationship of features in different time periods. The harmonic mean of the two comprehensively reflects the overall correlation degree of ESG parameter data, providing a quantitative basis for subsequent feature dimension reduction and fault warning weight calculation. In specific implementation, the parameters of each step need to be finely adjusted according to the equipment type and application scenario, such as the setting of the k value and the selection of the correlation time period, to ensure that the ESG correlation can accurately reflect the actual correlation characteristics of the equipment state.

[0044] Example 3: When extracting the core feature vector, it is necessary to construct a cross-correlation matrix of ESG correlations for different time periods. These different time periods can be multiple consecutive time periods arranged in chronological order. For example, a day can be divided into 24 hours, with each hour as a time period, or the time periods can be divided according to the operation cycle of the equipment. Taking the energy storage system in a zero-carbon park as an example, assuming that the system operates for 20 hours every day and shuts down for 4 hours for maintenance, then the daily operation time can be divided into 20 time periods, with each time period being 1 hour. After calculating the ESG correlation for each time period, arrange these correlation values in the order of time periods to construct a cross-correlation matrix. The elements in the cross-correlation matrix represent the correlation between ESG correlations in different time periods. For example, the element in the i-th row and j-th column of the matrix represents the degree of correlation between the ESG correlation in the i-th time period and the j-th time period.

[0045] After constructing the cross-correlation matrix, it is necessary to perform principal component analysis on it to obtain orthogonal eigenvectors. Principal component analysis is a dimensionality reduction technique that can transform high-dimensional data into low-dimensional principal components while retaining as much information as possible from the original data. In this process, first, the eigenvalues and eigenvectors of the cross-correlation matrix need to be calculated, and then the eigenvectors are orthogonalized so that they are perpendicular to each other. These orthogonal eigenvectors represent the main directions of data variation, and each eigenvector corresponds to an eigenvalue. The magnitude of the eigenvalue indicates the amount of information contained in that eigenvector.

[0046] It is necessary to select the eigenvectors whose principal component variance contribution rate exceeds the set threshold to form the core feature subspace. The choice of the set threshold needs to be determined according to the actual application scenario and the requirements for data dimensionality reduction. For example, the set threshold can be 80%, which means that we only select those eigenvectors whose sum of eigenvalues accounts for 80% of the sum of all eigenvalues. The subspace formed by these eigenvectors is called the core feature subspace. Taking the energy storage system mentioned earlier as an example, assume that the cross-correlation matrix is a 20×20 matrix. After principal component analysis, 20 eigenvectors and their corresponding eigenvalues are obtained. If the sum of the variance contribution rates of the first 5 eigenvectors exceeds 80%, then these 5 eigenvectors are selected to form the core feature subspace.

[0047] Map the cross-correlation matrix to the core feature subspace to obtain the core feature vectors after dimensionality reduction. The mapping process is actually to project the original cross-correlation matrix onto the core feature subspace, so that high-dimensional data is transformed into low-dimensional data. Specifically, for the ESG correlation degree vector of each time period, perform a dot product operation with each eigenvector in the core feature subspace to obtain the coordinate values of that time period on the core feature subspace. These coordinate values form the core feature vectors after dimensionality reduction. For example, the original ESG correlation degree vector of each time period is 20-dimensional. After mapping to a 5-dimensional core feature subspace, the core feature vector of each time period becomes 5-dimensional.

[0048] In actual operation, attention needs to be paid to data preprocessing and the selection of eigenvectors. First, before constructing the cross-correlation matrix, it is necessary to standardize the ESG correlation degree data to eliminate the differences in data dimensions and value ranges of different time periods. Second, when selecting eigenvectors, it is necessary to comprehensively consider the variance contribution rate and the physical meaning of the eigenvectors. Some eigenvectors may have a high variance contribution rate but may not have practical physical meaning. In this case, it is necessary to make a trade-off and selection according to the specific situation.

[0049] In addition, the extraction of the core feature vector also needs to consider the influence of the device's operating state and environmental factors. For example, when the device is in different operating modes, the distribution of the ESG correlation degree may change. At this time, it is necessary to reconstruct the cross-correlation matrix and perform principal component analysis to ensure that the core feature vector can accurately reflect the current operating state of the device. At the same time, changes in environmental factors such as temperature and humidity may also affect the calculation results of the ESG correlation degree, and thus affect the extraction of the core feature vector. Therefore, in practical applications, these environmental factors need to be appropriately processed and compensated.

