Remote operation and maintenance and fault diagnosis method and system for energy storage system

Abstract

The invention discloses a remote operation and maintenance and fault diagnosis method and system for an energy storage system, and relates to the field of fault diagnosis of energy storage equipment, a factory three-dimensional model is established, the binding relationship between a battery and the model is determined, data is collected from a battery management system, and after the steps of battery internal resistance estimation, data sequence labeling, abnormal value processing and Z-score standardization preprocessing, the fault diagnosis of the energy storage system is completed. And key characteristics of charge-discharge multiplying power and internal resistance change rate are extracted. And a multi-layer perceptron is selected to establish an artificial neural network model to realize a fault diagnosis prediction target. Once an abnormal battery is identified, the position of the abnormal battery in the plant global coordinate system is determined by means of coordinate information stored in the hash table. The fault prediction accuracy is improved by utilizing comprehensive data modeling, model parameters can be adaptively optimized, an abnormal battery is efficiently and accurately identified, and stable operation of an energy storage system is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of energy storage equipment fault diagnosis, and in particular to a method and system for remote operation and maintenance and fault diagnosis of an energy storage system. Background Art

[0002] In today's energy storage field, the scale of energy storage systems is becoming increasingly large, especially in large energy storage plants, which are often equipped with thousands of battery cells to meet the growing demand for energy storage and supply. However, as the scale of energy storage systems expands, their operation, maintenance and fault diagnosis face many challenges.

[0003] Traditional energy storage system operation and maintenance and fault diagnosis technologies have many limitations. In terms of battery positioning, there is a lack of accurate and efficient positioning methods, and it is difficult to quickly determine the specific location of many batteries in the factory. This means that during operation and maintenance and troubleshooting, staff usually need to spend a lot of manpower and time to search one by one, like looking for a needle in a haystack. It is not only inefficient, but also prone to omissions or misjudgments, greatly increasing the cost of operation and maintenance. In battery data management, the data is not highly correlated, and the relevant information of each battery is relatively scattered, making it difficult to intuitively and conveniently obtain detailed data of each battery for analysis and decision-making. The battery address, sensor, and tag information have not been effectively integrated with the physical location of the battery and other operating data, resulting in many inconveniences in data statistics, analysis, and making operation and maintenance decisions, and the potential value of the data cannot be fully utilized. For battery health status monitoring, especially the monitoring of the key indicator of battery resistance, the monitoring is not accurate enough. Traditional methods often cannot accurately grasp the real-time changes in battery resistance. As an important indicator reflecting the health status of the battery, inaccurate monitoring of battery resistance makes it difficult to detect potential battery problems in advance, such as battery aging and internal short circuit faults. Once these potential problems develop into actual failures, they may have a serious impact on the overall operation of the energy storage system, reduce the reliability of the energy storage system, shorten the service life of the battery, and thus increase operating costs and the instability of energy supply. In terms of fault diagnosis, traditional technologies mostly rely on relatively simple empirical judgments or simple analysis models based on limited data. These methods fail to fully utilize various data during battery operation, such as voltage, current, resistance, and temperature, and lack effective integration with a large number of historical fault conditions. Therefore, the prediction accuracy of battery failure types is low, and it is difficult to take preventive measures in advance. Often, a response can only be made after a failure occurs, resulting in a passive operation and maintenance work, which cannot effectively guarantee the stable operation of the energy storage system.

[0004] In summary, the existing energy storage system operation and maintenance and fault diagnosis technology can no longer meet the growing demand for efficient operation and maintenance and accurate fault diagnosis of large energy storage systems. In view of this, there is an urgent need for a more advanced, comprehensive and efficient energy storage system remote operation and maintenance and fault diagnosis method to overcome the above problems and improve the operation and maintenance efficiency, reliability and overall operation stability of the energy storage system. Summary of the invention

[0005] The purpose of the present invention is to provide a remote operation and maintenance and fault diagnosis method and system for energy storage systems. By establishing a three-dimensional model of the plant area and determining the binding relationship between the battery and the model, the voltage, current, and temperature data are collected from the battery management system and the battery internal resistance is estimated. After data sequence labeling, abnormal value processing, and Z-score standardization preprocessing, the key features of the charge and discharge rate and the internal resistance change rate are extracted; then a multi-layer perceptron is selected to establish an artificial neural network model to achieve the fault diagnosis prediction goal; once the abnormal battery is identified, its position in the global coordinate system of the plant area is determined with the help of the coordinate information stored in the hash table;

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for remote operation and maintenance and fault diagnosis of an energy storage system, characterized in that it comprises the following steps:

[0008] S1: Establish a three-dimensional model of the factory area, define a global coordinate system for the entire factory area, determine the origin position and coordinate axis direction with a corner of the factory area as the origin, determine the coordinate value of the battery compartment, and determine the coordinate value of the battery relative to the local coordinate system of the battery compartment;

[0009] S2: Determine the binding relationship, create a hash table to store the information of the battery and the battery compartment and the binding relationship between them, insert a node into the bucket of the hash table to store the relevant information of the battery;

[0010] S3: Collect the voltage, current, and temperature data of each battery at different times from the battery management system to estimate the battery internal resistance;

[0011] S4: Sequence the collected battery data and annotate the data associated with historical fault conditions;

[0012] S5: Outlier processing was performed on the battery data, and the Z-score standardization method was used to ensure that variables of different magnitudes had the same weight and comparability in the modeling process;

[0013] S6: Extract features, including charge and discharge rate, change rate of battery internal resistance, battery power, and thermal performance;

[0014] S7: Establishing an artificial neural network model: importing the features extracted in the previous steps to achieve the fault diagnosis prediction goal; thereby identifying the battery fault type;

[0015] S8: Determine the position of the abnormal battery, and determine the position of the abnormal battery in the global coordinate system of the entire factory through the normalized coordinate information of the abnormal battery in the local coordinate system of the battery compartment stored in the hash table and the normalized coordinate information of the battery compartment in the global coordinate system;

[0016] S9: Establish a remote operation and maintenance center to provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on fault diagnosis results and the real-time operation status of the energy storage system;

[0017] The three-dimensional model of the factory area is established in step S1 by first defining a global coordinate system for the entire factory area, determining the origin position and the direction of the coordinate axis with a corner of the factory area as the origin, and establishing the x, y, and z coordinate axes respectively;

[0018] For each battery compartment, measure its relative position in the factory area, and set the coordinates of the lower left corner of the battery compartment in the global coordinate system to The coordinates of the upper right vertex are

[0019] Normalize the coordinates of the battery compartment to facilitate subsequent calculations and modeling;

[0020] Define the normalization function N(X, X min , X max ), and its calculation formula is:

[0021]

[0022] Among them, X represents the coordinate value to be normalized, X min Indicates the coordinate value to be normalized, X max It is the maximum coordinate value range limit value in the corresponding coordinate direction;

[0023] Normalize the coordinates of the battery compartment. In the X direction, the normalized coordinates are:

[0024]

[0025] in is the minimum coordinate value of all battery compartments in the x direction, is the maximum coordinate value of all battery compartments in the x direction;

[0026] Similarly, normalization is performed in the y and z directions to obtain and

[0027] For each battery j, the initial placement position of the lower left vertex in the battery compartment is set to its coordinates in the global coordinate system as

[0028] First, convert the battery coordinates to the local coordinate system of the battery compartment to which it belongs. The calculation formula is as follows:

[0029] In the x-direction:

[0030]

[0031] in represents the coordinate value of battery j in the x direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the x direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the x direction in the global coordinate system;

[0032] In the y direction:

[0033]

[0034] in represents the coordinate value of battery j in the y direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the y direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the y direction in the global coordinate system;

[0035] In the z direction:

[0036]

[0037] in represents the coordinate value of battery j in the z direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the z direction in the global coordinate system, is the coordinate value of the lower left vertex of the battery compartment in the z direction in the global coordinate system;

[0038] Normalize the coordinates of the battery in the local coordinate system of the battery compartment, and use the normalization function N(x, x min , x max ), in the x direction, the normalized battery coordinates are:

[0039]

[0040] in is the minimum local coordinate value of all batteries in the battery compartment in the x direction, is the maximum local coordinate value of all batteries in the battery compartment in the x direction;

[0041] Similarly, the corresponding normalization operations are performed in the y and z directions to obtain and

[0042] In step S2, the binding relationship is determined by defining a spatial hash function H(x n ,y n , z n ), which is used to map the coordinates in the three-dimensional space to the index position in a hash table. The formula is:

[0043] H(x n ,y n , z n )=((int)(x n *p x +y n *p y +z n *p z )%m)

[0044] where x n ,y n ,z n is the coordinate value of the battery in the corresponding direction after the previous coordinate transformation and normalization operation, p x、 p y、 p z is a pre-selected prime number used to increase the randomness and dispersion of the hash function, so that different coordinates can be more evenly distributed in the hash table to avoid excessive hash conflicts. m is the size of the hash table, which is determined according to the coordinate range and precision to be processed. (int) represents the rounding operation, and % is the remainder operation. Create a hash table to store the information of the battery and the battery compartment and the binding relationship between them. Each bucket of the hash table can store a linked list structure to handle possible hash conflicts.

[0045] For each battery compartment, the normalized coordinates The hash index is calculated by the spatial hash function:

[0046]

[0047] In the hash table h i Insert a node into the bucket to store the relevant information of the battery compartment. The battery compartment number is A. i , normalized coordinates

[0048] For each battery j, its coordinates normalized in the local coordinate system of the battery compartment are The hash index is calculated by the spatial hash function:

[0049]

[0050] In the hash table h j Insert a node into the bucket to store battery-related information, including normalized coordinates Battery compartment number A i 、Battery address A j , sensor number S j , Tag T j At the same time, in the node storing battery information, a pointer is set to point to the node of the corresponding battery compartment in the hash table, so as to establish a binding relationship between the battery and the battery compartment. Through this binding algorithm based on spatial hashing, the battery can be effectively bound to the three-dimensional model, and when querying and processing battery-related information, the corresponding battery and battery compartment can be quickly located through the hash table, which improves the efficiency of data processing. In practical applications, the hash function and hash table structure can be optimized and adjusted according to specific circumstances.

