A deep learning-based energy storage power station thermal runaway early warning method and system

Abstract

The present invention provides a thermal runaway early warning method and system for energy storage power stations based on deep learning. The method comprises: inputting preprocessed battery cluster data into a trained prediction model to output a temperature forecast value for a period of time in the future; performing an abnormal state determination on the temperature forecast value based on a set abnormal state determination rule; the training process of the prediction model comprises: using continuous historical moment feature data, extracting features using a convolutional neural network, extracting time series features using an LSTM, and taking the output of the LSTM as the features of the historical data; selecting feature data at the predicted moment, extracting time series features using an LSTM, concatenating the feature data of each time step with the features of the historical data, and inputting the data into a layer of a fully connected neural network to predict the temperature value at the future moment. The advantages of the present invention are: it can determine the abnormal state of the battery at an early stage, has advantages in cost and complexity compared to existing technologies, and is easy to reuse and promote.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of energy storage power station control, and particularly relates to an energy storage power station thermal runaway early warning method and system based on deep learning. BACKGROUND

[0002] In recent years, renewable energy such as wind power and photovoltaic has developed rapidly, and the importance of energy storage technology in the field of energy storage and application is increasingly prominent. Lithium ion batteries are widely used in energy storage systems (ESS) due to their high energy density, low self-discharge rate, long cycle life and no memory effect. However, with the problems of long-time load operation of the system and battery aging, the safety risk of the battery is particularly important. In recent years, dozens of energy storage fire accidents have occurred worldwide, most of which are related to thermal runaway of energy storage systems.

[0003] The thermal runaway of the battery module mainly includes three stages: first, due to environmental or battery problems, internal short circuit of the battery, local temperature rise, high temperature causing a series of side reactions to further accelerate temperature rise. Then, the battery cells are closely spaced, heat is transmitted, and thermal runaway is transmitted. Finally, causing safety accidents such as fire and explosion. Therefore, it is urgent to develop a real-time and effective method to prevent the spread of thermal runaway and avoid safety accidents.

[0004] The existing safety strategy for thermal runaway generally sets different risk levels for the current battery temperature, generates an alarm signal, and then manually intervenes to solve the safety problem. This method lacks a certain flexibility and is affected by different environments, and the results may have a certain delay.

[0005] The invention patent with the application publication number CN115842183A: a thermal runaway early warning system and method for lithium ion battery energy storage cabin, uses multiple detection devices, including a pressure detection module, a sound detection module and a particle detection module, installed in the shell of the battery pack unit. The monitoring host is connected to the detection modules of all detection devices, and is used to receive and analyze the monitoring signals, and triggers an alarm when an abnormal signal is received.

[0006] This technical solution uses a series of detection modules and sensors to alarm thermal runaway. However, for large-scale energy storage systems, a large number of detection devices need to be installed and managed, increasing the cost and complexity of the system, and requiring higher technical personnel in the later maintenance.

[0007] The application patent with the application publication number CN115856675A: a battery thermal runaway early warning method and system, a storage medium, are used for quickly detecting and judging the abnormal state of a battery pack. The technical solution of the application comprises the following steps: first, detecting the temperature of a single battery; then, judging whether the temperature of the single battery meets the set temperature rise and voltage; if the condition is not met, detecting the temperature of adjacent devices; then, judging whether the temperature of the adjacent devices meets the set temperature rise and voltage; if the condition is met, it indicates that the single battery is abnormal, otherwise, it indicates that the battery pack is abnormal.

[0008] The technical solution only judges the state of the current battery and adjacent battery by setting a threshold value. Since the temperature changes under different batteries and environmental conditions are different, it is difficult to accurately set the threshold value, which may cause false alarms or delayed alarm situations, and it is difficult to reuse. SUMMARY

[0009] The purpose of the present application is to overcome the defects of the prior art that require the installation and management of a large number of detection devices or cannot be reused.

[0010] In order to achieve the above-mentioned purpose, the present application provides a thermal runaway early warning method for energy storage power stations based on deep learning, which comprises:

[0011] The preprocessed battery cluster data is input into the trained prediction model, and the temperature prediction value in the future period of time is output;

[0012] Based on the set abnormal state judgment rule, the temperature prediction value is judged for abnormal state;

[0013] The training process of the prediction model comprises:

[0014] Using continuous historical time feature data, the convolutional neural network is used to extract the features, the LSTM is used to extract the time sequence features, and the output of the LSTM is taken as the features of the historical data;

[0015] After selecting the feature data of the prediction time and extracting the time sequence features using the LSTM, the feature data of each time step is spliced with the features of the historical data, and input into a fully connected neural network to predict the temperature value at the future time;

[0016] The mean square error loss function is used for loss calculation and back propagation.