[0050] The entire process of extracting the core feature vector, through steps such as constructing the cross-correlation matrix, principal component analysis, feature vector selection, and mapping, realizes the dimensionality reduction processing of the ESG correlation degree data, retains the main features of the data, reduces the dimension of the data at the same time, and improves the efficiency and accuracy of subsequent processing. In specific implementation, it is necessary to reasonably select parameters such as time period division and threshold setting according to the type, operating characteristics, and application scenarios of the device to ensure that the core feature vector can accurately reflect the state characteristics of the device and provide a reliable basis for subsequent fault warning weight calculation and fault warning. For example, for zero-carbon devices such as wind turbines that are greatly affected by weather, the time period division can be finer and the threshold setting can be appropriately increased to capture more feature changes; while for relatively stable energy storage systems, the time period division can be relatively rough and the threshold setting can be appropriately reduced to reduce the calculation amount.

[0051] Example 4: When obtaining the fault warning weights for each time period, it is first necessary to normalize the core feature vector and calculate its cosine similarity with the preset reference vector. Taking the energy storage system of a certain photovoltaic power station as an example, assume that the core feature vector for a certain time period extracted through the steps of Example 3 is a numerical sequence with 5 dimensions, such as [0.23, 0.41, -0.17, 0.35, 0.12]. During normalization, the vector needs to be converted into a unit vector, that is, by calculating the square root of the sum of the squares of the numerical values of each dimension, and then dividing the numerical value of each dimension by this square root to make the vector length 1. Assume that after normalization, [0.37, 0.66, -0.27, 0.56, 0.19] is obtained. The preset reference vector can be determined according to the statistical mean of the historical core feature vectors during normal operation of the equipment. For example, by analyzing the core feature vectors of this energy storage system under normal conditions in the past month, the calculated reference vector is [0.30, 0.50, 0.10, 0.40, 0.20]. At this time, the calculation of the cosine similarity needs to be based on the dot product of the two unit vectors, that is, the sum of the products of the corresponding numerical values of each dimension. The closer the result is to 1, the more consistent the two directions are. For example, the calculated cosine similarity between the core feature vector of this time period and the reference vector is 0.37×0.30 + 0.66×0.50 + (-0.27)×0.10 + 0.56×0.40 + 0.19×0.20, and the specific value needs to be obtained according to actual calculation. This value reflects the degree of deviation of the current feature from the normal state.

[0052] Next, input the similarity value into a bidirectional long short-term memory network (Bi-LSTM), and generate an initial weight after transformation by the forward layer and the backward layer. The structure of the Bi-LSTM network needs to consider the bidirectionality of the time series. For example, set two hidden layers, each layer containing 50 neurons, to process the similarity sequence of consecutive time periods. Still taking the energy storage system of the photovoltaic power station as an example, assume that the similarity value calculated currently is for the 10th time period. The network will receive the similarity sequence from the 1st to the 10th time periods simultaneously. The forward layer processes from the 1st time period to the 10th time period to capture the feature trends of past time periods, and the backward layer processes from the 10th time period to the 1st time period to capture the feature dependencies of future time periods (backward derivation based on historical data). The forget gate, input gate, and output gate inside the network will adjust the weights according to the input data. For example, when the similarity value of a certain time period suddenly decreases (indicating that the feature deviates from the normal state), the forward layer will enhance the influence of this time period on subsequent weights, and the backward layer will verify whether this deviation belongs to abnormal fluctuations through historical data. After the forward and backward calculations of the network, an initial weight value between 0 and 1 is output, and this value synthesizes the dynamic feature changes in the time series.

[0053] Then, the initial weights are smoothed by moving average filtering, and the fault warning weights for each period are output. The size of the moving average window needs to be set according to the frequency of equipment state changes. For example, for an energy storage system, the window size can be set to 5, that is, the arithmetic mean of the initial weights for the current period and the previous 4 periods is calculated. Suppose the initial weight for the 10th period is 0.82, and the initial weights for the previous 4 periods are 0.75, 0.78, 0.80, and 0.81 respectively. Then the weight after moving average is (0.82 + 0.75 + 0.78 + 0.80 + 0.81) ÷ 5, and the specific value is obtained according to actual calculation. The smoothing process can eliminate the sudden change of weights caused by short-term data fluctuations. For example, when the similarity value suddenly changes due to a short-term abnormality of the sensor in a certain period, the moving average filtering can weaken the influence of this abnormal value through the mean value of historical weights, making the warning weight more in line with the actual state trend of the equipment.

