A lithium battery energy storage station cloud edge coordination intelligent fire-fighting linkage system and implementation method

Abstract

The application discloses a lithium battery energy storage station cloud-edge collaborative intelligent fire-fighting linkage system and an implementation method. It comprises a fire service cloud platform system, a network layer, a sensing layer and a control layer. The sensing layer reports various data in real time. The network layer extracts a feature vector set to construct an effective feature parameter identification classifier. The control layer realizes comprehensive linkage fire extinguishing control. The fire service cloud platform system is used for processing data logic of a program and is responsible for accessing data in a database. The application has the beneficial effect that a plurality of operating state parameters representing lithium batteries of an energy storage station are preprocessed on an edge gateway side, such as feature extraction, statistical analysis, dimension reduction optimization and the like, and then trained and dynamically identified through an improved BP neural network model deployed in an intelligent fire service cloud platform. According to the identification result, early warning information is sent out. In combination with a fire-fighting linkage control system, a complete cloud-edge collaborative intelligent fire-fighting linkage control method is formed, so that the operating safety risk is greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium battery energy storage stations, in particular to a lithium battery energy storage station cloud-edge collaborative intelligent fire-fighting linkage system and implementation method. BACKGROUND

[0002] With the proposal of the double carbon target and the development of digital traction of new power system construction, electrochemical energy storage stations play a very important role in green energy conversion due to their relatively mature technology and high cost performance. However, as the installed capacity of energy storage stations continues to increase, safety hazards of lithium battery energy storage stations begin to emerge. There are numerous fire and explosion accidents in domestic and foreign energy storage power stations, not only causing significant property losses, but also seriously threatening the safety of personnel's lives, so safety monitoring of lithium battery energy storage stations cannot be delayed.

[0003] In the current research on lithium battery early warning, some domestic research literature methods are based on the temperature of energy storage batteries for early warning, analyze the internal temperature of the battery according to the battery thermal model and electrochemical impedance spectrum, and establish a corresponding model to predict the overall temperature distribution of the battery. The temperature of the electric field is transmitted from the inside to the surface with a time difference of several seconds, and the calculation process of the thermal model is relatively complex. There are also methods for detecting the content of specific gases in the battery cabin to determine the thermal runaway of the battery, and sampling and detecting the gases after thermal decomposition to establish a multi-level early warning system for energy storage power stations. Due to the slow diffusion of gases caused by the blocking of the battery rack, the early warning sensitivity is poor, and the location information of the faulty battery cannot be directly obtained. In the system for detecting and analyzing acoustic signals, an acoustic emission measurement system is applied, and the correlation between the battery fault signal and the acoustic emission spectrum is statistically analyzed, the characteristic values of the de-noised acoustic signals are extracted, and a recognition classifier is constructed according to the extracted characteristic values. However, in the process of extracting the lithium battery fault signal, the internal information of the signal cannot be obtained, and it is also susceptible to noise interference. SUMMARY

[0004] The present application is to overcome the above-mentioned deficiencies in the prior art, and provides a lithium battery energy storage station cloud-edge collaborative intelligent fire-fighting linkage system and implementation method with small operation safety risk.

[0005] In order to achieve the above-mentioned purpose, the following technical solutions are adopted in the present application:

[0006] A lithium battery energy storage station cloud-edge collaborative intelligent fire-fighting linkage system, comprising a fire service cloud station system, a network layer, a sensing layer and a control layer, wherein:

[0007] The sensing layer comprises various sensor devices communicating with the gateway through RS485 modbusrtu protocol and reporting various data to the network layer in real time;

[0008] The network layer comprises a gateway with an edge computing function, a network switch and a firewall arranged at the energy storage station, and is mainly responsible for classifying and analyzing various sensor multi-source heterogeneous data reported by the sensing layer, projecting high-dimensional feature samples to low dimensions through wavelet transform, extracting feature information contained in the original samples, retaining positively correlated feature vectors, and reporting the extracted feature vector set to the fire service platform for training to construct an effective feature parameter recognition classifier.

[0009] The control layer realizes comprehensive linkage fire extinguishing control by associating different types of detection devices, combining early warning information issued by the fire service cloud platform through the network layer, and the like.

[0010] The front end of the fire service cloud platform system adopts an MVC development mode and is connected with the network layer, is used for processing data logic of programs, and is responsible for accessing data in a database.

[0011] In view of the characteristics of complex structure, great operation safety risk and fast thermal runaway speed of current large-scale lithium battery energy storage batteries, and in combination with the fact that current monitoring of lithium battery energy storage stations stays at a single device or terminal and is completely isolated from a fire extinguishing system, the application provides a lithium battery energy storage station cloud-edge collaborative intelligent fire extinguishing linkage system and an implementation method thereof.