[0017] As an improvement of the above-mentioned method, the battery cluster data comprises: time, total power, total current, SOC, and battery or copper bar single temperature.

[0018] As an improvement of the above-mentioned method, the pre-processing process of the battery cluster data comprises:

[0019] The linear filling method is used to process the null value in the data.

[0020] The maximum and minimum normalization method is used to normalize the data.

[0021] The sliding window is used to convert the data into supervised learning sequence feature data.

[0022] As an improvement of the above method, the historical time feature data includes: total power, total current, SOC, ambient temperature of the battery cluster, and the predicted temperature of the single battery or copper bar.

[0023] As an improvement of the above method, the training process of the prediction model further includes: using the five-fold cross-validation method to cyclically train and comprehensively evaluate the data set.

[0024] As an improvement of the above method, the setting method of the abnormal state determination rule includes:

[0025] When the error of M-minute data and the current data exceeds the abnormal determination threshold during the prediction of temperature data after N minutes, the abnormal state is determined; wherein N>M; the threshold is the error statistics of the test results after five-fold cross-validation, specifically, the difference between the test set prediction results and the actual value is counted, and the maximum error under the condition that the actual value is greater than the prediction value is selected as the abnormal determination threshold.

[0026] The application also provides a deep learning-based energy storage power station thermal runaway early warning system, which is realized based on the above method, and the system includes:

[0027] The data preprocessing module is used for preprocessing the collected battery cluster data.

[0028] The calculation module is used for inputting the battery cluster data processed by the data preprocessing module into the trained prediction model, and outputting the temperature prediction value in the future period of time; and

[0029] The abnormal state determination module is used for determining the abnormal state of the temperature prediction value according to the set abnormal state determination rule.

[0030] As an improvement of the above system, the system further includes:

[0031] The training module is used for training the prediction model using historical data.

[0032] Compared with the prior art, the application has the following advantages:

[0033] The application is based on a CNN and an LSTM neural network, uses historical temperature data to predict the current temperature, and analyzes the difference between the predicted value and the actual value, so that the abnormal state of the battery can be determined in the early stage, and the early warning time of thermal runaway is preposed. Moreover, the application is completely based on the conventional container sensor data, and compared with the use of multiple sensors for determination, it has more advantages in cost and complexity, and is easy to reuse and promote. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 A technical roadmap of the energy storage power station thermal runaway early warning method based on deep learning is shown.

[0035] Figure 2 A combination of historical features and features at each time step prediction time is shown. DETAILED DESCRIPTION

[0036] The technical solutions of the application will be described in detail below with reference to the drawings.

[0037] The application proposes an energy storage power station thermal runaway early warning method based on deep learning based on the monitoring data of energy storage containers, so as to realize early and accurate temperature anomaly warning and prevent the spread of thermal runaway and promote the safe development of the energy storage industry.

[0038] The application uses various data of the battery management system, including total power, total current, SOC (State of Charge, charge state, an index used to describe the current charging level or remaining capacity of a rechargeable battery or energy storage system. It is usually expressed in percentage, indicating the proportion of the remaining energy in the battery relative to its maximum capacity), environmental temperature, and historical battery cell temperature and copper bar temperature, etc. features, based on the method of combining CNN and LSTM deep neural network to predict the current battery or copper bar temperature, and combining the actual data to analyze the difference, so as to determine the abnormal state of the battery in the early stage and avoid the occurrence of thermal runaway.

[0039] Figure 1 The technical roadmap of the application is a kind of energy storage power station thermal runaway early warning method based on deep learning, and the overall design idea of the method is: since the battery / copper bar temperature and the battery charging and discharging state have high correlation, and the charging and discharging state has randomness, the historical feature data and the predicted temperature feature data at the prediction time are combined to predict the battery / copper bar temperature at the future time, and then the model error analysis is carried out to set the threshold, and finally the error analysis is carried out with the actual value to determine the abnormality.

[0040] The energy storage power station thermal runaway early warning method based on deep learning comprises: temperature prediction is performed in a manner combining historical data and prediction time characteristics. Branch 1 uses continuous historical time characteristic data including (total power, total current, SOC, ambient temperature, and predicted single battery / copper bar temperature) to use a 1D convolutional neural network to extract features, and then uses LSTM to extract time sequence features, and the output of the last time slice of the LSTM is taken as the feature of the historical data. Branch 2 selects the features of the total power, total current, SOC, and ambient temperature of the prediction time, uses LSTM to extract the time sequence features, and then splices each time step with the historical features, and inputs them into a layer of fully connected neural network to predict the temperature value at the future time. Finally, the mean square error loss function is used for loss calculation and back propagation.