[0054] In practical applications, the update mechanism of the preset reference vector is particularly important. For example, when the energy storage system of a photovoltaic power station switches from the daytime charging mode to the nighttime discharging mode, the ESG parameter characteristics of the equipment will change significantly. At this time, the reference vector needs to be recalculated based on the historical core feature vectors under the new working conditions. In specific operations, the reference vector can be set to be updated once an hour, and adjusted based on the sliding window mean value of the core feature vectors within the past 1 hour to ensure that the reference vector matches the current operating mode. In addition, the parameter training of the Bi-LSTM network needs to be based on the historical operation data of the equipment. For example, collect the core feature vectors and corresponding fault labels of this energy storage system in different states such as normal, minor fault, and severe fault, and optimize the network weights through the backpropagation algorithm, so that the network can more accurately convert the similarity sequence into the initial weight reflecting the fault probability.

[0055] The input sequence length of the bidirectional long short-term memory network also needs to be set according to the fault latency period of the equipment. For faults such as the capacity attenuation of energy storage batteries that develop slowly, the latency period may last for several days. Therefore, the input sequence can be set to include the daily core feature vector similarity values for the past 7 days; for faults such as overheating of the power module of an inverter that develop rapidly, the latency period may be only a few hours, and the input sequence can be shortened to the period data for the past 12 hours. This dynamic adjustment mechanism can enable the network to better capture the time characteristics of different types of faults.

[0056] The window size of the moving average filtering also needs to be combined with the sensitivity requirements of the warning. If you want to detect sudden faults in a timely manner, the window size can be reduced (such as set to 3) to make the weight more sensitive to recent changes; if you are more concerned about the long-term trend to reduce false alarms, the window size can be increased (such as set to 7). For example, during the high-temperature period in summer, the charge and discharge efficiency of the energy storage system may show a short-term decline due to environmental temperature fluctuations. A larger window size can avoid misjudging such normal fluctuations as fault warning signals.

[0057] In the process of generating the entire fault warning weight, through similarity calculation, two-way time series feature extraction, and smoothing processing, the dynamic quantification of device state features is realized. Taking the energy storage system of a photovoltaic power station as an example, when the cosine similarity between the core feature vector and the reference vector in a certain period decreases, it indicates that the ESG parameter features deviate from the normal state. The Bi-LSTM network will combine the historical and future feature trends to judge whether the deviation is persistent. If it is confirmed that the deviation is continuous and exceeds the smoothed weight threshold, a higher warning weight will be generated, providing a basis for subsequent warning parameter adjustment. In specific implementation, the parameters of each step need to be finely adjusted according to the device type, such as the update frequency of the reference vector, the number of hidden layer neurons of the Bi-LSTM, the window size of the moving average, etc., to ensure that the warning weight can accurately reflect the fault risk and adapt to the dynamic operating conditions of the device.

[0058] Example 5: When generating warning parameters, the fault warning weight needs to be multiplied by the reference threshold matrix point by point to obtain the weight adjustment parameter. Taking the energy storage converter of a wind farm as an example, assume that the fault warning weight for a certain period obtained through the steps of Example 4 is a 3×1 matrix, such as [0.85, 1.2, 0.9], corresponding to the warning weights of carbon footprint intensity, clean energy consumption, and charge and discharge efficiency of the energy storage system respectively. The reference threshold is set according to the historical data during the normal operation of the device. For example, the reference threshold for carbon footprint intensity is , the reference threshold for clean energy consumption is 800 kWh, and the reference threshold for the charge and discharge efficiency of the energy storage system is 0.9. When performing matrix point multiplication, the weight is multiplied by each reference threshold correspondingly to obtain the weight adjustment parameter, that is, the adjustment parameter for carbon footprint intensity is , the adjustment parameter for clean energy consumption is 800×1.2 = 960 kWh, and the adjustment parameter for the charge and discharge efficiency of the energy storage system is 0.9×0.9 = 0.81. This process dynamically scales the reference threshold through the weight, enabling the threshold to be adjusted according to the deviation degree of the current device state features.

[0059] Calculate the deviation between the weight adjustment parameter and the reference threshold, and dynamically correct the deviation through an extended Kalman filter. The calculation of the deviation is the weight adjustment parameter minus the reference threshold. For example, the deviation of carbon footprint intensity is , the deviation of clean energy consumption is 960 - 800 = 160 kWh, and the deviation of the charge-discharge efficiency of the energy storage system is 0.81 - 0.9 = -0.09. The Extended Kalman Filter is used to handle state estimation in nonlinear systems, and it requires establishing the state equation and observation equation of the system. Taking the operating state of the energy storage converter as an example, the state equation can describe the dynamic changes of internal parameters of the device, such as the degree of capacitor aging, the temperature of power devices, etc., and the observation equation correlates these internal states with the measurable deviation of ESG parameters. The filter will calculate the optimal estimate of the current state based on the state estimate at the previous moment and the current observation value, so as to correct the deviation. For example, when it is detected that the deviation of clean energy consumption suddenly increases, the filter will combine the historical operation data of the device and the current operating conditions to determine whether the deviation is caused by a device failure or external factors such as grid load fluctuations, and correct the deviation accordingly.