[0051] In step S3, the battery internal resistance is estimated by collecting the voltage, current, and temperature data of each battery at different times from the battery management system (BMS). When the battery is in the process of charging and discharging, the battery internal resistance can be estimated by using the modified formula of Ohm's law. The battery internal resistance R b It can be calculated by the following formula:

[0052]

[0053] Among them, R b is the internal resistance of the battery, V oc is the open circuit voltage of the battery, that is, the voltage value of the battery when there is no external load. This value usually needs to be measured after the battery has been stationary for a long enough time, because at this time the chemical reaction inside the battery reaches a relatively stable state, and its voltage can be approximately regarded as the open circuit voltage. V is the voltage value actually measured at a certain moment in the charge and discharge process of the battery, and I is the current value of the battery corresponding to the measured voltage V during the charge and discharge process;

[0054] By measuring the voltage V and current I and taking the average value, a more accurate estimate of the battery internal resistance can be obtained. The average value of the battery internal resistance is It can be calculated as:

[0055]

[0056] in, It represents the average value of the battery internal resistance obtained after multiple measurements and averaging. n is the total number of times the voltage and current are measured. V iis the voltage value of the battery at the i-th measurement, where i ranges from 1 to n, representing different measurements, I i is the current value of the battery at the i-th measurement. Similarly, the value range of i is from 1 to n, V i is the voltage value obtained at the same time;

[0057] In step S4, the collected battery data is sequenced and annotated with the historical fault condition related data. Specifically, the collected battery data is sequenced, and the voltage data sequence is {V i}, the current data sequence is {I i}, the temperature data sequence is {T i}, where i = 1, 2, ..., n, the average value of the battery internal resistance is calculated from the measured data And set its data sequence to {R i};

[0058] The data associated with the historical fault conditions are annotated, and the historical fault records of each battery in the corresponding time period are collected, and the fault types are grouped as F = {f1, f2, ..., f m}, where m is the number of fault types. For each data point i, a labeling function is defined Its value comes from the fault type set F or is null, and is used to annotate the collected battery data with the corresponding historical fault conditions, that is:

[0059]

[0060] In step S5, the abnormal value processing of the battery data is specifically to process the data with noise. i}(i=1,2,…,n), use mean filtering to remove measurement noise, the calculation formula is:

[0061]

[0062] in, is the voltage value after mean filtering, which can more smoothly reflect the real situation of battery voltage and reduce noise interference. m is the filter window size, which determines the number of adjacent data points involved in the filter calculation. V j is the data point in the original voltage data sequence, where the value of j ranges from i-m+1 to i+m-1, that is, the data points involved in the calculation The adjacent data points of

[0063] Similarly, for the current data series {I i}、Resistance data sequence {R i} and temperature data series {T i}The same mean filtering method is used for noise processing. The current value after filtering is The resistance value is and the temperature value is

[0064] To process the voltage outliers, first calculate the mean μ of the voltage data sequence V and standard deviation σ V , the mean calculation formula is:

[0065]

[0066] where μ V is the voltage data sequence {V i}, which reflects the average level of voltage data, n is the total number of data points in the voltage data sequence, that is, the number of times the voltage data is collected, and the standard deviation calculation formula is:

[0067]

[0068] where σ V is the voltage data sequence {V i}, which measures the dispersion of voltage data relative to the mean. According to the 3σ principle, if V i Satisfy |V i -μ V |>3σ V , then determine V i As an outlier, delete it from the data series;

[0069] The same method is applied to current, resistance and temperature data to process the abnormal values ​​in their respective data sequences. Let the processed current data sequence be {I' i}, the resistance data sequence is {R' i} and the temperature data sequence is {T' i}; The Z-score standardization method is used to ensure that variables of different magnitudes have the same weight and comparability in the modeling process;

[0070] For the voltage data sequence {V i}, first calculate the mean μ V and standard deviation σ V , the standardized voltage value The calculation formula is:

[0071]

[0072] in, is the voltage value after Z-score standardization. Through this standardization process, voltage data of different magnitudes will be converted to have the same mean value, so that each variable can participate in the calculation in the modeling process. V is the mean of the voltage data series, σ V is the standard deviation of the voltage data series;

[0073] Similarly, for the current data series {I i}、Resistance data sequence {R i} and temperature data series {T i} also performs Z-score normalization, and the normalized current value is The resistance value is and the temperature value is

[0074] The feature extraction in step S6 is specifically to extract the feature, wherein the charge and discharge rate is calculated as:

[0075]

[0076] The charge and discharge rate of the battery reflects the ratio of the current intensity of the battery during the charge and discharge process to its rated capacity. It is an important indicator for measuring the charge and discharge speed of the battery. i is the battery current value measured at the corresponding timestamp i, C is the rated capacity of the battery, which is the total amount of electricity that the battery can store or release as specified during design;

[0077] The internal resistance change rate is calculated using the following formula:

[0078]

[0079] RR is the rate of change of the battery's internal resistance, which reflects the change of the battery's internal resistance over time and is of great significance for monitoring the battery's health status and performance changes. i is the battery resistance value calculated at the corresponding time stamp i, R i-1 is the battery resistance value calculated at timestamp i-1, which is equal to R i is the resistance value at adjacent time points, t i Indicates the time point corresponding to timestamp i, t i-1 is the time point at timestamp i-1, which is the same as t i are adjacent time points;

[0080] For power characteristic calculation, the calculation formula is:

[0081] P=V i ×I i

[0082] Where P is the battery power characteristic, which is a combination of two basic parameters, voltage and current, and reflects the power output or input of the battery at a certain moment. It is of great significance for analyzing the performance and working status of the battery. V i is the battery voltage value measured at the corresponding timestamp i, I i is the battery current value measured at the corresponding timestamp i;

[0083] The calculation formula for thermal performance related characteristics is:

[0084] HT=T i ×R i

[0085] HT is a characteristic related to the thermal performance of the battery, which is a combination of two parameters: temperature and resistance. i is the battery temperature value measured at the corresponding timestamp i, R i is the battery resistance value calculated at the corresponding timestamp i;

[0086] Assume that the charge and discharge rate feature sequence obtained after the above feature extraction is {CR i}, the characteristic sequence of internal resistance change rate is {RR i}, the power characteristic sequence is {P i}, the thermal performance related characteristic sequence is {HT i};

[0087] The establishment of the artificial neural network model in step S7 specifically includes selecting a multi-layer perceptron (MLP) as the artificial neural network model to process battery-related data to achieve the fault diagnosis prediction goal, and determining the number of nodes in each layer;

[0088] The number of input layer nodes is set to n, which corresponds to receiving the pre-processed battery data, which includes voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps; let the vector of the eigenvalues ​​obtained after the above feature extraction be the number of input layer nodes n;

[0089] Hidden layer setting, there are h hidden layers, for the i-th hidden layer (i=1,2,…,h), let the number of nodes be n hi ,The role of the hidden layer is to perform nonlinear transformation on the input data and ,explore the complex relationships in the data. Different settings of the number of ,hidden layer nodes can affect the expressive ability and ,generalization ability of the model;

[0090] The number of nodes in the output layer, let the number of nodes in the output layer be m, which corresponds to the prediction target; let there be, k different fault types, and the output value m of each node represents the possibility of the corresponding fault type;

[0091] Define connection weights and biases

[0092] Set the connection weight matrix. For the connection weight matrix from the input layer to the first hidden layer, its elements Represents the connection weight from the i-th node in the input layer to the j-th node in the first hidden layer; these connection weights determine the degree of influence of the information of the input layer nodes on each node in the first hidden layer

[0093] For the connection weight matrix from the kth hidden layer (k>1) to the k+1th hidden layer, its elements are represents the connection weight from the i-th node in the k-th hidden layer to the j-th node in the k+1-th hidden layer, which is used to control the transmission and transformation of information between different hidden layers;

[0094] Set the bias. For the jth node in the first hidden layer, set its bias to Bias is an adjustable parameter that, together with the connection weight, affects the output value of each node in the first hidden layer. It can be seen as an offset adjustment of the input information before the neuron is activated, so that the neuron can better adapt to different data distributions and functional relationships.

[0095] For the jth node in the kth hidden layer, let its bias be It plays a similar role in the calculation of each hidden layer;

[0096] Perform forward propagation calculations. During the forward propagation process, data is gradually propagated from the input layer to the hidden layer to the output layer. Each layer obtains output values ​​according to the corresponding calculation rules, and these output values ​​continue to propagate as inputs to the next layer.