[0041] The specific process of the energy storage power station thermal runaway early warning method based on deep learning comprises:

[0042] Step 1: data preprocessing

[0043] Step 1-1: data extraction: for the required features of the model, more than half a year of battery cluster data is extracted, the sampling frequency is 1 min, and the data includes time, total power, total current, SOC, and battery / copper bar single temperature. Linear filling method is used to process the null value.

[0044] Step 1-2: data normalization: the maximum and minimum normalization method is used to normalize the data:

[0045]

[0046] Wherein, x is the input feature, x max is the maximum value of the current input feature, and x min is the minimum value of the current input feature.

[0047] Step 1-3: feature engineering: the sliding window data is converted into supervised learning sequence feature data.

[0048] Step 2: model construction

[0049] Step 2-1: data set division: the five-fold cross-validation method is used to cyclically train and comprehensively evaluate the data set.

[0050] Step 2-2: extraction of historical features: the total power, total current, SOC, ambient temperature, and predicted single battery / copper bar temperature of the battery cluster are used as historical data to input a 1D convolutional neural network to extract features to the network, the convolution kernel size is set to 3, the channel number is set to 128, the step is 1, and the activation function uses ReLU activation function, as shown in formula (2).

[0051]

[0052] Then the historical features are input into the LSTM recurrent neural network for time series feature extraction, and the output features at the last time are selected as the historical features. The mathematical expression of LSTM is as follows:

[0053] f t =σ(W f ·[h t-1 ,x t ]+b f ) (3)

[0054] i t =σ(W i ·[h t-1 ,x t ]+b i ) (4)

[0055]

[0056]

[0057] o t =σ(W o ·[h t-1 ,x t ]+b o ) (7)

[0058] h t =o t *tanh(C t ) (8)

[0059] Wherein, W is the input weight, b is the bias value, f represents the forgetting gate, represent the cell state at the current time step, i represents the input gate, and o represents the output gate. First, formula (3) shows that the hidden state h t-1 is input to the forgetting gate together with the input x t , and the output 0 represents complete forgetting, and 1 represents complete memory. Similarly, formulas (5-6) are for updating the cell state. First, a new cell state is generated, and then the information of the cell state at the last time is added to the new cell state, but the information at the last time here is controlled information through the forgetting gate. Formula (7) is the output gate, which is controlled by the current output and the output at the last time. Finally, formula (8) is used for nonlinear operation as the final output.

[0060] Step 2-3: Feature extraction at the predicted time: the total power, total current, SOC, and ambient temperature of the battery cluster at the predicted time are selected as feature data for feature engineering and input into another LSTM network for time series feature extraction, and the output of each time step is selected as the final output of the network.

[0061] Step 2-4: Combine the historical features with the predicted time features at each time step and input them into a fully connected neural network for synchronous output. The final output is the temperature prediction value for a future period of time. The process is shown in Figure 2

[0062] Loss function: the mean square error loss function is used for loss calculation, as shown in equation (9).

[0063]

[0064] where N represents the number of samples, y i represents the sample label, represents the predicted value.

[0065] Step 3: Abnormality determination

[0066] Step 3-1: Set the abnormality determination threshold

[0067] Statistical errors of the test results after five-fold cross-validation are calculated, specifically the difference between the predicted results and the actual values of the test set is calculated (the sample in this experiment is in the order of ten million), and the distribution is observed. Finally, the maximum error value under the condition that the actual value is greater than the predicted value is selected as the abnormality determination threshold.

[0068] Step 3-2: Set the abnormal state determination rule

[0069] Set the abnormal state determination rule: for example, if the error between 5 minutes of data and the current data exceeds the set abnormality determination threshold, the system determines that the current 10 minutes of temperature data is abnormal and sends an alarm.

[0070] The application also provides a deep learning-based energy storage power station thermal runaway early warning system, which is realized based on the above method. The system comprises:

[0071] A data preprocessing module is configured to preprocess the collected battery cluster data.

[0072] A calculation module is configured to input the battery cluster data processed by the data preprocessing module into a trained prediction model and output temperature prediction values for a future period of time.

[0073] An abnormal state determination module is configured to determine the abnormal state of the temperature prediction values according to the set abnormal state determination rule.

[0074] ​a training module configured to train the prediction model using historical data.

[0075] The application can also provide a computer device, comprising: at least one processor, a memory, at least one network interface and a user interface. The various components in the device are coupled together by a bus system. It can be understood that the bus system is used to realize the connection communication between the components. In addition to including a data bus, the bus system also includes a power bus, a control bus and a status signal bus.

