New energy equipment fault monitoring and analyzing method

Abstract

The invention relates to the technical field of new energy, in particular to a new energy equipment fault monitoring and analyzing method. According to the invention, the integrity and real-time performance of data are improved through data fusion and real-time state estimation, and the reliability and real-time performance of fault detection are enhanced; a nonlinear estimation method is combined with a physical model and a data driving model, so that the processing capacity of a complex nonlinear system is improved, and the defects in nonlinear system processing are overcome; multi-scale data fusion and comprehensive fault analysis are adopted, comprehensive analysis of different levels of data of the energy storage equipment is achieved, comprehensiveness and accuracy of fault detection are improved, the problems that in the prior art, accuracy and integrity of sensor data are seriously relied on, and an electric power system is often affected by noise, faults, loss and the like are solved, and the system reliability is improved. And meanwhile, the problems that a state estimation method is difficult to meet the requirement of real-time monitoring of a new energy power system and a traditional linearization processing method is poor in effect when a strong nonlinear system is processed are solved.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and in particular to a new energy equipment fault monitoring and analysis method. Background Art

[0002] At present, the power system state estimation StateEstimation, SE methods usually adopt such as weighted least squares WLS, Kalman filter KF and extended Kalman filter EKF, which are heavily dependent on the accuracy and completeness of sensor data.

[0003] Chinese Patent Publication No.: CN117332920A discloses a method for analyzing the evolution of faults in the operation of a new energy station, which belongs to the technical field of management of new energy stations. The present invention includes fault evolution prediction and fault analysis, which can realize the prediction of fault damage of new energy station equipment and the analysis, location and remediation of fault damage; can predict the loss of some new energy stations in advance; can reduce the loss of new energy stations during daily operation; can help staff understand the impact of environmental changes on new energy station equipment before the environment changes, facilitate workers to change maintenance plans, facilitate staff to arrange time reasonably, and can fully reduce the impact of environmental factors on the operation of new energy station equipment; when there are abnormalities in the monitoring data, the data is collected and checked multiple times, which can effectively reduce the impact caused by the abnormal data collected by the detection device, so as to ensure the accuracy of the abnormal state analysis, have good use effect, and have good use prospects.

[0004] However, sensor data in new energy equipment such as wind power generation and solar power generation are often affected by noise, failure and loss, resulting in inaccurate estimation results; in addition, the state estimation method is complex in calculation and cannot meet the needs of real-time monitoring of new energy power systems; the power output of new energy systems is highly volatile, and traditional methods are difficult to respond quickly; and the characteristics of some new energy equipment such as wind turbines and solar panels are often nonlinear, and traditional linear processing methods such as extended Kalman filtering are not effective when dealing with strongly nonlinear systems. Therefore, a new energy equipment fault monitoring and analysis method is urgently needed to solve such problems. Summary of the invention

[0005] To this end, the present invention provides a new energy equipment fault monitoring and analysis method to overcome the problems in the prior art that it heavily relies on the accuracy and completeness of sensor data, and the power system is often affected by noise, failures and loss, resulting in inaccurate estimation results; and the state estimation method is difficult to meet the needs of real-time monitoring of new energy power systems; traditional linear processing methods such as extended Kalman filtering are not effective when processing strongly nonlinear systems.

[0006] To achieve the above object, the present invention provides a new energy equipment fault monitoring and analysis method, comprising:

[0007] Step S1, multi-source and multi-scale data fusion, integrating multiple sensor data (voltage, current, temperature, vibration) of wind turbines, solar panels and energy storage devices, and combining data at different levels (battery cells, battery packs, BMS, etc.), and performing comprehensive analysis through data fusion algorithms; the fused data set has high integrity and high accuracy, improves the integrity and accuracy of the data, and enhances the reliability of fault detection;

[0008] Step S2, state estimation, based on the fused data, combined with the chemical reaction model and the recursive neural network model, adopts a hybrid state estimation method to perform real-time state estimation, and outputs a real-time updated device state estimation value; improves the system response speed and real-time performance, and adapts to the high volatility of power output of the new energy system;

[0009] Step S3, equipment status modeling and prediction, based on real-time status estimation and historical data, combined with the time series prediction Seq2Seq model, the equipment status is modeled and predicted; based on the fault prediction results based on the deep learning model in step S3, the estimation accuracy is improved to solve the error accumulation and estimation deviation problems of traditional methods when dealing with nonlinear systems;