[0097] Input layer to the first hidden layer,

[0098] The input vector is X = [x1, x2, ..., x n ], where x i The calculation formula of the output vector Y1 of the first hidden layer node is:

[0099]

[0100] in, is the output value of the jth node in the first hidden layer. It is the result of the calculation from the input layer to the first hidden layer, and will be used as the input of the next layer. f is the activation function oSigmoid function: Its output value is between 0 and 1. It can map the input value to an interval that is easy to understand and process. The output function is the probability estimation oRelU function of the fault occurrence: f(x) = max(0, x). When x>0, the output is x itself, and when x≤0, the output is 0. The connection weight from the i-th node in the input layer to the j-th node in the first hidden layer determines the influence of the information of the input layer node on the j-th node in the first hidden layer. is the bias of the jth node in the first hidden layer, which is used to offset the input information and affects the output value of the jth node in the first hidden layer together with the connection weight;

[0101] The propagation from the kth hidden layer to the k+1th hidden layer, when k>1, the output vector Y of the kth hidden layer node k The calculation formula is:

[0102]

[0103] Where Z j It represents the output value of the jth node in the output layer. It is the final output of the entire neural network and has different meanings depending on the specific prediction target fault type. f is the Sigmoid activation function used to output the probability of fault occurrence. is the connection weight from the i-th node of the last hidden layer to the j-th node of the output layer, which is used to control the transmission and transformation of information from the hidden layer to the output layer. is the bias of the jth node in the output layer, which adjusts the input information and affects the output value of the jth node in the output layer together with the connection weight. i Here, it refers to the corresponding element of the input vector of the last hidden layer node, which is obtained by processing the output vector of the h-1th hidden layer accordingly;

[0104] Define the training set and loss function

[0105] Assume that the training set data is {(X1,Y1),(X2,Y2),…,(X N ,N N )}, where X i is the input vector, including voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps. i is the corresponding output vector, indicating the possibility of occurrence of different fault types;

[0106] The loss function uses the mean square error (MSE) function and defines the loss function as:

[0107]

[0108] Where L represents the value of the loss function, which measures the difference between the model output and the actual output. The smaller the value of the loss function, the closer the model's prediction result is to the actual situation, and the better the model performance. N is the number of samples in the training set data, that is, the number of input-output pairs contained in the training set. i The model is based on the input vector X i The output value obtained by forward propagation is the prediction result of the neural network for the corresponding input, Y i is the same as the input vector X i The corresponding actual output value is specifically the real label determined by the actual fault condition of the battery;

[0109] Back propagation algorithm: The back propagation algorithm is based on the principle of gradient descent. It calculates the gradient of the loss function for the connection weights and biases of each layer, and then adjusts these parameters according to the gradient, so that the value of the loss function is continuously reduced, thereby optimizing the model.

[0110] Calculate the loss function L for the output value Z of the output layer node j The gradient is:

[0111]

[0112] in Represents the loss function L for the output value Z of the output layer node j The gradient of the loss function reflects the gradient of the loss function as Z j The rate of change, Z j is the output value of the jth node in the output layer, Y j is the actual output value corresponding to the jth node in the output layer;

[0113] According to the chain rule, the gradient of the loss function for the connection weights and biases of each layer is calculated step by step. For the connection weights from the output layer to the last hidden layer, The gradient calculation formula is:

[0114]

[0115] in, The bias of the jth node in the output layer is obtained by derivation according to the calculation formula of the output vector of the output layer node. The gradient calculation formula is:

[0116]

[0117] For the connection weights and biases between hidden layers, we also use a similar method. According to the chain rule, we gradually calculate the gradients of each layer to get the gradient of the loss function. For the connection weights from the kth hidden layer to the k+1th hidden layer, The gradient calculation formula is:

[0118]

[0119] in, The bias of the jth node in the kth hidden layer is obtained by derivation according to the calculation formula of the corresponding hidden layer node output vector. The gradient calculation formula is:

[0120]

[0121] According to the calculated gradients of the connection weights and biases of each layer, these parameters are adjusted according to a certain learning rate η. The update formula is:

[0122]

[0123] in, is the updated connection weight value, which is the original connection weight Based on the loss function, the gradient and learning rate are adjusted. η is the learning rate, which is a hyperparameter that controls the step size of the model parameter update. The update formula is:

[0124]

[0125] The connection weights and biases of other layers are updated according to the same update formula. By continuously repeating the above process, i.e. forward propagation to calculate the output value, calculate the loss function, back propagation to calculate the gradient, and update the parameters, the value of the loss function is continuously reduced until the predetermined training termination condition is reached, thereby achieving model optimization;

[0126] Through the above detailed artificial neural network (ANN) modeling process, using the battery voltage, current, resistance, temperature data and related historical fault conditions for training, a model that can effectively predict battery failure conditions can be constructed, providing strong support for the operation and maintenance management of the energy storage system.

[0127] In step S8, the abnormal battery location is determined. Specifically, in the process of establishing the three-dimensional modeling of the factory area, each battery stores relevant information in the hash table, including the battery number, normalized coordinates, and the battery compartment number A to which it belongs. i, the battery failure results calculated by the artificial neural network are also integrated into this data structure, and the failure results output by the artificial neural network for each battery are set to a vector F = [f1, f2, ..., f m ], for each battery, store its fault result vector F in the hash table, and in the node storing battery-related information, add a new field to store this fault result vector.

[0128] When the artificial neural network model identifies the battery fault vector F = [f1, f2, …, f m ], the battery is marked as an abnormal battery by adding a new field, and its battery number, the number of the battery compartment to which it belongs, and the normalized coordinate information in the battery compartment are recorded. For the abnormal battery, first obtain its normalized coordinate value in the local coordinate system of the battery compartment, and use the normalized coordinate information in the local coordinate system of the battery compartment stored in the hash table, as well as the normalized coordinate information of the battery compartment in the global coordinate system, to determine the position of the abnormal battery in the global coordinate system of the entire factory;

[0129] In step S9, a remote operation and maintenance center is established to transmit the collected data to the remote operation and maintenance center through wired or wireless communication, and a server is set up in the remote operation and maintenance center to receive the data transmitted from the energy storage system, classify and store the received data according to time series and equipment type, and provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation status of the energy storage system.

[0130] The remote operation and fault diagnosis method of the energy storage system achieves efficient operation through a series of orderly steps: first, a three-dimensional model of the plant is established and the binding relationship between the battery and the model is determined to lay the foundation for subsequent precise positioning; then, voltage, current, and temperature data are collected from the battery management system and the internal resistance of the battery is estimated. After data sequence labeling, outlier processing, and Z-score standardization preprocessing, the key features of charge and discharge rate and internal resistance change rate are extracted; then, a multi-layer perceptron is used to establish an artificial neural network model, and the number of nodes, connection weights, and biases of each layer are set. The model is optimized through forward propagation, loss function based on mean square error, and back propagation algorithm to achieve the fault diagnosis prediction goal; once an abnormal battery is identified, its position in the global coordinate system of the plant is determined with the help of coordinate information stored in the hash table; finally, the remote operation and maintenance center receives and classifies the data, and provides detailed operation and maintenance guidance and suggestions for on-site operation and maintenance personnel based on the fault diagnosis results and real-time operation status, thereby realizing comprehensive and accurate monitoring, diagnosis, and operation and maintenance management of the energy storage system.

[0131] Compared with the prior art, the present invention has the following beneficial effects:

[0132] By establishing a three-dimensional model of the factory and binding each battery to the three-dimensional model, and ensuring that the battery's address, sensor, and label are consistent, it is possible to accurately spatially locate each battery in the factory. This allows staff to quickly determine the location of specific batteries during operation and maintenance and troubleshooting, greatly shortening the time to find faulty batteries and improving operation and maintenance efficiency. In large-scale energy storage plants, there are tens of thousands of batteries, and accurate positioning can avoid blind searches and reduce manpower and time costs.

[0133] Tightly binding the battery to the three-dimensional model and keeping the relevant information consistent will help to more conveniently manage the battery-related data. Detailed information on each battery can be intuitively obtained based on the three-dimensional model, which is convenient for data analysis, statistics and subsequent operation and maintenance decision-making.

[0134] By measuring resistance, the resistance of the battery can be more accurately understood. Since battery resistance is one of the important indicators reflecting the health status of the battery, more accurate resistance monitoring can detect potential battery problems in advance, such as battery aging and internal short circuit faults, so that maintenance measures can be taken in time to extend the battery life and improve the overall reliability of the energy storage system.

[0135] By collecting all-round data on the voltage, current, resistance, and temperature of each battery, and building models based on big data and artificial intelligence, and combining various historical fault conditions for modeling training, this comprehensive data utilization method enables the established model to have a deeper understanding of the operating status and failure mode of the battery. The artificial intelligence model has a strong nonlinear fitting ability. Through training with a large amount of data, it can more accurately predict the type of failure that may occur in the battery and take preventive measures in advance, which significantly improves the accuracy of fault prediction compared to existing technologies.

[0136] As new battery data is continuously collected and the model is continuously trained based on historical failure conditions, the model parameters can be continuously adjusted and optimized according to actual operating conditions to better adapt to changes in battery performance and emerging failure modes.

[0137] By accurately measuring various battery parameters, using comprehensive modeling analysis and combining it with historical fault conditions, abnormal batteries can be identified more efficiently and accurately. Once an abnormal battery is identified, its position in the factory can be quickly located with the help of a three-dimensional model, facilitating timely and targeted maintenance or replacement operations, reducing the impact of abnormal batteries on the overall operation of the energy storage system, and ensuring the stable operation of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0138] Figure 1 This is a flow chart of a method for remote operation and maintenance and fault diagnosis of an energy storage system according to the present invention; DETAILED DESCRIPTION

[0139] The technical solutions in the embodiments of the present invention will be fully described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0140] like Figure 1 As shown, a method for remote operation and maintenance and fault diagnosis of an energy storage system includes the following steps:

[0141] S1: Establish a 3D model of the factory: define a global coordinate system for the entire factory, determine the origin position and coordinate axis direction with a corner of the factory as the origin, determine the coordinate value of the battery compartment, and determine the coordinate value of the battery relative to the local coordinate system of the battery compartment;

[0142] S2: Determine the binding relationship: Create a hash table to store the information of the battery and the battery compartment and the binding relationship between them, insert a node into the bucket of the hash table to store the relevant information of the battery;

[0143] S3: Collect the voltage, current, and temperature data of each battery at different times from the battery management system to estimate the battery internal resistance;

[0144] S4: Sequence the collected battery data and annotate the data associated with historical fault conditions;

[0145] S5: Outlier processing of battery data: The Z-score standardization method is used to make variables of different magnitudes have the same weight and comparability in the modeling process;

[0146] S6: Extract features: including charge and discharge rate, change rate of battery internal resistance, battery power, and thermal performance;

[0147] S7: Establishing an artificial neural network model: Importing the features extracted in the previous steps to achieve the fault diagnosis prediction goal, thereby identifying the battery fault type;

[0148] S8: Determine the position of the abnormal battery: Determine the position of the abnormal battery in the global coordinate system of the entire factory area through the normalized coordinate information of the abnormal battery in the local coordinate system of the battery compartment stored in the hash table and the normalized coordinate information of the battery compartment in the global coordinate system;

[0149] S9: Establish a remote operation and maintenance center to provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation status of the energy storage system.