[0012] As a preferred, the fire service cloud platform system comprises an application service layer, a data analysis layer and a data management layer, the data management layer is connected with the network layer, the application service layer and the data analysis layer, SQLServer and MYSQL databases are adopted in the data management layer, corresponding database tables are designed according to functions provided by each module of the system, and meanings and constraint relationships of fields in the database tables are displayed through the tables; the data analysis layer provides analysis, processing, recognition, matching and judgment services for time-sensitive data reported by the gateway and non-time-sensitive data stored in the data management layer; and the application service layer provides an operation interface for system users and provides on-site device state display and manual control functions for the system users.

[0013] The application further provides an implementation method of the lithium battery energy storage station cloud-edge collaborative intelligent fire extinguishing linkage, and the method comprises the following steps.

[0014] (1) The perception layer acquires data and reports to the network layer through the modbusrtu protocol;

[0015] (2) The network layer uses edge computing technology to pre-process data and extract a feature vector matrix, which is reported to the fire service cloud platform by the gateway;

[0016] (3) The improved BP neural network model deployed in the fire service cloud platform simulates and trains the first 30 data and obtains a feature vector recognition library; at the same time, the feature vector recognition library is constantly updated according to the subsequent uploaded data;

[0017] (4) Researchers manually input a historical year's lithium battery energy storage station fire data as an anomaly database into the fire service cloud platform. The anomaly database serves as an important reference for predicting fires, and the database is constantly updated when receiving subsequent abnormal data, thereby improving the accuracy of the judgment;

[0018] (5) The improved BP neural network model is used to dynamically identify and diagnose the subsequent collected data by comparing them with the normal data in the feature vector recognition library; when abnormal data appears, the anomaly database is compared to analyze and diagnose whether a fault has occurred inside the energy storage station; if so, a fire event file is established separately;

[0019] (6) For the result of diagnosing the fault, a fire control command is sent to the edge side fire linkage control layer at the same time as the warning information is sent;

[0020] (7) After receiving the fire extinguishing command, the fire linkage control layer immediately starts the alarm information, opens the fire extinguishing material safety valve, releases the fire extinguishing material, and cuts off all electrical circuits in the lithium battery energy storage warehouse; at the same time, the gateway sends the fire information to the fire service cloud platform, notifying the fire personnel and electrical maintenance personnel to arrive at the scene in time for subsequent fire investigation and electrical maintenance work;

[0021] (8) The fire service cloud platform analyzes this fire, corrects the fire event file, and updates the anomaly database.

[0022] As a preferred, in step (2), the data pre-processing step using edge computing technology is as follows:

[0023] (21) Obtain the wavelet basis function:

[0024] Set the wavelet transform L 2 (R) represents a signal with limited energy, and its Fourier transform is Ψ(w); when The condition is met:

[0025]

[0026] Let Ψ(ω) be the mother wavelet spectrum function, w represents frequency, t represents time, and the mother wavelet A wavelet sequence can be obtained through proper linear transformation:

[0027]

[0028] In the above formula, a and b are the scale and translation factors of the mother wavelet, respectively;

[0029] (22) Discretization of the above formula gives the discrete wavelet function:

[0030]

[0031] (23) For any input sound signal f(t) ∈ L 2 (R), the discrete wavelet transform is:

[0032]

[0033] (24) The discrete wavelet is decomposed into four layers according to the approximation value and the noise value, as shown in the following formula:

[0034] w f (j, k) = C A4 +C D4 +C D3 +C D2 +C D1

[0035] In the formula, C Ai is the wavelet decomposition approximation part, and C Di represents the part where the noise is located.

[0036] (25) The threshold size is determined according to the sample estimation;

[0037] (26) The wavelet coefficients are reorganized according to the threshold value;

[0038] (27) The wavelet energy spectrum after processing the wavelet coefficients is inversely transformed to reconstruct the time signal

[0039]

[0040] In the above formula, a = 2 j , b = 2 j k, j, and k are positive integers.

[0041] (28) Establishing a de-noising evaluation index function, here the signal-to-noise ratio is introduced to evaluate the signal de-noising performance evaluation index, for the de-noising effect that fails to meet the requirements, the reconstructed signal is returned to step (24) again for decomposition, and the second threshold judgment is carried out, and then reconstruction is carried out, if the signal after the second reconstruction still cannot meet the evaluation index requirement, the signal is discarded, and the evaluation index function is as follows:

[0042]

[0043] (29) The reconstructed signal after wavelet de-noising and meeting the required signal-to-noise ratio evaluation index needs to be subjected to principal component analysis PCA for eigenvalue extraction.