[0010] Step S4, fault detection, uses a fusion model of physical modeling, i.e., a battery model, and data-driven random forest modeling to perform fault detection; Step S4 outputs more accurate and robust fault detection and prediction results, and enhances the ability to identify complex fault modes;

[0011] Step S5, hierarchical fault detection, based on the multi-source and multi-scale data fusion results, establishes a hierarchical fault detection model based on battery cells, battery packs and BMS, and uses hierarchical analysis method to perform fault inference; outputs hierarchical fault detection results and comprehensive fault analysis reports, improves the ability to identify complex fault modes, and provides comprehensive fault diagnosis;

[0012] Step S6, fault warning, establishes a fault prediction model based on historical data and real-time data, uses long short-term memory network LSTM combined with attention mechanism to predict faults; outputs intelligent warning signals and detailed fault prediction reports; thereby improving the accuracy and foresight of fault prediction.

[0013] Furthermore, the multi-source and multi-scale data fusion method in step S1 is:

[0014] Perform data cleaning, remove outliers and noise, and normalize data from different sensors to the same scale;

[0015] Kalman filtering is used to fuse multi-source data through state prediction and update:

[0016] Initial state estimate: Initial error covariance matrix: P0;

[0017] To predict the state, the state transfer equation is expressed as: in is the predicted state at time k, F is the state transfer matrix, is the state estimation at time k-1, B is the control matrix, u k is the control vector, i.e., the external input;

[0018] The error covariance prediction is denoted as P k|k-1 =FP k-1|k-1 F T +Q,P k|k-1 That is, the prediction error covariance at time k, Q is the process noise covariance matrix, F T is the transpose of the state transfer matrix;

[0019] Calculate the Kalman gain, which is expressed as: K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , K k That is, the Kalman gain at time k,

[0020] H is the observation matrix, H T is the transpose of the observation matrix, R is the observation noise covariance matrix;

[0021] Update the status. The status update process is as follows: where z k is the observed value at time k;

[0022] Perform error covariance update, the error covariance update process is: P k|k =(IK k H)P k|k-1 , where I is the identity matrix;

[0023] Furthermore, the multi-source and multi-scale data fusion method in step S1 also includes:

[0024] Since particle filtering is applicable to nonlinear and non-Gaussian systems, particle filtering is used here to represent the probability distribution of the state through a set of weighted particles:

[0025] Generate N particles and initialize the particle set: Represents the state of the i-th particle at the initial time (t=0), and initializes the particle weight: in represents the weight of the i-th particle at the initial moment. At the initial moment, the weights of all particles are equal, and the sum of their weights is 1;

[0026] Perform importance sampling and state transfer: in represents the i-th particle at time k, represents the state of particle k-1 at a known time Next, the particle state x k The probability distribution of

[0027] Weight update: represents the weight of the i-th particle at time k, represents the state of the i-th particle at a known time k Next, we observe z k The probability of represents the state sequence of k-1 particles at a known time and the observation sequence z 1:k Next, the particle state Importance distribution of

[0028] Normalize and normalize the weights: in represents the normalized weight of the i-th particle at time k, represents the sum of all particle weights at time k;

[0029] Resample according to the normalized weights Resample to get a new particle set

[0030] Furthermore, the state estimation method in step S2 includes:

[0031] Acquire real-time data from the multi-source and multi-scale data fusion results of step S1, including voltage, current, temperature, and vibration;

[0032] Perform initial state estimation, including the initial voltage, current, temperature, and vibration state variables of the equipment, and initialize the error covariance matrix;

[0033] Use the battery's chemical reaction model to predict the battery state;

[0034] Use recurrent neural network, RNN, for state prediction and state estimation based on historical observation data;

[0035] The prediction results of the chemical reaction model and the recursive neural network model are weighted and fused to obtain a fused prediction state;

[0036] Perform Kalman filter update, calculate Kalman gain, use observations to update predicted state, and obtain updated state estimate;

[0037] Furthermore, the state estimation method in step S2 also includes:

[0038] Update the error covariance matrix to reflect the uncertainty of the new state estimate;

[0039] Perform importance sampling and generate new particle states according to the state transition model;

[0040] Update particle weights and adjust the weight of each particle according to the observed data;

[0041] Normalize particle weights and resample according to the normalized particle weights;

[0042] The resampled particle states are weighted averaged to obtain the final real-time state estimate;

[0043] Further, the chemical reaction model in step S2 is expressed as: Where V is the battery voltage, E0 is the standard electrode potential, I is the current, R is the internal resistance, k is the Boltzmann constant, T is the temperature, n is the number of electrons, F is the Faraday constant, Q is the remaining charge, Q max is the maximum battery capacity, ln is the natural logarithm;