[0150] The three-dimensional model of the factory area is established in step S1 by first defining a global coordinate system for the entire factory area, determining the origin position and the direction of the coordinate axis with a corner of the factory area as the origin, and establishing the x, y, and z coordinate axes respectively;

[0151] For each battery compartment, measure its relative position in the factory area, and set the coordinates of the lower left corner of the battery compartment in the global coordinate system to The coordinates of the upper right vertex are

[0152] Normalize the coordinates of the battery compartment and define the normalization function N(X, X min , X max ), and its calculation formula is:

[0153]

[0154] Among them, X represents the coordinate value to be normalized, X min Indicates the coordinate value to be normalized, X max It is the maximum coordinate value range limit value in the corresponding coordinate direction;

[0155] Normalize the coordinates of the battery compartment. In the X direction, the normalized coordinates are:

[0156]

[0157] in is the minimum coordinate value of all battery compartments in the x direction, is the maximum coordinate value of all battery compartments in the x direction;

[0158] Similarly, normalization is performed in the y and z directions to obtain and

[0159] For each battery j, the initial placement position of the lower left vertex in the battery compartment is set to its coordinates in the global coordinate system as

[0160] First, convert the battery coordinates to the local coordinate system of the battery compartment to which it belongs. The calculation formula is as follows:

[0161] In the x-direction:

[0162]

[0163] in represents the coordinate value of battery j in the x direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the x direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the x direction in the global coordinate system;

[0164] In the y direction:

[0165]

[0166] in represents the coordinate value of battery j in the y direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the y direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the y direction in the global coordinate system;

[0167] In the z direction:

[0168]

[0169] in represents the coordinate value of battery j in the z direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the z direction in the global coordinate system, is the coordinate value of the lower left vertex of the battery compartment in the z direction in the global coordinate system;

[0170] Normalize the coordinates of the battery in the local coordinate system of the battery compartment using the normalization function N(x, x min , x max ), in the x direction, the normalized battery coordinates are:

[0171]

[0172] in is the minimum local coordinate value of all batteries in the battery compartment in the x direction, is the maximum local coordinate value of all batteries in the battery compartment in the x direction;

[0173] Similarly, the corresponding normalization operations are performed in the y and z directions to obtain and

[0174] In step S2, the binding relationship is determined by defining a spatial hash function H(x n ,y n , z n ), which is used to map the coordinates in the three-dimensional space to the index position in a hash table. The formula is:

[0175] H(x n ,y n , z n )=((int)(x n *p x +y n *p y +z n *p z )%m)

[0176] where xn ,y n ,z n is the coordinate value of the battery in the corresponding direction after the previous coordinate transformation and normalization operation, p x、 p y、 p z is a pre-selected prime number used to increase the randomness and dispersion of the hash function, m is the size of the hash, (int) represents the rounding operation, and % is the remainder operation;

[0177] Create a hash table to store the information of batteries and battery compartments and the binding relationship between them. Each bucket of the hash table can store a linked list structure to handle possible hash conflicts.

[0178] For each battery compartment, the normalized coordinates The hash index is calculated by the spatial hash function:

[0179]

[0180] In the hash table h i Insert a node into the bucket to store the relevant information of the battery compartment. The battery compartment number is A. i , normalized coordinates

[0181] For each battery j, its coordinates normalized in the local coordinate system of the battery compartment are The hash index is calculated by the spatial hash function:

[0182]

[0183] In the hash table h j Insert a node into the bucket to store battery-related information, including normalized coordinates Battery compartment number A i

[0184] 、Battery address A j , sensor number S j , Tag T j At the same time, in the node storing the battery information, a pointer is set to point to the node of the battery compartment in the hashtag, thereby establishing a binding relationship between the battery and the battery compartment.

[0185] In step S3, the battery internal resistance is estimated by collecting the voltage, current and temperature data of each battery at different times from the battery management system BMS. When the battery is in the process of charging and discharging, the battery internal resistance is estimated by using the modified formula of Ohm's law. The battery internal resistance R b Calculated by the following formula:

[0186]

[0187] Among them, R b is the internal resistance of the battery, V oc is the open circuit voltage of the battery, V is the voltage value actually measured at a certain moment during the charge and discharge process of the battery, and I is the current value of the battery corresponding to the measured voltage V during the charge and discharge process;

[0188] By measuring the voltage V and current I and taking the average value, a more accurate estimate of the battery internal resistance can be obtained. The average value of the battery internal resistance is It can be calculated as:

[0189]

[0190] in, It represents the average value of the battery internal resistance obtained after multiple measurements and averaging. n is the total number of times the voltage and current are measured. V i is the voltage value of the battery at the i-th measurement, where i ranges from 1 to n, representing different measurements, I i is the current value of the battery at the i-th measurement. Similarly, the value range of i is from 1 to n, V i is the voltage value obtained at the same time.

[0191] In step S4, the collected battery data is sequenced and annotated with the historical fault condition related data. Specifically, the collected battery data is sequenced, and the voltage data sequence is {V i}, the current data sequence is {I i}, the temperature data sequence is {T i}, where i = 1, 2, ..., n, the average value of the battery internal resistance is calculated from the measured data And set its data sequence to {R i};

[0192] The data associated with the historical fault conditions are annotated, and the historical fault records of each battery in the corresponding time period are collected, and the fault types are grouped as F = {f1, f2, ..., f m}, where m is the number of fault types. For each data point i, a labeling function is defined

[0193]

[0194] Its value comes from the fault type set F or is a null value, and is used to annotate the collected battery data with the corresponding historical fault conditions.

[0195] In step S5, the abnormal value processing of the battery data is specifically to process the data with noise. i}(i=1,2,…,n), use mean filtering to remove measurement noise, the calculation formula is:

[0196]

[0197] in, is the voltage value after mean filtering, m is the filter window size, V j is a data point in the original voltage data sequence, where the value of j ranges from i-m+1 to i+m-1;

[0198] Similarly, for the current data series {I i}、Resistance data sequence {R i} and temperature data series {T i}The same mean filtering method is used for noise processing. The current value after filtering is The resistance value is and the temperature value is

[0199] To process the voltage outliers, first calculate the mean μ of the voltage data sequence V and standard deviation σ V , the mean calculation formula is:

[0200]

[0201] where μ V is the voltage data sequence {V i}, n is the total number of data points in the voltage data sequence, and the standard deviation is calculated as:

[0202]

[0203] where σ V is the voltage data sequence {V i}, which measures the dispersion of voltage data relative to the mean. According to the 3σ principle, if V i Satisfy |V i -μ V |>3σ V , then determine V i As an outlier, delete it from the data series;

[0204] The same method is applied to current, resistance and temperature data to process the abnormal values ​​in their respective data sequences. Let the processed current data sequence be {I' i}, the resistance data sequence is {R' i ] and the temperature data series is {T'i};

[0205] The Z-score standardization method is used to ensure that variables of different magnitudes have the same weight and comparability in the modeling process;

[0206] For the voltage data sequence {V i}, first calculate the mean μ V and standard deviation σ V , the standardized voltage value The calculation formula is:

[0207]

[0208] in, is the voltage value after Z-score standardization, where μ V is the mean of the voltage data series, σ V is the standard deviation of the voltage data series;

[0209] Similarly, for the current data series {I i}、Resistance data sequence {R i} and temperature data series {T i} also performs Z-score normalization, and the normalized current value is The resistance value is and the temperature value is

[0210] The feature extraction in step S6 is specifically to extract the feature, wherein the charge and discharge rate is calculated as:

[0211]

[0212] Where CR represents the charge and discharge rate of the battery, I i is the battery current value measured at the corresponding timestamp i, and C is the rated capacity of the battery;

[0213] The internal resistance change rate is calculated using the following formula:

[0214]

[0215] Where RR is the rate of change of the battery internal resistance, R i is the battery resistance value calculated at the corresponding time stamp i, R i-1 is the battery resistance value calculated at timestamp i-1, which is equal to R i is the resistance value at adjacent time points, t i Indicates the time point corresponding to timestamp i, t i-1 is the time point at timestamp i-1, which is the same as t i are adjacent time points;

[0216] For power characteristic calculation, the calculation formula is:

[0217] P=V i ×I i

[0218] Where P is the battery power characteristic, V i is the battery voltage value measured at the corresponding timestamp i, I i is the battery current value measured at the corresponding timestamp i;

[0219] The calculation formula for thermal performance related characteristics is:

[0220] HT=T i ×R i

[0221] Where HT is a characteristic related to battery thermal performance, T i is the battery temperature value measured at the corresponding timestamp i, R i is the battery resistance value calculated at the corresponding timestamp i;

[0222] Assume that the charge and discharge rate feature sequence obtained after the above feature extraction is {CR i}, the characteristic sequence of internal resistance change rate is {RR i}, the power characteristic sequence is {P i}, the thermal performance related characteristic sequence is {HT i}.