[0076] The user interface can include a display, a keyboard or a pointing device, for example, a mouse, a trackball, a touchpad or a touch screen.

[0077] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM) or a flash memory. The volatile memory can be a Random Access Memory (RAM) used as an external cache. By way of example, but not limitation, many forms of RAM can be used, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM) and Direct Rambus RAM (DRRAM). The memory described herein is intended to include, but not be limited to, these and any other suitable types of memory.

[0078] In some embodiments, the memory stores elements, executable modules or data structures, or a subset thereof, or an extended set thereof: an operating system and an application program.

[0079] The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, and the like, for implementing various basic services and processing hardware-based tasks. The application programs include various application programs, such as a media player (Media Player), a browser (Browser), and the like, for implementing various application services. The program for implementing the method of the embodiments of the present disclosure can be included in the application programs.

[0080] In the above-described embodiments, the processor can be configured to, by invoking a program or an instruction stored in a memory, specifically, a program or an instruction stored in an application program:

[0081] perform the steps of the above-described method.

[0082] The above-described method can be applied to a processor or implemented by the processor. The processor can be an integrated circuit chip having a signal processing capability. In the implementation process, the steps of the above-described method can be completed by hardware integrated logic circuits in the processor or by software form instructions. The above-described processor can be a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The above-described methods, steps and logic block diagrams can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the above-described method can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, or other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and combines the hardware to complete the steps of the above-described method.

[0083] It is understood that the embodiments described in the present invention may be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit may be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or a combination thereof.

[0084] For software implementation, the technology of the present invention can be implemented by executing the functional modules (e.g., procedures, functions, etc.) of the present invention. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0085] The present invention may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, each step in the above method embodiment can be implemented.

[0086] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention and are intended to be encompassed by the claims of the present invention.

Claims

1. A deep learning-based thermal runaway early warning method for an energy storage power station, the method comprising: The pre-processed battery cluster data is input into the trained prediction model to output the temperature prediction value for a period of time in the future; Based on the set abnormal state judgment rules, the temperature prediction value is judged as abnormal state; The processing of the prediction model includes: Using continuous historical moment feature data, we use convolutional neural networks to extract features, use LSTM to extract time series features, and take the output of LSTM as the feature of historical data; Select the feature data at the prediction moment, use LSTM to extract the time series features, then splice the feature data of each time step with the features of the historical data, and input it into a fully connected neural network to predict the temperature value at the future moment; The method for setting abnormal state judgment rules includes: When predicting temperature data N minutes later, if the error between the data of M minutes and the current data exceeds the abnormality judgment threshold, it is judged to be an abnormal state; where N>M; the threshold is the error statistics of the test results after five-fold cross-validation, specifically the difference between the predicted results of the test set and the actual value, and the maximum error value under the condition that the actual value is greater than the predicted value is selected as the abnormality judgment threshold.

2. The thermal runaway early warning method for energy storage power stations based on deep learning according to claim 1 is characterized in that: Battery cluster data includes: time, total power, total current, SOC, and battery or copper busbar temperature.

3. The thermal runaway early warning method for energy storage power stations based on deep learning according to claim 1 is characterized in that: The preprocessing process of battery cluster data includes: The empty values ​​in the data are processed by linear filling; The maximum and minimum normalization method is used to normalize the data; Use sliding windows to convert data into sequence feature data for supervised learning.

4. The thermal runaway early warning method for energy storage power stations based on deep learning according to claim 1 is characterized in that: Historical characteristic data, including the total power, total current, SOC, ambient temperature of the battery cluster, and the predicted temperature of the single battery or copper busbar.

5. The thermal runaway early warning method for energy storage power station based on deep learning according to claim 1 is characterized in that: The training process of the prediction model includes: using a five-fold cross-validation method to perform cyclic training and comprehensive evaluation on the data set.

6. A deep learning-based thermal runaway early warning system for an energy storage power station, implemented based on any of the methods described in claims 1-5, characterized in that: The system comprises: Data preprocessing module, used to preprocess the collected battery cluster data; A calculation module, which is used to input the battery cluster data processed by the data preprocessing module into the trained prediction model and output the temperature prediction value for a period of time in the future; and The abnormal state judgment module is used to judge the abnormal state of the temperature prediction value according to the set abnormal state judgment rules.

7. The deep learning-based thermal runaway early warning system for energy storage power stations according to claim 6, characterized in that: The system further comprises: The training module is used to train the prediction model using historical data.

Images