[0044] Use recurrent neural network (RNN) for state prediction and state estimation based on historical observation data;

[0045] The RNN model is represented as: t =σ(W h x t +U h h t-1 +b h ), y t =W y h t +b y ), where h t is the hidden state at time t, x t is the input observation data at time t, including voltage, current, W h , U h , W y is the weight matrix, b h and b y is the bias, σ is the activation function, y t is the output at time t, i.e., a certain state estimation value of the device, including voltage, current, temperature, and vibration;

[0046] The prediction results of the chemical reaction model and the recursive neural network model are weighted and fused to obtain a fused prediction state;

[0047] Furthermore, the device status modeling prediction method in step S3 is:

[0048] Collect the real-time state estimation values ​​and historical data obtained from step S2, including voltage, current, temperature, and vibration;

[0049] Format the input data into a time series forecast format, divided into input sequences and target sequence Among them, each x T and T Contains multiple state variables, T represents the length of the time series;

[0050] Build a Seq2Seq model that encodes the input sequence X into latent states and decodes the latent states into an output sequence

[0051] Input sequence X is fed into the LSTM unit to generate hidden states and cell states:

[0052] h t ,c t =LSTM(x t ,h t-1 ,c t-1 ), where h t is the hidden state at time t, c t is the cell state at time t, x t is the input data at time t, h t-1 is the hidden state at time t-1, c t-1 is the cell state at time t-1;

[0053] The last hidden state h of the encoder T and cell state c T As the initial state of the decoder, the decoder generates the predicted output sequence

[0054] in is the predicted output at time t, is the predicted output at time t-1, h′ t is the hidden state of the decoder at time t, c′ t is the cell state of the decoder at time t, h′ t-1 is the hidden state of the decoder at time t-1, c′ t-1 is the cell state of the decoder at time t-1;

[0055] Furthermore, the device status modeling prediction method in step S3 also includes:

[0056] Use mean square error MSE as the loss function to measure the predicted value The difference from the actual value Y: L is the loss value, y t is the actual value at time t, is the predicted value at time t;

[0057] The model parameters are updated using the back-propagation algorithm and the Adam optimizer. The update process is expressed as: Where θ is the model parameter, η is the learning rate, is the gradient of the loss function with respect to the model parameters;

[0058] Furthermore, the fault detection step in step S4 includes:

[0059] The data in step S3 and the real-time monitoring data are normalized, and the battery model is a chemical reaction model, that is,

[0060] Use the preprocessed data to train a random forest model and build a classifier for fault detection: in is the predicted output, N is the number of decision trees, f i (x) is the prediction of the i-th decision tree for input x;

[0061] Select key characteristics, including voltage, current, temperature, and vibration, as model inputs;

[0062] A weighted fusion strategy is used to fuse the output results of the physical model and the random forest model: final =αy phys +(1-α)y rf , where y final is the final fused output, y phys is the output of the physical model, y rf is the output of the random forest model, α is the weighting coefficient;

[0063] Set a threshold and determine whether the device is faulty based on the output of the fusion model;

[0064] Furthermore, the hierarchical fault detection method in step S5 includes:

[0065] Collect multi-source and multi-scale data fusion results, including data from battery cells, battery packs, and battery management systems (BMS);

[0066] Establish a fault detection model for each battery cell, using key characteristics of battery voltage, current, and temperature for modeling;

[0067] Monitor the status of each battery cell in real time and detect abnormal conditions, including voltage imbalance and overtemperature;

[0068] Use the voltage, current, and temperature characteristics of the battery pack to build a fault detection model;

[0069] Monitor the status of the battery pack in real time and detect abnormal conditions, including battery pack voltage imbalance and SOC abnormality;

[0070] Establish a fault detection model for the battery management system, using system-level data including total voltage, total current, temperature, and SOC for modeling;

[0071] Monitor the status of the BMS in real time and detect system-level anomalies, including communication failures and sensor failures;

[0072] Construct a hierarchical analysis model to comprehensively analyze the fault detection results of battery cells, battery packs and BMS;

[0073] Combine the data and test results at each level to perform fault inference, use the hierarchical analysis method to determine the root cause of the fault, and assess the severity and scope of the fault.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] The present invention improves the integrity and real-time performance of data through data fusion and real-time state estimation, thereby enhancing the reliability and real-time performance of fault detection.