[0223] The establishment of the artificial neural network model in step S7 specifically includes selecting a multi-layer perceptron MLP as the artificial neural network model to process battery-related data to achieve the fault diagnosis prediction goal, and determining the number of nodes in each layer;

[0224] The number of input layer nodes is set to n, which corresponds to receiving the pre-processed battery data, which includes voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps; let the vector of the eigenvalues ​​obtained after the above feature extraction be the number of input layer nodes n;

[0225] Hidden layer setting, there are h hidden layers, for the i-th hidden layer (i=1,2,…,h), let the number of nodes be

[0226] The number of nodes in the output layer, let the number of nodes in the output layer be m, which corresponds to the prediction target; let there be, k different fault types, and the output value m of each node represents the possibility of the corresponding fault type;

[0227] Define the connection weights and biases, set the connection weight matrix, and for the connection weight matrix from the input layer to the first hidden layer, its elements represents the connection weight from the i-th node in the input layer to the j-th node in the first hidden layer;

[0228] For the connection weight matrix from the kth hidden layer (k>1) to the k+1th hidden layer, its elements are represents the connection weight from the i-th node in the k-th hidden layer to the j-th node in the k+1-th hidden layer;

[0229] Set the bias. For the jth node in the first hidden layer, set its bias to

[0230] For the jth node in the kth hidden layer, let its bias be It plays a similar role in the calculation of each hidden layer;

[0231] Perform forward propagation calculations, where data is gradually propagated from the input layer through the hidden layer to the output layer. Each layer obtains output values ​​according to the corresponding calculation rules, and these output values ​​continue to propagate as inputs to the next layer.

[0232] From the input layer to the first hidden layer, the input vector is X = [x1, x2, ..., x n ], where x i The calculation formula of the output vector Y1 of the first hidden layer node is:

[0233]

[0234] in, is the output value of the jth node in the first hidden layer, and f is the activation function oSigmoid function: Its output value is between 0 and 1, and the output function is the probability estimation oRelU function of fault occurrence: f(x) = max90, x), and its output is x itself when x>0, and the output is 0 when x≤0; The connection weight from the i-th node in the input layer to the j-th node in the first hidden layer determines the influence of the information of the input layer node on the j-th node in the first hidden layer. is the bias of the jth node in the first hidden layer, which is used to offset the input information and affects the output value of the jth node in the first hidden layer together with the connection weight;

[0235] The propagation from the kth hidden layer to the k+1th hidden layer, when k>1, the output vector Y of the kth hidden layer node k The calculation formula is:

[0236]

[0237] Where Z j represents the output value of the jth node in the output layer, f is the Sigmoid activation function used to output the probability of a fault. is the connection weight from the i-th node in the last hidden layer to the j-th node in the output layer, is the bias of the jth node in the output layer, x i Refers to the corresponding element of the input vector of the last hidden layer node;

[0238] Define the training set and loss function

[0239] Assume that the training set data is {(X1,Y1),(X2,Y2),…,(X N ,N N )}, where X i is the input vector, including voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps. i is the corresponding output vector, indicating the possibility of occurrence of different fault types;

[0240] The loss function uses the mean square error MSE function and defines the loss function as:

[0241]

[0242] Where L represents the value of the loss function. The smaller the value of the loss function, the closer the model's prediction result is to the actual situation, and the better the model's performance is. N is the number of samples in the training set data, that is, the number of input-output pairs contained in the training set. i The model is based on the input vector X i The output value obtained by forward propagation is the prediction result of the neural network for the corresponding input, Y i is the same as the input vector X i The corresponding actual output value is specifically the real label determined by the actual fault condition of the battery;

[0243] The back propagation algorithm is based on the principle of gradient descent. It calculates the gradient of the loss function for the connection weights and biases of each layer, and then adjusts these parameters according to the gradient, so that the value of the loss function is continuously reduced, thereby optimizing the model.

[0244] Calculate the loss function L for the output value Z of the output layer node j The gradient is:

[0245]

[0246] in Represents the loss function L for the output value Z of the output layer node j The gradient of Z j is the output value of the jth node in the output layer, Y j is the actual output value corresponding to the jth node in the output layer;

[0247] According to the chain rule, the gradient of the loss function for the connection weights and biases of each layer is calculated step by step. For the connection weights from the output layer to the last hidden layer, The gradient calculation formula is:

[0248]

[0249] in, The bias of the jth node in the output layer is obtained by derivation according to the calculation formula of the output vector of the output layer node. The gradient calculation formula is:

[0250]

[0251] For the connection weights and biases between hidden layers, we also use a similar method. According to the chain rule, we gradually calculate the gradients of each layer to get the gradient of the loss function. For the connection weights from the kth hidden layer to the k+1th hidden layer, The gradient calculation formula is:

[0252]

[0253] in, The bias of the jth node in the kth hidden layer is obtained by derivation according to the calculation formula of the corresponding hidden layer node output vector. The gradient calculation formula is:

[0254]

[0255] According to the calculated gradients of the connection weights and biases of each layer, these parameters are adjusted according to a certain learning rate η. The update formula is:

[0256]

[0257] in, is the updated connection weight value, η is the learning rate, which controls the step size of model parameter update, and for bias The update formula is:

[0258]

[0259] The connection weights and biases of other layers are updated according to the same update formula. By continuously repeating the above process, that is, forward propagation to calculate the output value, calculate the loss function, back propagation to calculate the gradient, and update the parameters, the value of the loss function is continuously reduced until the predetermined training termination condition is reached, thereby achieving model optimization.

[0260] In step S8, the abnormal battery location is determined. Specifically, in the process of establishing the three-dimensional model of the factory area, each battery stores relevant information in the hash table, including the battery number, normalized coordinates, and the battery compartment number to which it belongs. The battery failure result calculated by the artificial neural network is also integrated into this data structure, and the failure result output by the artificial neural network for each battery is set to a vector F = [f1, f2, ..., f m ], for each battery, store its fault result vector F in the hash table, and in the node storing battery-related information, add a new field to store this fault result vector;

[0261] When the artificial neural network model identifies the battery fault vector F = [f1, f2, …, f m ], the battery is marked as an abnormal battery by adding a new field, and its battery number, the number of the battery compartment to which it belongs, and the normalized coordinate information in the battery compartment are recorded. For the abnormal battery, first obtain its normalized coordinate value in the local coordinate system of the battery compartment, and use the normalized coordinate information in the local coordinate system of the battery compartment stored in the hash table, as well as the normalized coordinate information of the battery compartment in the global coordinate system, to determine the position of the abnormal battery in the global coordinate system of the entire factory.

[0262] In step S9, a remote operation and maintenance center is established to transmit the collected data to the remote operation and maintenance center through wired or wireless communication, and a server is set up in the remote operation and maintenance center to receive the data transmitted from the energy storage system, classify and store the received data according to time series and equipment type, and provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation status of the energy storage system.

[0263] A remote operation and maintenance and fault diagnosis system for an energy storage system, characterized in that it comprises a server and a processor, wherein a system program is stored in the server, and wherein the processor implements the steps of the method when executing the computer program.

[0264] The specific implementation method thereof is as follows: S1: establishing a three-dimensional model of the factory area, defining a global coordinate system for the entire factory area, determining the origin position and the direction of the coordinate axis with a corner of the factory area as the origin, determining the coordinate value of the battery compartment, and determining the coordinate value of the battery relative to the battery compartment; specifically, first defining a global coordinate system for the entire factory area, determining the origin position and the direction of the coordinate axis with a corner of the factory area as the origin, and establishing x, y, and z coordinate axes respectively;

[0265] For each battery compartment, measure its relative position in the factory area, and set the coordinates of the lower left corner of the battery compartment in the global coordinate system to The coordinates of the upper right vertex are

[0266] Normalize the coordinates of the battery compartment and define the normalization function N(X, X min , X max ), and its calculation formula is:

[0267]

[0268] Among them, X represents the coordinate value to be normalized, X min Indicates the coordinate value to be normalized, X max It is the maximum coordinate value range limit value in the corresponding coordinate direction;

[0269] Normalize the coordinates of the battery compartment. In the X direction, the normalized coordinates are:

[0270]

[0271] in is the minimum coordinate value of all battery compartments in the x direction, is the maximum coordinate value of all battery compartments in the x direction;

[0272] Similarly, normalization is performed in the y and z directions to obtain and

[0273] For each battery j, the initial placement position of the lower left vertex in the battery compartment is set to its coordinates in the global coordinate system as

[0274] First, convert the battery coordinates to the local coordinate system of the battery compartment to which it belongs. The calculation formula is as follows:

[0275] In the x-direction:

[0276]

[0277] in represents the coordinate value of battery j in the x direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the x direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the x direction in the global coordinate system;

[0278] In the y direction:

[0279]

[0280] in represents the coordinate value of battery j in the y direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the y direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the y direction in the global coordinate system;

[0281] In the z direction:

[0282]

[0283] in represents the coordinate value of battery j in the z direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the z direction in the global coordinate system, is the coordinate value of the lower left vertex of the battery compartment in the z direction in the global coordinate system;

[0284] Normalize the coordinates of the battery in the local coordinate system of the battery compartment using the normalization function N(x, x min , x max ), in the x direction, the normalized battery coordinates are:

[0285]

[0286] in is the minimum local coordinate value of all batteries in the battery compartment in the x direction, is the maximum local coordinate value of all batteries in the battery compartment in the x direction;

[0287] Similarly, the corresponding normalization operations are performed in the y and z directions to obtain and

[0288] S2: Determine the binding relationship, create a hash table to store the information of the battery and the battery compartment and the binding relationship between them, insert a node into the bucket of the hash table to store the relevant information of the battery; specifically, define a spatial hash function H(x n ,y n , z n), which is used to map the coordinates in the three-dimensional space to the index position in a hash table. The formula is:

[0289]

[0290] where x n ,y n ,z n is the coordinate value of the battery in the corresponding direction after the previous coordinate transformation and normalization operation, p x、 p y、 p z is a pre-selected prime number used to increase the randomness and dispersion of the hash function, m is the size of the hash, (int) represents the rounding operation, and % is the remainder operation;

[0291] Create a hash table to store the information of batteries and battery compartments and the binding relationship between them. Each bucket of the hash table can store a linked list structure to handle possible hash conflicts.