[0060] The corrected deviation is superimposed on the reference threshold to generate a warning parameter. For example, the warning parameter for the carbon footprint intensity is 0.5 + (-0.075 corrected value). Assuming the corrected deviation is -0.05, the warning parameter is ; the warning parameter for clean energy consumption is 800 + 160 corrected value. Assuming the corrected deviation is 120, the warning parameter is 920 kWh; the warning parameter for the charge-discharge efficiency of the energy storage system is 0.9 + (-0.09 corrected value). Assuming the corrected deviation is -0.07, the warning parameter is 0.83. These warning parameters are obtained by combining the fault warning weight of the current device state and the corrected deviation on the basis of the reference threshold, and can more accurately reflect the current fault risk degree of the device.

[0061] In actual operation, the setting of the reference threshold needs to be based on a large amount of historical operation data of the device. For example, for the energy storage converter of a wind farm, it is necessary to collect its normal operation data under different seasons and different load conditions, and determine the reference threshold range of each ESG parameter through statistical analysis. At the same time, the aging factor of the device needs to be considered. As the operation time of the device increases, the reference threshold may drift, so it is necessary to update the reference threshold regularly.

[0062] The parameter adjustment of the Extended Kalman Filter is also crucial. The process noise covariance and measurement noise covariance of the filter need to be set according to the actual operation of the device. For example, for an energy storage converter with a relatively stable operating state, the process noise covariance can be set smaller, while for a device that is greatly affected by the external environment, such as a wind turbine, the process noise covariance needs to be set larger to adapt to its larger state changes. In addition, the initial state estimate of the filter also needs to be as accurate as possible. Usually, the initial operating state of the device can be used as the initial state of the filter.

[0063] After generating the warning parameters, it is also necessary to compare them with the actual operating data of the device in real time. When the actual operating data exceeds the warning parameters, the fault warning process is triggered. For example, when the actual measured value of the carbon footprint intensity of the energy storage converter reaches the above, the system will issue a corresponding warning signal.

[0064] In the entire process of generating the warning parameters, through steps such as weight adjustment, deviation correction, and threshold superposition, the dynamic adjustment of the device fault warning threshold is achieved. Taking the energy storage converter in a wind farm as an example, when the device has a minor fault, the fault warning weight will increase accordingly. Through matrix dot multiplication, the reference threshold changes, and then the deviation is corrected by an extended Kalman filter. The finally generated warning parameters can timely reflect the fault state of the device and provide a reliable basis for the device's fault warning. In specific implementation, it is necessary to reasonably set the parameters of each step according to the type and operating characteristics of the device, such as the calculation method of the fault warning weight, the parameters of the extended Kalman filter, etc., to ensure that the generated warning parameters can accurately and timely reflect the fault risk of the device.

[0065] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0066] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A zero-carbon device fault warning method based on ESG scheduling, characterized in that, Including: Collecting multi-source ESG parameter data during the operation of zero-carbon equipment, where the ESG parameter data includes carbon footprint intensity, clean energy consumption, and charge-discharge efficiency of the energy storage system; Using a graph convolutional network to extract device state features from the ESG parameter data, obtaining a state feature matrix, and calculating the ESG correlation degree based on the covariance change of the state feature matrix; Performing feature dimensionality reduction on the cross-correlation matrix of the ESG correlation degrees at different time periods, extracting core feature vectors, and performing non-linear mapping on the core feature vectors through the ReLU activation function to obtain the fault warning weights for each time period; Dynamically adjusting the benchmark threshold of the state feature matrix according to the fault warning weights, generating warning parameters, and triggering the device fault warning process based on the warning parameters.

2. The zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, wherein, The device state feature extraction includes: Intercepting the ESG parameter data at a set sampling interval, constructing a state tensor and performing standardization processing; Using a multi-head graph convolutional layer to perform cross-device feature aggregation on the state tensor, and outputting the dot product sum of the adjacency matrix and the feature vector; Inputting the aggregation result into a gated recurrent unit, extracting multi-time scale features through the update gate and the reset gate, and screening the feature dimensions through an attention mask to generate a state feature matrix.