[0044] As preferred, in step (25), the selection and setting of the threshold value is the key in the wavelet de-noising process, by removing the wavelet coefficients that do not meet the set threshold value requirement and retaining the remaining coefficients, a good de-noising effect is achieved; according to the background noise characteristics of the battery prefabricated cabin and considering the computing capacity of the edge gateway, a fixed maximum minimum criterion method is used to determine the threshold value, so as to reduce the calculation overhead and improve the calculation efficiency; the maximum minimum criterion threshold value method is shown in the following formula:

[0045]

[0046] In the above formula, θ is the standard deviation of the noise, and n is the number of wavelet coefficients at each scale.

[0047] As preferred, in step (26), wavelet transform of a given signal is to expand the signal according to a certain wavelet function cluster, that is, to express the signal as a linear combination of a series of wavelet functions of different scales and different time shifts, wherein the coefficient of each term is called a wavelet coefficient, and specifically, the wavelet coefficients greater than the set threshold value are retained, and the remaining ones are zeroed, as shown in the following formula:

[0048]

[0049] Wherein, is the set of wavelet coefficients after judgment, x is the wavelet coefficient, and T is the threshold value.

[0050] As preferred, in step (29), the specific steps of the principal component analysis method are as follows:

[0051] (291) Assuming that the input data source has n samples and p indexes, an nxp sample matrix x can be formed:

[0052]

[0053] (292) Calculate the average value of the characteristic vector:

[0054]

[0055] (293) Find the covariance matrix S of the eigenvectors, i.e.:

[0056]

[0057] (294) Calculate the eigenvalues ​​and eigenvectors of the standardized sample covariance:

[0058] Eigenvalues: λ1, λ2, ..., λ n Feature vectors: v1, v2, ..., v n ;

[0059] (295) Calculate the contribution rate η of each component. i and cumulative contribution rate β k :

[0060]

[0061]

[0062] (296) Construct the eigenvector mapping matrix:

[0063] Sort the principal components by their contribution rates, select the p largest principal components that meet the requirements, and then use their corresponding p eigenvectors to form an eigenvector matrix V as a mapping matrix. Transform the denoised and reconstructed signal into the feature space, i.e.:

[0064] Y = V T X

[0065] In the above formula: Y is the projected data matrix, i.e., the principal component matrix of the original signal after PCA processing; V is the eigenvector matrix, V = [v1, v2, ..., v p ], V T Let X be the transpose of the matrix, and let X be the average of the eigenvectors.

[0066] Preferably, in step (3), the improved BP neural network model is based on the following principle:

[0067] The output of the forward propagation of the neural network can be expressed as:

[0068] bp_output = f(∑WX)

[0069] In the above formula, W represents the weight matrix, X represents the input feature vector matrix, and f represents the activation function. The above formula provides the transmission information from the input layer to the hidden layer. During forward propagation, an error function is used to verify whether the training process has ended. Training ends when the output error is less than a predetermined value; otherwise, backward propagation is performed. The error function is calculated as follows:

[0070]

[0071] In the above formula, E represents the error function, and d j Indicates the actual output, o j This represents the expected output, j represents the neuron number in the network, and d represents the desired output. j o j Here, d and o represent the actual and expected outputs of the j-th neuron, respectively, while d and o represent the actual and expected outputs of the entire neuron. n refers to the total number of neurons in the network. The computational error of the output neural network can be calculated using the above formula. The error signal obtained from each layer is used to adjust the weights between neurons. The backpropagation process of the error can be represented as:

[0072]

[0073] In the above formula, v kj z represents the connection weight. k The above formula represents the hidden layer, k represents the number of hidden layers, and n represents the total number of neurons. Backpropagation of error is performed using this formula. The forward propagation of the SNN network goes from the encoding layer to the output layer. The state variables of the neurons are:

[0074]

[0075] In the above formula, O j Represents the state variables of an SNN. Represents the terminal weights in an SNN. Let represent the delayed signal of a neuron in the SNN, m represent the total number of neurons, i represent the impulse response coefficient, k and j are positive integers, and t refer to time. After adjusting all the weights of the neural network model using the above formula, the input signal propagates forward. Therefore, the output vector matrix of the SNN neural network output layer can be represented as:

[0076]

[0077] In the above formula, w i Represents the output layer weight coefficients, where η represents the weighting exponent. Let b represent the state vector. i Δw represents the pre-output vector of the output layer. r The updated weight coefficients are represented by r, which represents the number of neurons connecting the coding layer, and x. i This refers to the sample matrix with impulse response coefficient i; the output vector matrix of the improved BP neural network can be calculated using the above formula, and the above model can be directly used for the training and parameter identification of fire service pan-tilt units.