[0292] For each battery compartment, the normalized coordinates The hash index is calculated by the spatial hash function:

[0293]

[0294] In the hash table h i Insert a node into the bucket to store the relevant information of the battery compartment. The battery compartment number is A. i , normalized coordinates

[0295] For each battery j, its coordinates normalized in the local coordinate system of the battery compartment are The hash index is calculated by the spatial hash function:

[0296]

[0297] In the hash table h j Insert a node into the bucket to store battery-related information, including normalized coordinates Battery compartment number A i 、Battery address A j , sensor number S j , Tag T j At the same time, in the node storing the battery information, a pointer is set to point to the node of the battery compartment in the hashtag, thereby establishing a binding relationship between the battery and the battery compartment.

[0298] S3: Collect the voltage, current, and temperature data of each battery at different times from the battery management system to estimate the battery internal resistance; specifically, collect the voltage, current, and temperature data of each battery at different times from the battery management system (BMS). When the battery is in the process of charging and discharging, the modified formula of Ohm's law can be used to estimate the battery internal resistance. The battery internal resistance R b It can be calculated by the following formula:

[0299]

[0300] Among them, R b is the internal resistance of the battery, V oc is the open circuit voltage of the battery, V is the voltage value actually measured at a certain moment during the charge and discharge process of the battery, and I is the current value of the battery corresponding to the measured voltage V during the charge and discharge process;

[0301] By measuring the voltage V and current I and taking the average value, a more accurate estimate of the battery internal resistance can be obtained. The average value of the battery internal resistance is It can be calculated as:

[0302]

[0303] in, It represents the average value of the battery internal resistance obtained after multiple measurements and averaging. n is the total number of times the voltage and current are measured. V i is the voltage value of the battery at the i-th measurement, where i ranges from 1 to n, representing different measurements, I i is the current value of the battery at the i-th measurement. Similarly, the value range of i is from 1 to n, V i is the voltage value obtained at the same time.

[0304] S4: Sequence the collected battery data and annotate them with the historical fault condition related data; specifically, sequence the collected battery data, and the voltage data sequence is {V i}, the current data sequence is I i}, the temperature data sequence is {T i}, where i = 1, 2, ..., n, the average value of the battery internal resistance is calculated from the measured data And set its data sequence to {R i};

[0305] The data associated with the historical fault conditions are annotated, and the historical fault records of each battery in the corresponding time period are collected, and the fault types are grouped as F = {f1, f2, ..., f m}, where m is the number of fault types. For each data point i, a labeling function is defined

[0306]

[0307] Its value comes from the fault type set F or is a null value, and is used to annotate the collected battery data with the corresponding historical fault conditions.

[0308] S5: Process the outliers of the battery data and use the Z-score standardization method to make the variables of different magnitudes have the same weight and comparability in the modeling process. Specifically, the data is processed for noise. For the voltage data series {V i}(i=1,2,…,n), use mean filtering to remove measurement noise, the calculation formula is:

[0309]

[0310] in, is the voltage value after mean filtering, m is the filter window size, V j is a data point in the original voltage data sequence, where the value of j ranges from i-m+1 to i+m-1;

[0311] Similarly, for the current data series {I i}、Resistance data sequence {R i} and temperature data series {T i}The same mean filtering method is used for noise processing. The current value after filtering is The resistance value is and the temperature value is

[0312] To process the voltage outliers, first calculate the mean μ of the voltage data sequence V and standard deviation σ V , the mean calculation formula is:

[0313]

[0314] where μ V is the voltage data sequence {V i}, n is the total number of data points in the voltage data sequence, and the standard deviation is calculated as:

[0315]

[0316] where σ V is the voltage data sequence {V i}, which measures the dispersion of voltage data relative to the mean. According to the 3σ principle, if V i Satisfy |V i -μ V |>3σV , then determine V i As an outlier, delete it from the data series;

[0317] The same method is applied to current, resistance and temperature data to process the abnormal values ​​in their respective data sequences. Let the processed current data sequence be {I' i}, the resistance data sequence is {R' i} and the temperature data sequence is {T' i};

[0318] The Z-score standardization method is used to ensure that variables of different magnitudes have the same weight and comparability in the modeling process;

[0319] For the voltage data sequence {V i}, first calculate the mean μ V and standard deviation σ V , the standardized voltage value The calculation formula is:

[0320]

[0321] in, is the voltage value after Z-score standardization, where μ V is the mean of the voltage data series, σ V is the standard deviation of the voltage data series;

[0322] Similarly, for the current data series {I i}、Resistance data sequence {R i} and temperature data series {T i} also performs Z-score normalization, and the normalized current value is The resistance value is and the temperature value is

[0323] S6: Extract features, including charge and discharge rate, rate of change of battery internal resistance, battery power, and thermal performance; specifically, extract features, wherein the charge and discharge rate is calculated as:

[0324]

[0325] Where CR represents the charge and discharge rate of the battery, I i is the battery current value C measured at the corresponding timestamp i and is the rated capacity of the battery;

[0326] The internal resistance change rate is calculated using the following formula:

[0327]

[0328] Where RR is the rate of change of the battery internal resistance, R i is the battery resistance value calculated at the corresponding time stamp i, R i-1 is the battery resistance value calculated at timestamp i-1, which is equal to R i is the resistance value at adjacent time points, t i Indicates the time point corresponding to timestamp i, t i-1 is the time point at timestamp i-1, which is the same as t i are adjacent time points;

[0329] For power characteristic calculation, the calculation formula is:

[0330] P=V i ×I i

[0331] Where P is the battery power characteristic, V i is the battery voltage value measured at the corresponding timestamp i, I i is the battery current value measured at the corresponding timestamp i;

[0332] The calculation formula for thermal performance related characteristics is:

[0333] HT=T i ×R i

[0334] Where HT is a characteristic related to battery thermal performance, T i is the battery temperature value measured at the corresponding timestamp i, R i is the battery resistance value calculated at the corresponding time stamp i;

[0335] Assume that the charge and discharge rate feature sequence obtained after the above feature extraction is {CR i}, the characteristic sequence of internal resistance change rate is {RR i}, the power characteristic sequence is {P i}, the thermal performance related characteristic sequence is {HT i}.

[0336] S7: Establish an artificial neural network model to process battery-related data to achieve the fault diagnosis prediction goal; thereby identifying the battery fault type; specifically, select a multi-layer perceptron (MLP) as an artificial neural network model to process battery-related data to achieve the fault diagnosis prediction goal, and determine the number of nodes in each layer;

[0337] The number of input layer nodes is set to n, which corresponds to receiving the pre-processed battery data, which includes voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps; let the vector of the eigenvalues ​​obtained after the above feature extraction be the number of input layer nodes n;

[0338] Hidden layer setting, there are h hidden layers, for the i-th hidden layer (i=1,2,…,h), let the number of nodes be

[0339] The number of nodes in the output layer, let the number of nodes in the output layer be m, which corresponds to the prediction target; let there be, k different fault types, and the output value m of each node represents the possibility of the corresponding fault type;

[0340] Define connection weights and biases

[0341] Set the connection weight matrix. For the connection weight matrix from the input layer to the first hidden layer, its elements represents the connection weight from the i-th node in the input layer to the j-th node in the first hidden layer;

[0342] For the connection weight matrix from the kth hidden layer (k>1) to the k+1th hidden layer, its elements are represents the connection weight from the i-th node in the k-th hidden layer to the j-th node in the k+1-th hidden layer;

[0343] Set the bias. For the jth node in the first hidden layer, set its bias to

[0344] For the jth node in the kth hidden layer, let its bias be It plays a similar role in the calculation of each hidden layer;

[0345] Perform forward propagation calculations, where data is gradually propagated from the input layer through the hidden layer to the output layer. Each layer obtains output values ​​according to the corresponding calculation rules, and these output values ​​continue to propagate as inputs to the next layer.

[0346] Input layer to the first hidden layer,

[0347] The input vector is X = [x1, x2, ..., x n ], where x i The calculation formula of the output vector Y1 of the first hidden layer node is:

[0348]

[0349] in, is the output value of the jth node in the first hidden layer, and f is the activation function oSigmoid function: Its output value is between 0 and 1, and the output function is the probability estimation oRelU function of fault occurrence: f(x) = max(0,x), and its output is x itself when x>0, and the output is 0 when x≤0; The connection weight from the i-th node in the input layer to the j-th node in the first hidden layer determines the influence of the information of the input layer node on the j-th node in the first hidden layer. is the bias of the jth node in the first hidden layer, which is used to offset the input information and affects the output value of the jth node in the first hidden layer together with the connection weight;

[0350] The propagation from the kth hidden layer to the k+1th hidden layer, when k>1, the output vector Y of the kth hidden layer node k The calculation formula is:

[0351]

[0352] Where Z j represents the output value of the jth node in the output layer, f is the Sigmoid activation function used to output the probability of a fault. is the connection weight from the i-th node in the last hidden layer to the j-th node in the output layer, is the bias of the jth node in the output layer, x i Refers to the corresponding element of the input vector of the last hidden layer node;

[0353] Define the training set and loss function

[0354] Assume that the training set data is {(X1,Y1),(X2,Y2),…,(X N ,N N )}, where X i is the input vector, including voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps. i is the corresponding output vector, indicating the possibility of occurrence of different fault types;

[0355] The loss function uses the mean square error (MSE) function and defines the loss function as:

[0356]

[0357] Where L represents the value of the loss function. The smaller the value of the loss function, the closer the model's prediction result is to the actual situation, and the better the model's performance is. N is the number of samples in the training set data, that is, the number of input-output pairs contained in the training set. i The model is based on the input vector X i The output value obtained by forward propagation is the prediction result of the neural network for the corresponding input, Y i is the same as the input vector X i The corresponding actual output value is specifically the real label determined by the actual fault condition of the battery;

[0358] The back propagation algorithm is based on the principle of gradient descent. It calculates the gradient of the loss function for the connection weights and biases of each layer, and then adjusts these parameters according to the gradient, so that the value of the loss function is continuously reduced, thereby optimizing the model.