[0076] The present invention adopts a nonlinear estimation method combined with a physical model and a data-driven model to improve the processing capability of complex nonlinear systems and solve the deficiencies in nonlinear system processing.

[0077] The present invention adopts multi-scale data fusion and comprehensive fault analysis to achieve comprehensive analysis of data at different levels of energy storage equipment, thereby improving the comprehensiveness and accuracy of fault detection.

[0078] The present invention improves the accuracy and foresight of fault prediction through intelligent early warning and predictive maintenance, combines historical data with real-time data, and optimizes algorithms, thereby reducing equipment failure rate and maintenance costs;

[0079] It solves the problem that the existing technology relies heavily on the accuracy and completeness of sensor data, and the power system is often affected by noise, failure and loss. It also solves the problem that the state estimation method is difficult to meet the needs of real-time monitoring of new energy power systems, and the traditional linear processing method is not effective when dealing with strongly nonlinear systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a schematic diagram of the process of the new energy equipment fault monitoring and analysis method of the present invention. DETAILED DESCRIPTION

[0081] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0082] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0083] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0084] Example 1

[0085] See also Figure 1 The present invention provides a new energy equipment fault monitoring and analysis method, comprising:

[0086] Step S1, multi-source and multi-scale data fusion, integrating multiple sensor data (voltage, current, temperature, vibration) of wind turbines, solar panels and energy storage devices, and combining data at different levels (battery cells, battery packs, BMS, etc.), and performing comprehensive analysis through data fusion algorithms; the fused data set has high integrity and high accuracy, improves the integrity and accuracy of the data, and enhances the reliability of fault detection;

[0087] The multi-source and multi-scale data fusion method in step S1 is:

[0088] Perform data cleaning, remove outliers and noise, and normalize data from different sensors to the same scale;

[0089] Kalman filtering is used to fuse multi-source data through state prediction and update:

[0090] Initial state estimate: Initial error covariance matrix: P0;

[0091] To predict the state, the state transfer equation is expressed as: in is the predicted state at time k, F is the state transfer matrix, is the state estimation at time k-1, B is the control matrix, u k is the control vector, i.e., the external input;

[0092] The error covariance prediction is denoted as P k|k-1 =FP k-1|k-1 F T +Q,P k|k-1 That is, the prediction error covariance at time k, Q is the process noise covariance matrix, F T is the transpose of the state transfer matrix;

[0093] Calculate the Kalman gain, which is expressed as: K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , K k That is, the Kalman gain at time k,

[0094] H is the observation matrix, H T is the transpose of the observation matrix, R is the observation noise covariance matrix;

[0095] Update the status. The status update process is as follows: where z k is the observed value at time k;

[0096] Perform error covariance update, the error covariance update process is: P k|k =(IK k H)P k|k-1 , where I is the identity matrix;

[0097] The multi-source and multi-scale data fusion method in step S1 also includes:

[0098] Since particle filtering is applicable to nonlinear and non-Gaussian systems, particle filtering is used here to represent the probability distribution of the state through a set of weighted particles:

[0099] Generate N particles and initialize the particle set: Represents the state of the i-th particle at the initial time (t=0), and initializes the particle weight: in represents the weight of the i-th particle at the initial moment. At the initial moment, the weights of all particles are equal, and the sum of their weights is 1;

[0100] Perform importance sampling and state transfer: in represents the i-th particle at time k, represents the state of particle k-1 at a known time Next, the particle state x k The probability distribution of

[0101] Weight update: represents the weight of the i-th particle at time k, represents the state of the i-th particle at a known time k Next, we observe z k The probability of represents the state sequence of k-1 particles at a known time and the observation sequence z 1:k Next, the particle state Importance distribution of

[0102] Normalize and normalize the weights: in represents the normalized weight of the i-th particle at time k, represents the sum of all particle weights at time k;

[0103] Resample according to the normalized weights Resample to get a new particle set

[0104] Specifically, by integrating multiple sensor data from wind turbines, solar panels and energy storage devices, and combining data at different levels (such as battery cells, battery packs, BMS), high data integrity and accuracy are achieved; Kalman filtering and particle filtering data fusion algorithms are used to process noise and outliers in multi-source data to provide a more reliable fault detection basis; improving data integrity and accuracy, and significantly enhancing the reliability of fault detection;

[0105] Step S2, state estimation, based on the fused data, combined with the chemical reaction model and the recursive neural network model, adopts a hybrid state estimation method to perform real-time state estimation, and outputs a real-time updated device state estimation value; improves the system response speed and real-time performance, and adapts to the high volatility of power output of the new energy system;