[0078] The application has the beneficial effects that: by using edge computing technology and cloud edge cooperation, a plurality of operation state parameters characterizing lithium batteries of energy storage stations are preprocessed on an edge gateway side, such as feature extraction, statistical analysis, dimension reduction optimization, etc., and then trained and dynamically identified through an improved BP neural network model deployed on an intelligent fire control cloud station, early warning information is issued according to the identification result, a complete cloud edge cooperation intelligent fire control linkage control method is formed in combination with a fire control linkage control system, and the operation safety risk is greatly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0079] Figure 1 is a system structure block diagram of the application;

[0080] Figure 2 is a method flowchart of the application;

[0081] Figure 3 is a method flowchart of data preprocessing;

[0082] Figure 4 is an improved BP neural network model diagram. DETAILED DESCRIPTION

[0083] The application will be further described below in combination with the drawings and specific embodiments.

[0084] As Figure 1 described in the embodiments, a lithium battery energy storage station cloud edge cooperation intelligent fire control linkage system includes a fire service cloud station system, a network layer, a perception layer and a control layer, wherein:

[0085] The perception layer includes various sensor devices that communicate with the gateway through the RS485 modbusrtu protocol and report various data to the network layer in real time; the perception layer mainly includes various sensor devices, including battery metering devices, temperature sensors, sound wave sensors, gas sensors, infrared detectors, smoke detectors and the like;

[0086] The network layer includes a gateway with edge computing function, a network switch and a firewall deployed in the energy storage station, wherein the gateway with edge computing function is the core node of building the edge computing of the system, mainly responsible for classifying and analyzing the multi-source heterogeneous data of various sensors reported by the perception layer, then projecting the high-dimensional feature samples to low dimension through wavelet transform, extracting the feature information contained in the original samples, and retaining the positively correlated feature vectors, and reporting the extracted feature vector set to the fire service platform for training, and constructing an effective feature parameter recognition classifier;

[0087] The control layer mainly includes detector linkage alarm, alarm device, fire extinguishing material, fire extinguishing device linkage and regional multi-controller linkage, through the association of different types of detection devices, through the early warning information issued by the fire service cloud platform through the network layer, the comprehensive linkage fire control is realized;

[0088] The front end of the fire service cloud platform system adopts the development mode of MVC, is connected with the network layer, is used for processing the data logic of the program, is responsible for accessing the data in the database, the back end of the fire service cloud platform system adopts the lightweight server framework of Springboot. The fire service cloud platform system includes an application service layer, a data analysis layer and a data management layer, the application service layer, the data analysis layer and the data management layer all belong to part of the background data processing, the data management layer is connected with the network layer, the application service layer and the data analysis layer, wherein the SQL Server and the MYSQL database are adopted in the data management layer, the data management function is perfect, and the data saving safety is high, the database table can be flexibly designed, corresponding database tables are designed according to the functions provided by each module of the system, the meanings of each field in the database table and the constraint relationship are displayed through the table; the data analysis layer provides analysis, processing, identification, matching and judgment services for the time-sensitive data reported by the gateway and the non-time-sensitive data stored in the data management layer; the application service layer provides an operation interface for system users and provides on-site device state display and manual control functions for system users.

[0089] As shown in Figure 2 , the application also provides an implementation method of lithium battery energy storage station cloud-edge collaborative intelligent fire linkage, characterized in that, the method comprises the following steps:

[0090] (1) the perception layer acquires (sound wave, heat induction, temperature, gas) data and reports to the network layer through the modbusrtu protocol;

[0091] (2) the network layer uses edge computing technology to pre-process data, extracts a feature vector matrix, and reports to the fire service cloud platform through the gateway; as shown in Figure 3 , the step of using edge computing technology to pre-process data specifically comprises the following steps:

[0092] (21) a wavelet base function is acquired:

[0093] a wavelet transform L 2 (R) represents an energy-limited signal, and the Fourier transform thereof is Ψ(w); when the condition is met:

[0094]

[0095] called Ψ (ω) is the mother wavelet spectrum function, w represents frequency, t represents time, the mother wavelet A wavelet sequence can be obtained through proper linear transformation:

[0096]

[0097] In the above formula, a and b are the scaling and translation factors of the mother wavelet, respectively;

[0098] (22) Discretization of the above formula, the discrete wavelet function is:

[0099]