[0359] Calculate the loss function L for the output value Z of the output layer node j The gradient is:

[0360]

[0361] in Represents the loss function L for the output value Z of the output layer node j The gradient of Z j is the output value of the jth node in the output layer, Y j is the actual output value corresponding to the jth node in the output layer;

[0362] According to the chain rule, the gradient of the loss function for the connection weights and biases of each layer is calculated step by step. For the connection weights from the output layer to the last hidden layer, The gradient calculation formula is:

[0363]

[0364] in, The bias of the jth node in the output layer is obtained by derivation according to the calculation formula of the output vector of the output layer node. The gradient calculation formula is:

[0365]

[0366] For the connection weights and biases between hidden layers, we also use a similar method. According to the chain rule, we gradually calculate the gradients of each layer to get the gradient of the loss function. For the connection weights from the kth hidden layer to the k+1th hidden layer, The gradient calculation formula is:

[0367]

[0368] in, The bias of the jth node in the kth hidden layer is obtained by derivation according to the calculation formula of the corresponding hidden layer node output vector. The gradient calculation formula is:

[0369]

[0370] According to the calculated gradients of the connection weights and biases of each layer, these parameters are adjusted according to a certain learning rate η. The update formula is:

[0371]

[0372] in, is the updated connection weight value, η is the learning rate, which controls the step size of model parameter update, and for bias The update formula is:

[0373]

[0374] The connection weights and biases of other layers are updated according to the same update formula. By continuously repeating the above process, that is, forward propagation to calculate the output value, calculate the loss function, back propagation to calculate the gradient, and update the parameters, the value of the loss function is continuously reduced until the predetermined training termination condition is reached, thereby achieving model optimization.

[0375] S8: Determine the position of the abnormal battery, and determine the position of the abnormal battery in the global coordinate system of the entire factory through the normalized coordinate information of the battery compartment local coordinate system stored in the hash table, and the normalized coordinate information of the battery compartment in the global coordinate system; specifically, in the process of establishing the three-dimensional modeling of the factory, each battery stores relevant information in the hash table, including the battery number, normalized coordinates, and the battery compartment number A to which it belongs. i , the battery failure results calculated by the artificial neural network are also integrated into this data structure, and the failure results output by the artificial neural network for each battery are set to a vector F = [f1, f2, ..., f m ], for each battery, store its fault result vector F in the hash table, and in the node storing battery-related information, add a new field to store this fault result vector;

[0376] When the artificial neural network model identifies the battery fault vector F = [f1, f2, …, f m], the battery is marked as an abnormal battery by adding a new field, and its battery number, the number of the battery compartment to which it belongs, and the normalized coordinate information in the battery compartment are recorded. For the abnormal battery, first obtain its normalized coordinate value in the local coordinate system of the battery compartment, and use the normalized coordinate information in the local coordinate system of the battery compartment stored in the hash table, as well as the normalized coordinate information of the battery compartment in the global coordinate system, to determine the position of the abnormal battery in the global coordinate system of the entire factory.

[0377] S9: Establish a remote operation and maintenance center to provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation of the energy storage system. Establish a remote operation and maintenance center to transmit the collected data to the remote operation and maintenance center through wired or wireless communication, and set up a server in the remote operation and maintenance center to receive data transmitted from the energy storage system, classify and store the received data according to time series and equipment type, and provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation of the energy storage system.

Claims

1. A remote operation and maintenance and fault diagnosis method for an energy storage system, characterized in that: The following steps are involved: S1: Establish a 3D model of the factory: define a global coordinate system for the entire factory, determine the origin position and coordinate axis direction with a corner of the factory as the origin, determine the coordinate value of the battery compartment, and determine the coordinate value of the battery relative to the local coordinate system of the battery compartment; S2: Determine the binding relationship: Create a hash table to store the information of the battery and the battery compartment and the binding relationship between them, insert a node into the bucket of the hash table to store the relevant information of the battery; S3: Collect the voltage, current, and temperature data of each battery at different times from the battery management system to estimate the battery internal resistance; S4: Sequence the collected battery data and annotate the data associated with historical fault conditions; S5: Outlier processing of battery data: The Z-score standardization method is used to make variables of different magnitudes have the same weight and comparability in the modeling process; S6: Extract features: including charge and discharge rate, change rate of battery internal resistance, battery power, and thermal performance; S7: Establishing an artificial neural network model: importing the extracted features to achieve the fault diagnosis prediction goal, thereby identifying the battery fault type; S8: Determine the position of the abnormal battery: Determine the position of the abnormal battery in the global coordinate system of the entire factory area through the normalized coordinate information of the abnormal battery in the local coordinate system of the battery compartment stored in the hash table and the normalized coordinate information of the battery compartment in the global coordinate system; S9: Establish a remote operation and maintenance center to provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation status of the energy storage system.

2. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: The three-dimensional model of the factory area is established in step S1 by first defining a global coordinate system for the entire factory area, determining the origin position and the direction of the coordinate axis with a corner of the factory area as the origin, and establishing the x, y, and z coordinate axes respectively; For each battery compartment, measure its relative position in the factory area, and set the coordinates of the lower left corner of the battery compartment in the global coordinate system to The coordinates of the upper right vertex are Normalize the coordinates of the battery compartment and define the normalization function N(X, X min , X max ), and its calculation formula is: Among them, X represents the coordinate value to be normalized, X min Indicates the coordinate value to be normalized, X max It is the maximum coordinate value range limit value in the corresponding coordinate direction; Normalize the coordinates of the battery compartment. In the X direction, the normalized coordinates are: in is the minimum coordinate value of all battery compartments in the x direction, is the maximum coordinate value of all battery compartments in the x direction; Similarly, normalization is performed in the y and z directions to obtain and For each battery j, the initial placement position of the lower left vertex in the battery compartment is set to its coordinates in the global coordinate system as First, convert the battery coordinates to the local coordinate system of the battery compartment to which it belongs. The calculation formula is as follows: In the x-direction: in represents the coordinate value of battery j in the x direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the x direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the x direction in the global coordinate system; In the y direction: in represents the coordinate value of battery j in the y direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the y direction in the global coordinate system, is the coordinate value of the lower left corner of the battery compartment in the y direction in the global coordinate system; In the z direction: in represents the coordinate value of battery j in the z direction in the local coordinate system of the battery compartment, is the initial placement coordinate value of battery j in the z direction in the global coordinate system, is the coordinate value of the lower left vertex of the battery compartment in the z direction in the global coordinate system; Normalize the coordinates of the battery in the local coordinate system of the battery compartment using the normalization function N(x, x min , x max ), in the x direction, the normalized battery coordinates are: in is the minimum local coordinate value of all batteries in the battery compartment in the x direction, is the maximum local coordinate value of all batteries in the battery compartment in the x direction; Similarly, the corresponding normalization operations are performed in the y and z directions to obtain and 3. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: In step S2, the binding relationship is determined by defining a spatial hash function H(x n ,y n , z n ), which is used to map the coordinates in the three-dimensional space to the index position in a hash table. The formula is: H(x n ,y n ,z n )=((int)(x n *p x +y n *p y +z n *p z )%m) where x n ,y n ,z n is the coordinate value of the battery in the corresponding direction after the previous coordinate transformation and normalization operation, p x、 p y、 p z is a pre-selected prime number used to increase the randomness and dispersion of the hash function, m is the size of the hash, (int) represents the rounding operation, and % is the remainder operation; Create a hash table to store the information of batteries and battery compartments and the binding relationship between them. Each bucket of the hash table can store a linked list structure to handle possible hash conflicts. For each battery compartment, the normalized coordinates The hash index is calculated by the spatial hash function: In the hash table h i Insert a node into the bucket to store the relevant information of the battery compartment. The battery compartment number is A. i , normalized coordinates For each battery j, its coordinates normalized in the local coordinate system of the battery compartment are The hash index is calculated by the spatial hash function: In the hash table h j Insert a node into the bucket to store battery-related information, including normalized coordinates Battery compartment number A i 、Battery address A j , sensor number S j , Tag T j At the same time, in the node storing the battery information, a pointer is set to point to the node of the battery compartment in the hashtag, thereby establishing a binding relationship between the battery and the battery compartment.

4. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: In step S3, the battery internal resistance is estimated by collecting the voltage, current and temperature data of each battery at different times from the battery management system BMS. When the battery is in the process of charging and discharging, the battery internal resistance is estimated by using the modified formula of Ohm's law. The battery internal resistance R b Calculated by the following formula: Among them, R b is the internal resistance of the battery, V oc is the open circuit voltage of the battery, V is the voltage value actually measured at a certain moment during the charge and discharge process of the battery, and I is the current value of the battery corresponding to the measured voltage V during the charge and discharge process; By measuring the voltage V and current I and taking the average value, a more accurate estimate of the battery internal resistance can be obtained. The average value of the battery internal resistance is It can be calculated as: in, It represents the average value of the battery internal resistance obtained after multiple measurements and averaging. n is the total number of times the voltage and current are measured. V i is the voltage value of the battery at the i-th measurement, where i ranges from 1 to n, representing different measurements, I i is the current value of the battery at the i-th measurement. Similarly, the value range of i is from 1 to n, V i is the voltage value obtained at the same time.

5. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: In step S4, the collected battery data is sequenced and annotated with the historical fault condition related data. Specifically, the collected battery data is sequenced, and the voltage data sequence is {V i }, the current data sequence is {I i }, the temperature data sequence is {T i }, where i = 1, 2, ..., n, the average value of the battery internal resistance is calculated from the measured data And set its data sequence to {R i }; The data associated with the historical fault conditions are annotated, and the historical fault records of each battery in the corresponding time period are collected, and the fault types are grouped as F = {f1, f2, ..., f m }, where m is the number of fault types. For each data point i, a labeling function is defined Its value comes from the fault type set F or is a null value, and is used to annotate the collected battery data with the corresponding historical fault conditions.

6. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: In step S5, the abnormal value processing of the battery data is specifically to process the data with noise. i }(i=1,2,…,n), use mean filtering to remove measurement noise, the calculation formula is: in, is the voltage value after mean filtering, m is the filter window size, V j is a data point in the original voltage data sequence, where the value of j ranges from i-m+1 to i+m-1; Similarly, for the current data series {I i }、Resistance data sequence {R i } and temperature data series {T i }The same mean filtering method is used for noise processing. The current value after filtering is The resistance value is and the temperature value is To process the voltage outliers, first calculate the mean μ of the voltage data sequence V and standard deviation σ V , the mean calculation formula is: where μ V is the voltage data sequence {V i }, n is the total number of data points in the voltage data sequence, and the standard deviation is calculated as: where σ V is the voltage data sequence {V i }, which measures the dispersion of voltage data relative to the mean. According to the 3σ principle, if V i Satisfy |V i -μ V |>3σ V , then determine V i As an outlier, delete it from the data series; The same method is applied to current, resistance and temperature data to process the abnormal values ​​in their respective data sequences. Let the processed current data sequence be {I' i }, the resistance data sequence is {R' i } and the temperature data sequence is {T' i }; The Z-score standardization method is used to ensure that variables of different magnitudes have the same weight and comparability in the modeling process; For the voltage data sequence {V i }, first calculate the mean μ V and standard deviation σ V , the standardized voltage value The calculation formula is: in, is the voltage value after Z-score standardization, where μ V is the mean of the voltage data series, σ V is the standard deviation of the voltage data series; Similarly, for the current data series {I i }、Resistance data sequence {R i } and the temperature data series {T i } also performs Z-score normalization, and the normalized current value is The resistance value is and the temperature value is 7. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: The feature extraction in step S6 is specifically to extract the feature, wherein the charge and discharge rate is calculated as: Where CR represents the charge and discharge rate of the battery, I i is the battery current value measured at the corresponding timestamp i, and C is the rated capacity of the battery; The internal resistance change rate is calculated using the following formula: Where RR is the rate of change of the battery internal resistance, R i is the battery resistance value calculated at the corresponding time stamp i, R i-1 is the battery resistance value calculated at timestamp i-1, which is equal to R i is the resistance value at adjacent time points, t i Indicates the time point corresponding to timestamp i, t i-1 is the time point at timestamp i-1, which is the same as t i are adjacent time points; For power characteristic calculation, the calculation formula is: P=V i ×I i Where P is the battery power characteristic, V i is the battery voltage value measured at the corresponding timestamp i, I i is the battery current value measured at the corresponding timestamp i; The calculation formula for thermal performance related characteristics is: HT=T i ×R i Where HT is a characteristic related to battery thermal performance, T i is the battery temperature value measured at the corresponding timestamp i, R i is the battery resistance value calculated at the corresponding time stamp i; Assume that the charge and discharge rate feature sequence obtained after the above feature extraction is {CR i }, the characteristic sequence of internal resistance change rate is {RR i }, the power characteristic sequence is {P i }, the thermal performance related characteristic sequence is {HT i }.

8. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: The establishment of the artificial neural network model in step S7 specifically includes selecting a multi-layer perceptron MLP as the artificial neural network model to process battery-related data to achieve the fault diagnosis prediction goal, and determining the number of nodes in each layer; The number of input layer nodes is set to n, which corresponds to receiving the pre-processed battery data, which includes voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps; let the vector of the eigenvalues ​​obtained after the above feature extraction be the number of input layer nodes n; Hidden layer setting, there are h hidden layers, for the i-th hidden layer (i=1,2,…,h), let the number of nodes be The number of nodes in the output layer, let the number of nodes in the output layer be m, which corresponds to the prediction target; let there be, k different fault types, and the output value m of each node represents the possibility of the corresponding fault type; Define the connection weights and biases, set the connection weight matrix, and for the connection weight matrix from the input layer to the first hidden layer, its elements represents the connection weight from the i-th node in the input layer to the j-th node in the first hidden layer; For the connection weight matrix from the kth hidden layer (k>1) to the k+1th hidden layer, its elements are represents the connection weight from the i-th node in the k-th hidden layer to the j-th node in the k+1-th hidden layer; Set the bias. For the jth node in the first hidden layer, set its bias to For the jth node in the kth hidden layer, let its bias be It plays a similar role in the calculation of each hidden layer; Perform forward propagation calculations, where data is gradually propagated from the input layer through the hidden layer to the output layer. Each layer obtains output values ​​according to the corresponding calculation rules, and these output values ​​continue to propagate as inputs to the next layer. From the input layer to the first hidden layer, the input vector is X = [x1, x2, ..., x n ], where x i The calculation formula of the output vector Y1 of the first hidden layer node is: in, is the output value of the jth node in the first hidden layer, and f is the activation function oSigmoid function: Its output value is between 0 and 1, and the output function is the probability estimation oRelU function of fault occurrence: f(x) = max(0,x), and its output is x itself when x>0, and the output is 0 when x≤0; The connection weight from the i-th node in the input layer to the j-th node in the first hidden layer determines the influence of the information of the input layer node on the j-th node in the first hidden layer. is the bias of the jth node in the first hidden layer, which is used to offset the input information and affects the output value of the jth node in the first hidden layer together with the connection weight; The propagation from the kth hidden layer to the k+1th hidden layer, when k>1, the output vector Y of the kth hidden layer node k The calculation formula is: Where Z j represents the output value of the jth node in the output layer, f is the Sigmoid activation function used to output the probability of a fault. is the connection weight from the i-th node in the last hidden layer to the j-th node in the output layer, is the bias of the jth node in the output layer, x i Refers to the corresponding element of the input vector of the last hidden layer node; Define the training set and loss function Assume that the training set data is {(X1,Y1),(X2,Y2),…,(X N ,N N )}, where X i is the input vector, including voltage, current, resistance, temperature, and derived features such as charge and discharge rate, internal resistance change rate, power characteristics, and thermal performance extracted through the previous data feature engineering steps. i is the corresponding output vector, indicating the possibility of occurrence of different fault types; The loss function uses the mean square error MSE function and defines the loss function as: Where L represents the value of the loss function. The smaller the value of the loss function, the closer the model's prediction result is to the actual situation, and the better the model's performance is. N is the number of samples in the training set data, that is, the number of input-output pairs contained in the training set. i The model is based on the input vector X i The output value obtained by forward propagation is the prediction result of the neural network for the corresponding input, Y i is the same as the input vector X i The corresponding actual output value is specifically the real label determined by the actual fault condition of the battery; The back propagation algorithm is based on the principle of gradient descent. It calculates the gradient of the loss function for the connection weights and biases of each layer, and then adjusts these parameters according to the gradient, so that the value of the loss function is continuously reduced, thereby optimizing the model. Calculate the loss function L for the output value Z of the output layer node j The gradient is: in Represents the loss function L for the output value Z of the output layer node j The gradient of Z j is the output value of the jth node in the output layer, Y j is the actual output value corresponding to the jth node in the output layer; According to the chain rule, the gradient of the loss function for the connection weights and biases of each layer is calculated step by step. For the connection weights from the output layer to the last hidden layer, The gradient calculation formula is: in, The bias of the jth node in the output layer is obtained by derivation according to the calculation formula of the output vector of the output layer node. The gradient calculation formula is: For the connection weights and biases between hidden layers, we also use a similar method. According to the chain rule, we gradually calculate the gradients of each layer to get the gradient of the loss function. For the connection weights from the kth hidden layer to the k+1th hidden layer, The gradient calculation formula is: in, The bias of the jth node in the kth hidden layer is obtained by derivation according to the calculation formula of the corresponding hidden layer node output vector. The gradient calculation formula is: According to the calculated gradients of the connection weights and biases of each layer, these parameters are adjusted according to a certain learning rate η. The update formula is: in, is the updated connection weight value, η is the learning rate, which controls the step size of model parameter update, and for bias The update formula is: The connection weights and biases of other layers are updated according to the same update formula. By continuously repeating the above process, that is, forward propagation to calculate the output value, calculate the loss function, back propagation to calculate the gradient, and update the parameters, the value of the loss function is continuously reduced until the predetermined training termination condition is reached, thereby achieving model optimization.

9. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: In step S8, the abnormal battery location is determined. Specifically, in the process of establishing the three-dimensional modeling of the factory area, each battery stores relevant information in the hash table, including the battery number, normalized coordinates, and the battery compartment number A to which it belongs. i , the battery failure results calculated by the artificial neural network are also integrated into this data structure, and the failure results output by the artificial neural network for each battery are set to a vector F = [f1, f2, ..., f m ], for each battery, store its fault result vector F in the hash table, and in the node storing battery-related information, add a new field to store this fault result vector; When the artificial neural network model identifies the battery fault vector F = [f1, f2, …, f m ], the battery is marked as an abnormal battery by adding a new field, and its battery number, the number of the battery compartment to which it belongs, and the normalized coordinate information in the battery compartment are recorded. For the abnormal battery, first obtain its normalized coordinate value in the local coordinate system of the battery compartment, and use the normalized coordinate information in the local coordinate system of the battery compartment stored in the hash table, as well as the normalized coordinate information of the battery compartment in the global coordinate system, to determine the position of the abnormal battery in the global coordinate system of the entire factory.

10. A method for remote operation and maintenance and fault diagnosis of an energy storage system according to claim 1, characterized in that: In step S9, a remote operation and maintenance center is established to transmit the collected data to the remote operation and maintenance center through wired or wireless communication, and a server is set up in the remote operation and maintenance center to receive the data transmitted from the energy storage system, classify and store the received data according to time series and equipment type, and provide detailed operation and maintenance guidance and suggestions to on-site operation and maintenance personnel based on the fault diagnosis results and the real-time operation status of the energy storage system.

11. A remote operation and maintenance and fault diagnosis system for an energy storage system, characterized in that: The method comprises a server and a processor, wherein the server stores a system program, and is characterized in that the processor implements the steps of the method according to any one of claims 1 to 9 when executing the computer program.

Images