3. A zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, characterized in that, The calculating of the ESG correlation degree includes: Selecting the state feature matrix of any time period as the benchmark threshold, and calculating the covariance value of the feature vectors of its adjacent k time periods; Taking the ratio of the mean of the covariance values to the Euclidean distance of the benchmark threshold as the within-period correlation quantity; Calculating the cross-correlation coefficient between the benchmark threshold and the feature of the correlation period, and taking its absolute value as the between-period correlation quantity; Taking the harmonic mean of the within-period correlation quantity and the between-period correlation quantity as the ESG correlation degree.

4. A zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, characterized in that, The extracting of the core feature vectors includes: Constructing a cross-correlation matrix of the ESG correlation degrees at different time periods, and performing principal component analysis on it to obtain orthogonal feature vectors; Selecting the feature vectors with the principal component variance contribution rate exceeding the set threshold to form the core feature subspace; Mapping the cross-correlation matrix to the core feature subspace to obtain the core feature vectors after dimensionality reduction.

5. A zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, characterized in that, The obtaining of the fault warning weights for each time period includes: Performing normalization processing on the core feature vectors, and calculating their cosine similarity with the preset reference vector; Inputting the similarity value into a bidirectional long short-term memory network, and generating initial weights after transformation by the forward layer and the backward layer; Smoothing the initial weights through moving average filtering, and outputting the fault warning weights for each time period.

6. The zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, wherein The generating of the warning parameters includes: Performing matrix dot multiplication on the fault warning weights and the benchmark threshold to obtain a weight adjustment parameter; Calculating the deviation amount between the weight adjustment parameter and the benchmark threshold, and dynamically correcting the deviation amount through an extended Kalman filter; Superimposing the corrected deviation amount on the benchmark threshold to generate warning parameters.

7. A zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, characterized in that It also includes: Constructing a multi-dimensional device state space based on the warning parameters, and extracting the extreme points and covariance mutation regions of the space; When the density of the extreme points exceeds the set threshold or the span of the covariance mutation region is greater than the limit value, it is determined as the fault warning state, and multi-level warning instructions are generated.

8. A zero-carbon equipment fault warning method based on ESG scheduling according to claim 7, characterized in that, The constructing of the multi-dimensional device state space includes: Mapping the warning parameters to a multi-dimensional coordinate system according to the operation time sequence, and generating a state distribution point set; The spatial reconstruction of the state distribution point set is carried out by using the kernel density estimation method, and the covariance gradient and kurtosis coefficient of the reconstructed space are calculated; The noise of the gradient distribution is suppressed by the Gaussian filtering algorithm to eliminate random interference.

9. The zero-carbon equipment fault warning method based on ESG scheduling according to claim 1, characterized in that, It also includes: Establish a joint optimization model of ESG parameter data and equipment warning thresholds, and use the non-dominated sorting genetic algorithm to solve the optimal warning strategy; Couple the optimal warning strategy with the warning parameters in real time to generate a fault warning plan that maximizes the ESG benefits; The solution of the joint optimization model includes: Define the objective function as the weighted absolute value sum of the ESG index deviation and the warning miss rate, and the constraint condition is the safe operating range of the equipment; Perform Pareto optimal transformation on the objective function and decompose it into a warning threshold sub-problem and an ESG scheduling sub-problem; Solve the two sub-problems alternately and iteratively until convergence, and output the optimal warning strategy that meets the constraint conditions.

10. A zero-carbon device fault warning system based on ESG scheduling, characterized in that, It includes: A multi-source data acquisition module, which is used to obtain the carbon footprint intensity, clean energy consumption, and energy storage system charge and discharge efficiency data during the operation in real time, and construct a multi-dimensional ESG parameter time series matrix; A state feature extraction module, which uses a graph convolutional network to perform cross-device feature aggregation on the multi-dimensional ESG parameter time series matrix, captures the device state features at multiple time scales through a gated recurrent unit, and outputs a state feature matrix; An ESG correlation calculation module, which generates the intra-period correlation quantity and the inter-period correlation quantity according to the covariance change of the state feature matrix, and calculates their harmonic mean as the ESG correlation degree; A cross-correlation matrix decomposition module, which constructs a cross-correlation matrix for the ESG correlation degrees of different periods, and extracts the core eigenvectors through principal component analysis; A weight mapping module, which uses the ReLU activation function to perform non-linear mapping on the core eigenvectors to generate the fault warning weights for each period; A parameter optimization module, which dynamically adjusts the reference threshold based on the fault warning weights, and corrects the parameter deviation amount through an extended Kalman filter to generate warning parameters; A real-time warning module, which converts the warning parameters into equipment fault warning signals and transmits them to the monitoring platform through the industrial Internet to trigger the warning instructions.

Images