[0100] (23) For any input sound signal f(t) ∈ L 2 (R), the discrete wavelet transform is:

[0101]

[0102] (24) Four-layer decomposition of the discrete wavelet according to the approximation value and the noise value, as shown in the following formula:

[0103] w f (j, k) = C A4 + C D4 + C D3 + C D2 + C D1

[0104] In the formula, C Ai is the wavelet decomposition approximation part, C Di then represents the part where the noise is located; wavelet function layering refers to decomposing the part where the noise is located into four specific parts, namely the low-frequency part and the high-frequency parts in three directions (horizontal, vertical and diagonal), which is one of the steps of standard wavelet transform;

[0105] (25) Determine the threshold size according to the sample estimation; the selection and setting of the threshold value are the key in the wavelet denoising process, by removing the wavelet coefficients that do not meet the set threshold requirements and retaining the remaining coefficients, a good denoising effect is achieved; according to the background noise characteristics of the battery prefabricated cabin and considering the computing capacity of the edge gateway, a fixed maximum minimum criterion method is directly used to determine the threshold value, so as to reduce the calculation overhead and improve the calculation efficiency; the maximum minimum criterion threshold value method is as shown in the following formula:

[0106]

[0107] In the above formula, θ is the standard deviation of the noise, and n is the number of wavelet coefficients at each scale;

[0108] (26) According to threshold recombination wavelet coefficient; wavelet transform for a given signal is to expand the signal according to a certain wavelet function cluster, that is, the signal is expressed as a linear combination of a series of wavelet functions with different scales and different time shifts, and each coefficient is called a wavelet coefficient, which is: leave the wavelet coefficient greater than the set threshold, and the rest is zero processing, as shown in the following formula:

[0109]

[0110] Wherein, is the wavelet coefficient set after determination, x is the wavelet coefficient, and T is the threshold;

[0111] (27) Inverse transform of the wavelet energy spectrum after processing the wavelet coefficient, reconstruct the time signal

[0112]

[0113] In the above formula, a = 2 j , b = 2 j k, j, k are positive integers; j, k in steps (21) to (27) above have no actual meaning, which means sequence number parameter;

[0114] (28) Establishing a denoising evaluation index function, here the signal-to-noise ratio is introduced to evaluate the signal denoising performance evaluation index, for the denoising effect that fails to meet the requirements, the reconstructed signal is returned to step (24) again for decomposition, and after the second threshold determination, the reconstructed signal is reconstructed again, if the reconstructed signal still cannot meet the evaluation index requirements, the signal is discarded, the relationship between the reconstructed time signal in step (27) and the evaluation index function is shown in the following formula, which is simply speaking that the square difference is summed and averaged, and then logarithmic operation is performed, n is the number of wavelet coefficients at each scale, f(n) refers to the reconstructed time signal with n as the variable, and the evaluation index function is as follows:

[0115]

[0116] (29) The reconstructed signal after wavelet denoising and meeting the signal-to-noise ratio evaluation index needs to be subjected to principal component analysis (PCA) for eigenvalue extraction; the specific steps of principal component analysis are as follows:

[0117] (291) Assuming that the input data source has n samples and p indexes (because wavelet denoising and PCA eigenvalue extraction are two relatively independent parts, here n and p are only assumed parameters and do not correspond to the previous steps), an n×p sample matrix x can be constructed:

[0118]

[0119] (292) Calculate the average of the eigenvectors:

[0120]

[0121] (293) Obtain the covariance matrix S of the eigenvectors, that is:

[0122]

[0123] (294) Calculate the eigenvalues and eigenvectors of the normalized sample covariance:

[0124] Eigenvalues: λ1, λ2,..., λ n , eigenvectors: v1, v2,..., v n ;

[0125] (295) Calculate the contribution rate η i and cumulative contribution rate β k of each component:

[0126]

[0127]

[0128] (296) Construct the eigenvector mapping matrix:

[0129] Sort the principal components by contribution rate, select the largest p principal components that meet the requirements (cumulative contribution rate is higher than 85%), then form the eigenvector matrix V corresponding to the p principal components as the mapping matrix, and transform the denoised reconstructed signal to the feature space, that is:

[0130] Y = V T X

[0131] In the above formula: Y is the projected data matrix, that is, the principal component matrix of the original signal after PCA processing; V is the eigenvector matrix, V = [v1, v2,..., v p ], V T is the transpose matrix, and X is the average of the eigenvectors;

[0132] Based on wavelet denoising and PCA principal component analysis, the original data collected by the sensor is denoised and dimensionally reduced, and the data set of multiple variables is converted into a data set of fewer new variables. At the same time, the principal components are normalized and reported to the improved BP neural network model of the fire service cloud platform for training and dynamic identification and diagnosis.