[0106] The state estimation method in step S2 includes:

[0107] Acquire real-time data from the multi-source and multi-scale data fusion results of step S1, including voltage, current, temperature, and vibration;

[0108] Perform initial state estimation, including the initial voltage, current, temperature, and vibration state variables of the equipment, and initialize the error covariance matrix;

[0109] Use the battery's chemical reaction model to predict the battery state;

[0110] Use recurrent neural network, RNN, for state prediction and state estimation based on historical observation data;

[0111] The prediction results of the chemical reaction model and the recursive neural network model are weighted and fused to obtain a fused prediction state;

[0112] Perform Kalman filter update, calculate Kalman gain, use observations to update predicted state, and obtain updated state estimate;

[0113] The state estimation method in step S2 also includes:

[0114] Update the error covariance matrix to reflect the uncertainty of the new state estimate;

[0115] Perform importance sampling and generate new particle states according to the state transition model;

[0116] Update particle weights and adjust the weight of each particle according to the observed data;

[0117] Normalize particle weights and resample according to the normalized particle weights;

[0118] The resampled particle states are weighted averaged to obtain the final real-time state estimate;

[0119] Specifically, a hybrid state estimation method is used to perform real-time state estimation by combining the chemical reaction model and the recursive neural network (RNN) model. The chemical reaction model provides physical state prediction of the battery, while the RNN model predicts the state through historical observation data. The weighted fusion of the two models makes the state prediction more accurate and comprehensive. Through the update process of the Kalman filter, the system can reflect the dynamic state of new energy equipment in real time and adapt to the high volatility of power output, thereby improving the system response speed and real-time performance.

[0120] The chemical reaction model in step S2 is expressed as: Where V is the battery voltage, E0 is the standard electrode potential, I is the current, R is the internal resistance, k is the Boltzmann constant, T is the temperature, n is the number of electrons, F is the Faraday constant, Q is the remaining charge, Q max is the maximum battery capacity, ln is the natural logarithm;

[0121] Use recurrent neural network (RNN) for state prediction and state estimation based on historical observation data;

[0122] The RNN model is represented as: t =σ(W h x t +U h h t-1 +b h ), y t =W y h t +b y ), where h t is the hidden state at time t, x t is the input observation data at time t, including voltage, current, W h , U h , W y is the weight matrix, b h and b y is the bias, σ is the activation function, y t is the output at time t, i.e., an estimated value of a certain state of the device, including voltage, current, temperature, and vibration;

[0123] The prediction results of the chemical reaction model and the recursive neural network model are weighted and fused to obtain a fused prediction state;

[0124] In addition, the problem of insufficient fault detection accuracy is optimized; that is, it relies on predefined fault modes, and has a low recognition rate for complex and diverse fault conditions, especially new fault modes that have never been seen before;

[0125] The present invention performs intelligent early warning and predictive maintenance, combines historical data with real-time data; establishes a fault prediction model based on historical data and real-time data, realizes early warning of faults, and improves the accuracy and foresight of fault prediction;

[0126] Step S3, equipment status modeling and prediction, based on the real-time status estimation value and historical data, combined with the time series prediction Seq2Seq model, the equipment status is modeled and predicted;

[0127] The equipment status modeling prediction method in step S3 is:

[0128] Collect the real-time state estimation values ​​and historical data obtained from step S2, including voltage, current, temperature, and vibration;

[0129] Format the input data into a time series forecast format, divided into input sequences and target sequence Among them, each x Tand T Contains multiple state variables, T represents the length of the time series;

[0130] Build a Seq2Seq model that encodes the input sequence X into latent states and decodes the latent states into an output sequence

[0131] Input sequence X is fed into the LSTM unit to generate hidden states and cell states:

[0132] h t ,c t =LSTM(x t ,h t-1 ,c t-1 ), where h t is the hidden state at time t, c t is the cell state at time t, x t is the input data at time t, h t-1 is the hidden state at time t-1, c t-1 is the cell state at time t-1;

[0133] The last hidden state h of the encoder T and cell state c T As the initial state of the decoder, the decoder generates the predicted output sequence

[0134] in is the predicted output at time t, is the predicted output at time t-1, h′ t is the hidden state of the decoder at time t, c′ t is the cell state of the decoder at time t, h′ t-1 is the hidden state of the decoder at time t-1, c′ t-1 is the cell state of the decoder at time t-1;

[0135] Specifically, a time series prediction model is used to model and predict the device status. The Seq2Seq model processes the input sequence through LSTM units, generates hidden states and cell states, and then decodes them into output sequences. Combining historical data and real-time state estimates, the Seq2Seq model can effectively capture the long-term dependencies of device status, improve the accuracy of prediction, and solve the error accumulation and estimation bias problems of traditional methods when dealing with nonlinear systems.