[0133] (3) The fire service cloud holder is deployed based on the improved BP neural network model, the first 30 times of data are simulated and trained, and the characteristic vector recognition library is obtained; at the same time, the characteristic vector recognition library is constantly updated according to the subsequent uploaded data; since the four most important operating state parameters of the lithium battery energy storage station, i.e. sound wave, gas concentration, thermal induction and temperature, have a certain priori and mutual correlation in the probability distribution of the state space in the time dimension, they are not completely independent of each other, therefore, the pulse neural network SNN is introduced here, and the time information is calculated and transmitted by using a specific neuron model, so that the new neural network model can effectively process the time sequence of the four operating state parameters, and the gradient corresponding to the pulse neuron is calculated in the form of back propagation, and the gradient is updated, although this model adopts a two-time neural network training architecture, the hardware resource requirement of the cloud holder is very high, which reduces the training speed to a certain extent, but greatly improves the recognition accuracy, as shown in Figure 4 The specific principle of the improved BP neural network model is as follows:

[0134] The result of the forward propagation output of the neural network can be represented as:

[0135] bp_output=f(∑WX)

[0136] In the above formula, W represents the weight matrix, X represents the input characteristic value vector matrix (this X matrix corresponds to the Y matrix in the above), and f represents the activation function; the transmission information from the input layer to the hidden layer is obtained by the above formula, and an error function is used in the forward propagation process to verify whether the training process is ended, and the training is ended when the output error is less than the predetermined value, and the reverse transmission is performed if the ending condition is not met, and the error function calculation process is as follows:

[0137]

[0138] In the above formula, E represents the error function, d j represents the actual output, o j represents the expected output, j represents the neuron number in the network, d j , o j indicates the actual output and the expected output of the jth neuron, and d, o indicates the actual output and the expected output of all neurons, and n indicates the total number of neurons in the network, and the calculation error of the output neural network can be calculated by the above formula, and the error signal obtained by each layer is used to adjust the weight between neurons, and the error back propagation process can be represented as:

[0139]

[0140] In the above formula, v kj represents the connection weight, and z krepresents the implicit layer, k represents the number of hidden layers, n represents the total number of neurons, and the error back propagation is completed through the above formula; the forward propagation form of the SNN network is from the encoding layer to the output layer, and the state variable of the neuron is:

[0141]

[0142] In the above formula, O j represents the state variable of the SNN, represents the end weight in the SNN, represents the delay signal of the neuron in the SNN, m represents the total number of neurons, i represents the pulse response coefficient, k and j are positive integers (k and j are consistent with the foregoing), and t refers to time; all weights of the neural network model are adjusted through the above formula, the input signal is forward propagated, and therefore the output vector matrix of the SNN neural network output layer can be represented as:

[0143]

[0144] In the above formula, w i represents the output layer weight coefficient, η represents the weighting index, represents the state vector, b i represents the output layer pre-vector, Δw r represents the updated weight coefficient, r represents the number of neurons connected to the encoding layer, i is consistent with the foregoing, and x i refers to the i sample matrix with the pulse response coefficient; the output vector matrix of the improved BP neural network can be calculated through the above formula, and the above model can be directly used for training and parameter identification of the fire service cloud platform;

[0145] (4) The researchers manually input a historical year's fire data (containing temperature, gas and other parameters) of a lithium battery energy storage station as an abnormal database into the fire service cloud platform. The abnormal database serves as an important reference for predicting fires, and the database is updated continuously when subsequent abnormal data is received, thereby improving the accuracy of judgment.

[0146] (5) The improved BP neural network model is used for dynamic identification and diagnosis of the subsequently collected data, and the data is compared with the normal data in the feature vector identification library; when abnormal data appears, the abnormal database is compared, the abnormal data is analyzed, and it is diagnosed whether a fault occurs inside the energy storage station; if so, a fire event file is established separately;

[0147] (6) The result of diagnosing the fault is immediately sent to the edge side fire linkage control layer together with the warning information and the fire control command;

[0148] (7) After receiving the fire extinguishing instruction, the control layer of the fire linkage opens the alarm information, opens the fire extinguishing material safety valve, releases the fire extinguishing material (the main function of the fire extinguishing material is to isolate air, which can prevent fire), and cuts off all electrical circuits in the lithium battery energy storage warehouse; at the same time, the gateway sends the open fire information to the fire service cloud platform, informs the fire personnel and electrical maintenance personnel to arrive at the scene in time to do follow-up fire investigation and electrical maintenance work;

[0149] (8) The fire service cloud platform analyzes this fire, corrects the fire event file, and updates the abnormal database.

[0150] In view of the characteristics of current large-scale lithium battery energy storage battery structure, large operation safety risk, fast thermal runaway speed and the like, and in combination with the current monitoring of the lithium battery energy storage station remaining in a single device or terminal and the complete isolation of the fire fighting system, the application proposes a lithium battery energy storage station cloud edge collaborative intelligent fire fighting linkage system and an implementation method thereof, uses edge computing technology and cloud edge collaboration, carries out feature extraction, statistical analysis, dimension reduction optimization and the like preprocessing on a plurality of operation state parameters representing the lithium battery of the energy storage station at the edge gateway side, then trains and dynamically identifies based on an improved BP neural network model deployed in the intelligent fire fighting cloud platform, issues an early warning information according to the identification result, and combines the fire fighting linkage control system to form a complete cloud edge collaborative intelligent fire fighting linkage control method.

Claims

1. A method for implementing intelligent fire-fighting linkage of lithium battery energy storage station cloud edge collaboration, characterized in that, Specifically comprises the following steps: (1) the perception layer acquires data and reports to the network layer through the modbusrtu protocol; (2) the network layer uses edge computing technology to preprocess the data, extracts the feature vector matrix, and reports it to the fire service cloud platform through the gateway; (3) the improved BP neural network model deployed in the fire service cloud platform simulates and trains the first 30 times of data and obtains the feature vector identification library; at the same time, the feature vector identification library is constantly updated according to the subsequent uploaded data; (4) researchers manually input a historical year's lithium battery energy storage station fire data as an anomaly database into the fire service cloud platform. The anomaly database serves as an important reference for predicting fires, and the database is constantly updated when receiving subsequent abnormal data, thereby improving the accuracy of the judgment; (5) the improved BP neural network model is used to dynamically identify and diagnose the subsequent collected data by comparing it with the normal data in the feature vector identification library; when abnormal data appears, the anomaly database is compared to analyze and diagnose whether a fault has occurred inside the energy storage station; if so, a fire event file is established separately; (6) the result of diagnosing the fault immediately sends a warning message to the control layer of the edge side fire linkage, and sends a fire control command to the control layer; (7) after receiving the fire extinguishing command, the control layer of the fire linkage immediately starts the alarm information, opens the safety valve of the fire extinguishing material, releases the fire extinguishing material, and cuts off all electrical circuits in the lithium battery energy storage warehouse; at the same time, the gateway sends the fire information to the fire service cloud platform, notifies the firefighters and electrical maintenance personnel to arrive at the scene in time to do subsequent fire investigation and electrical maintenance work; (8) the fire service cloud platform analyzes the fire and corrects the fire event file, and updates the anomaly database; In step (2), the steps of data preprocessing using edge computing technology are as follows: (21) obtain the wavelet basis function: Setting up a wavelet transform L 2 (R) denotes a signal with finite energy whose Fourier transform is Ψ(w); when the condition is met: say The mother wavelet is defined as Ψ(ω), which is the mother wavelet's spectral function. ω represents frequency, and t represents time. A wavelet sequence can be obtained through appropriate linear transformation: In the formula, a and b are the stretching and translation factors of the mother wavelet; (22) discretize the above formula to obtain the discrete wavelet function: (23) For any input acoustic signal f(t) e L 2 The discrete wavelet transform of (R) is: (24) decompose the discrete wavelet into four layers according to the approximation value and the noise value, as shown in the following formula: w f (j, k) = C A4 +C D4 +C D3 +C D2 +C D1 where C Ai is the approximation part of the wavelet decomposition, C Di denotes the part where the noise is located; (25) determine the threshold size according to the sample estimation; (26) recombine the wavelet coefficients according to the threshold value; (27) inverse transform the wavelet energy spectrum after processing the wavelet coefficients, and reconstruct the time signal In the above formula, where a = 2 j b = 2 j k, j, k are positive integers; (28) establish a denoising evaluation index function, which introduces the signal-to-noise ratio to evaluate the signal denoising performance evaluation index. For the denoising effect that fails to meet the requirements, the reconstructed signal is returned to step (24) for decomposition again, and the second threshold determination is performed before reconstruction. If the twice reconstructed signal still cannot meet the evaluation index requirements, the signal is discarded. The evaluation index function is as follows: (29) the reconstructed signal after wavelet denoising and meeting the required signal-to-noise ratio evaluation index needs to be subjected to principal component analysis PCA for feature value extraction.