[0136] The device status modeling prediction method in step S3 also includes:

[0137] Use mean square error MSE as the loss function to measure the predicted value The difference from the actual value Y: L is the loss value, y t is the actual value at time t, is the predicted value at time t;

[0138] The model parameters are updated using the back-propagation algorithm and the Adam optimizer. The update process is expressed as: Where θ is the model parameter, η is the learning rate, is the gradient of the loss function with respect to the model parameters;

[0139] Step S4, fault detection, using a fusion model of physical modeling, i.e., a battery model, and data-driven random forest modeling to perform fault detection;

[0140] The fault detection step in step S4 includes:

[0141] The data in step S3 and the real-time monitoring data are normalized, and the battery model is a chemical reaction model, that is,

[0142] Use the preprocessed data to train a random forest model and build a classifier for fault detection: in is the predicted output, N is the number of decision trees, f i (x) is the prediction of the i-th decision tree for input x;

[0143] Select key characteristics, including voltage, current, temperature, and vibration, as model inputs;

[0144] A weighted fusion strategy is used to fuse the output results of the physical model and the random forest model: final =αy phys +(1-α)y rf , where y final is the final fused output, y phys is the output of the physical model, y rf is the output of the random forest model, α is the weighting coefficient;

[0145] Set a threshold and determine whether the device is faulty based on the output of the fusion model;

[0146] Specifically, a fusion model is established for fault detection using physical modeling and data-driven random forest model; the physical model provides the battery status based on chemical reactions, while the random forest model performs fault classification in a data-driven manner; the fusion strategy improves the accuracy and robustness of fault detection and enhances the ability to identify complex fault modes; a threshold is set to determine whether a device has failed based on the output of the fusion model to ensure the reliability of fault detection;

[0147] Step S5, hierarchical fault detection, based on the multi-source and multi-scale data fusion results, a hierarchical fault detection model based on battery cells, battery packs and BMS is established, and fault inference is performed using a hierarchical analysis method;

[0148] The hierarchical fault detection method in step S5 includes:

[0149] Collect multi-source and multi-scale data fusion results, including data from battery cells, battery packs, and battery management systems (BMS);

[0150] Establish a fault detection model for each battery cell, using key characteristics of battery voltage, current, and temperature for modeling;

[0151] Monitor the status of each battery cell in real time and detect abnormal conditions, including voltage imbalance and overtemperature;

[0152] Use the voltage, current, and temperature characteristics of the battery pack to build a fault detection model;

[0153] Monitor the status of the battery pack in real time and detect abnormal conditions, including battery pack voltage imbalance and SOC abnormality;

[0154] Establish a fault detection model for the battery management system, using system-level data including total voltage, total current, temperature, and SOC for modeling;

[0155] Monitor the status of the BMS in real time and detect system-level anomalies, including communication failures and sensor failures;

[0156] Construct a hierarchical analysis model to comprehensively analyze the fault detection results of battery cells, battery packs and BMS;

[0157] Combine the data and test results at each level to infer the fault, use the hierarchical analysis method to determine the root cause of the fault, and assess the severity and impact of the fault;

[0158] Specifically, based on the results of multi-source and multi-scale data fusion, a hierarchical fault detection model for battery cells, battery packs and BMS is established; each level of the model monitors its status in real time and detects abnormal conditions; by building a hierarchical analysis model, the data and detection results of each level are comprehensively analyzed to determine the root cause of the fault and assess the severity and scope of the fault; the hierarchical fault detection model provides more comprehensive fault diagnosis and improves the ability to identify complex fault modes;

[0159] Step S6, fault warning, establishes a fault prediction model based on historical data and real-time data, uses long short-term memory network LSTM combined with attention mechanism to predict faults; outputs intelligent warning signals and detailed fault prediction reports; thereby improving the accuracy and foresight of fault prediction.