2. The implementation method of the lithium battery energy storage station cloud edge coordination intelligent fire fighting linkage according to claim 1, characterized in that, In step (25), the selection and setting of the threshold value is the key in the wavelet denoising process, by removing the wavelet coefficients which do not meet the set threshold value requirements, and retaining the remaining coefficients, so as to achieve good denoising effect; according to the background noise characteristics of the battery prefabricated cabin, and considering the computing power of the edge gateway, the fixed maximum minimum criterion method is used to determine the threshold value, so as to reduce the calculation overhead and improve the calculation efficiency; the maximum minimum criterion threshold value method is as follows: In the above formula, θ is the standard deviation of the noise, and n is the number of wavelet coefficients at each scale.

3. The implementation method of the intelligent fire-fighting linkage of the lithium battery energy storage station cloud edge collaboration according to claim 1, characterized in that, In step (26), wavelet transform of a given signal is to expand the signal according to a certain wavelet function cluster, that is, to express the signal as a linear combination of a series of wavelet functions with different scales and different time shifts, wherein the coefficient of each term is called wavelet coefficient, and the specific is: the wavelet coefficients greater than the set threshold value are left, and the rest is zero processing, as shown in the following formula: wherein is the set of wavelet coefficients after thresholding, x is a wavelet coefficient, and T is a threshold value.

4. The implementation method of the lithium battery energy storage station cloud edge coordination intelligent fire fighting linkage according to claim 1, characterized in that, in In step (29), the specific steps of principal component analysis method are as follows: (291) Assuming that the input data source has n samples and p indexes, an nxp sample matrix x can be constructed: (292) Calculate the average value of the eigenvector: (293) Calculate the covariance matrix S of the eigenvector, that is: (294) Calculate the eigenvalue and eigenvector of the normalized sample covariance: Eigenvalue :λ1,λ2,...,λ n , eigenvectors: v1, v2,..., v n ; (295)calculating the contribution rate η of each component i and the cumulative contribution rate β k : (296) Construct the eigenvector mapping matrix: Sort the principal components according to the contribution rate, select the largest p principal components that meet the requirements, and then form the eigenvector matrix V as the mapping matrix, and transform the denoising reconstructed signal to the feature space, that is: Y = V T X In the above formula: Y is the data matrix after projection, that is, the principal component matrix of the original signal after PCA processing; V is the eigenvector matrix, V = [v1, v2,..., vn], V p is the transpose matrix, and X is the average value of the eigenvector. T ​ 5. The implementation method of the intelligent fire-fighting linkage of the lithium battery energy storage station cloud edge collaboration according to claim 1, characterized in that, In step (3), the specific principle of the improved BP neural network model is as follows: The output result of the forward propagation of the neural network can be represented as: bp_output=f(∑WX) In the above formula, W represents the weight matrix, X represents the input eigenvector matrix, and f represents the activation function; the transmission information from the input layer to the hidden layer is obtained through the above formula, and an error function is used in the forward propagation process to verify whether the training process is ended, and the training is ended when the output error is less than the predetermined value, and the reverse transmission is performed if the end condition is not met, and the error function calculation process is as follows: where E represents the error function, d j represents the actual output, o j represents the desired output, j represents the neuron number in the network, d j , o j represents the actual output and the desired output of the jth neuron, and d, o represents the actual output and the desired output of all neurons, n represents the total number of neurons in the network. The calculation error of the output neural network can be calculated by the above formula. The error signal obtained by each layer is used to adjust the weight between neurons. The error back propagation process can be represented as: In the above formula, v kj represents the connection weight, z k represents the hidden layer, k represents the number of hidden layers, n represents the total number of neurons, and error back propagation is completed by the above formula; the forward propagation form of the SNN network is from the encoding layer to the output layer, and the state variable of the neuron is: In the above formula, O j denotes the state variable of the SNN, denotes the end weight in the SNN, denotes the delay signal of the neuron in the SNN, m denotes the total number of neurons, i denotes the pulse response coefficient, k and j are positive integers, and t refers to time; all the weights of the neural network model are adjusted by the above formula, the input signal is forward propagated, and therefore the output vector matrix of the output layer of the SNN neural network can be represented as: w i denotes the output layer weight coefficient, and η denotes the weighting index, denotes the state vector, and b i denotes the output layer pre-vector, and Δw r denotes the updated weight coefficient, r denotes the number of neurons connected to the encoding layer, and x i denotes the impulse response coefficient i sample matrix; the output vector matrix of the improved BP neural network can be calculated by the above formula, and the above model can be directly used for training and parameter identification of the fire service cloud platform.

Images