[0160] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for monitoring and analyzing faults of new energy equipment. It is characterized in that include: Step S1, multi-source and multi-scale data fusion, integrating multiple sensor data of wind turbines, solar panels and energy storage devices, and combining data at different levels to conduct comprehensive analysis through data fusion algorithms; Step S2, state estimation, based on the fused data, combined with the chemical reaction model and the recursive neural network model, a hybrid state estimation method is used to perform real-time state estimation, and output a real-time updated device state estimation value; Step S3, equipment status modeling and prediction, based on real-time status estimation and historical data, combined with the time series prediction Seq2Seq model, to model and predict the status of the equipment; Step S4, fault detection, using a fusion model of physical modeling, i.e., a battery model, and data-driven random forest modeling to perform fault detection; Step S5, hierarchical fault detection, based on the multi-source and multi-scale data fusion results, a hierarchical fault detection model based on battery cells, battery packs and BMS is established, and fault inference is performed using a hierarchical analysis method; Step S6, fault warning, based on historical data and real-time data, establish a fault prediction model, and use the long short-term memory network LSTM combined with the attention mechanism to perform fault prediction; Output intelligent early warning signals and detailed fault prediction reports.

2. A method for monitoring and analyzing faults of new energy equipment according to claim 1, It is characterized in that The multi-source and multi-scale data fusion method in step S1 is: Perform data cleaning, remove outliers and noise, and normalize data from different sensors to the same scale; Kalman filtering is used to fuse multi-source data through state prediction and update: Initial state estimate: Initial error covariance matrix: P 0 ; To predict the state, the state transfer equation is expressed as: in is the predicted state at time k, F is the state transfer matrix, is the state estimation at time k-1, B is the control matrix, u k is the control vector, i.e., the external input; The error covariance prediction is denoted as P k|k-1 =FP k-1|k-1 F T +Q,P k|k-1 That is, the prediction error covariance at time k, Q is the process noise covariance matrix, F T is the transpose of the state transfer matrix; Calculate the Kalman gain, which is expressed as: K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , K k That is, the Kalman gain at time k, H is the observation matrix, H T is the transpose of the observation matrix, R is the observation noise covariance matrix; Update the status. The status update process is as follows: where z k is the observed value at time k; Perform error covariance update, the error covariance update process is: P k|k =(IK k H)P k|k-1 , where I is the identity matrix.

3. A method for monitoring and analyzing faults of new energy equipment according to claim 2, It is characterized in that The multi-source and multi-scale data fusion method in step S1 also includes: Particle filtering is used to represent the probability distribution of the state through a set of weighted particles: Generate N particles and initialize the particle set: Represents the state of the i-th particle at the initial time (t=0), and initializes the particle weight: in represents the weight of the i-th particle at the initial moment. At the initial moment, the weights of all particles are equal, and the sum of their weights is 1; Perform importance sampling and state transfer: in represents the i-th particle at time k, represents the state of particle k-1 at a known time Next, the particle state x k The probability distribution of Weight update: represents the weight of the i-th particle at time k, represents the state of the i-th particle at a known time k Next, we observe z k The probability of represents the state sequence of k-1 particles at a known time and the observation sequence z 1:k Next, the particle state Importance distribution of Normalize and normalize the weights: in represents the normalized weight of the i-th particle at time k, represents the sum of all particle weights at time k; Resample according to the normalized weights Resample to get a new particle set 4. A method for monitoring and analyzing faults of new energy equipment according to claim 3, It is characterized in that The state estimation method in step S2 includes: Acquire real-time data from the multi-source and multi-scale data fusion results of step S1, including voltage, current, temperature, and vibration; Perform initial state estimation, including the initial voltage, current, temperature, and vibration state variables of the equipment, and initialize the error covariance matrix; Use the battery's chemical reaction model to predict the battery state; Use recurrent neural network, RNN, for state prediction and state estimation based on historical observation data; The prediction results of the chemical reaction model and the recursive neural network model are weighted and fused to obtain a fused prediction state; Perform a Kalman filter update, calculate the Kalman gain, use the observations to update the predicted state, and obtain an updated state estimate.

5. A method for monitoring and analyzing faults of new energy equipment according to claim 4, It is characterized in that The state estimation method in step S2 also includes: Update the error covariance matrix to reflect the uncertainty of the new state estimate; Perform importance sampling and generate new particle states according to the state transition model; Update particle weights and adjust the weight of each particle according to the observed data; Normalize particle weights and resample according to the normalized particle weights; The resampled particle states are weighted averaged to obtain the final real-time state estimate.

6. A method for monitoring and analyzing faults of new energy equipment according to claim 5, It is characterized in that The chemical reaction model in step S2 is expressed as: Where V is the battery voltage, E 0 is the standard electrode potential, I is the current, R is the internal resistance, k is the Boltzmann constant, T is the temperature, n is the number of electrons, F is the Faraday constant, Q is the remaining charge, Q max is the maximum battery capacity, ln is the natural logarithm; Use recurrent neural network (RNN) for state prediction and state estimation based on historical observation data; The RNN model is represented as: t =σ(W h x t +U h h t-1 +b h ), y t =W y h t +b y ), where h t is the hidden state at time t, x t is the input observation data at time t, including voltage, current, W h , U h , W y is the weight matrix, b h and b y is the bias, σ is the activation function, y t is the output at time t, i.e., a certain state estimation value of the device, including voltage, current, temperature, and vibration; The prediction results of the chemical reaction model and the recursive neural network model are weighted and fused to obtain the fused prediction state.

7. A method for monitoring and analyzing faults of new energy equipment according to claim 6, It is characterized in that The equipment status modeling prediction method in step S3 is: Collect the real-time state estimation values ​​and historical data obtained from step S2, including voltage, current, temperature, and vibration; Format the input data into a time series forecast format, divided into input sequences and target sequence Among them, each x T and T Contains multiple state variables, T represents the length of the time series; Build a Seq2Seq model that encodes the input sequence X into latent states and decodes the latent states into an output sequence Input sequence X is fed into the LSTM unit to generate hidden states and cell states: h t ,c t =LSTM(x t ,h t-1 ,c t-1 ), where h t is the hidden state at time t, c t is the cell state at time t, x t is the input data at time t, h t-1 is the hidden state at time t-1, c t-1 is the cell state at time t-1; The last hidden state h of the encoder T and cell state c T As the initial state of the decoder, the decoder generates the predicted output sequence in is the predicted output at time t, is the predicted output at time t-1, h′ t is the hidden state of the decoder at time t, c′ t is the cell state of the decoder at time t, h′ t-1 is the hidden state of the decoder at time t-1, c′ t-1 is the cell state of the decoder at time t-1.

8. A method for monitoring and analyzing faults of new energy equipment according to claim 7, It is characterized in that The device status modeling prediction method in step S3 also includes: Use mean square error MSE as the loss function to measure the predicted value The difference from the actual value Y: L is the loss value, yt is the actual value at time t, is the predicted value at time t; The model parameters are updated using the back-propagation algorithm and the Adam optimizer. The update process is expressed as: θ-η▽ θ L, where θ is the model parameter, η is the learning rate, and η▽ θ L is the gradient of the loss function with respect to the model parameters.

9. A method for monitoring and analyzing faults of new energy equipment according to claim 8, It is characterized in that The fault detection step in step S4 includes: The data in step S3 and the real-time monitoring data are normalized, and the battery model is a chemical reaction model, that is, Use the preprocessed data to train a random forest model and build a classifier for fault detection: in is the predicted output, N is the number of decision trees, f i (x) is the prediction of the i-th decision tree for input x; Select key characteristics, including voltage, current, temperature, and vibration, as model inputs; A weighted fusion strategy is used to fuse the output results of the physical model and the random forest model: final =αy phys +(1-α)y rf , where y final is the final fused output, y phys is the output of the physical model, y rf is the output of the random forest model, α is the weighting coefficient; Set the threshold and determine whether the device fails based on the output of the fusion model.

10. A method for monitoring and analyzing faults of new energy equipment according to claim 9, It is characterized in that The hierarchical fault detection method in step S5 includes: Collect multi-source and multi-scale data fusion results, including data from battery cells, battery packs, and battery management systems (BMS); Establish a fault detection model for each battery cell, using key characteristics of battery voltage, current, and temperature for modeling; Monitor the status of each battery cell in real time and detect abnormal conditions, including voltage imbalance and overtemperature; Use the voltage, current, and temperature characteristics of the battery pack to build a fault detection model; Monitor the status of the battery pack in real time and detect abnormal conditions, including battery pack voltage imbalance and SOC abnormality; Establish a fault detection model for the battery management system, using system-level data including total voltage, total current, temperature, and SOC for modeling; Monitor the status of the BMS in real time and detect system-level anomalies, including communication failures and sensor failures; Construct a hierarchical analysis model to comprehensively analyze the fault detection results of battery cells, battery packs and BMS; Combine the data and test results at each level to perform fault inference, use the hierarchical analysis method to determine the root cause of the fault, and assess the severity and scope of the fault.

Images