High-capacity battery energy storage power station safety situation early warning method and device
Through multi-dimensional signal data and deep learning models, the key safety features of energy storage power stations are extracted, and the problem that existing systems are difficult to predict multiple types of safety accidents is solved, achieving more accurate early warning and more efficient safety management.
Patent Information
- Application Number
- CN202411101013.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-15
AI Technical Summary
The existing energy storage power station safety accident warning system is difficult to effectively predict multiple types of safety accidents, especially fire, explosion, chemical and electrical risks, resulting in insufficient warning time and affecting the safe operation of the power station.
By acquiring multi-dimensional signal data (electricity, temperature, sound, gas), using improved random forest method and long-term short-term memory network model, key safety feature signals are extracted, and early warning thresholds for different types of safety accidents are set to achieve super-supported Short-term forecasts and warnings for multiple types of safety accidents.
The power station's ability to predict multiple types of safety accidents has been greatly improved, and the early warning time has been advanced, which has improved the safe operation level of the power station and ensured more effective safety measures.
Smart Images

Figure CN118965074A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method and a device for early warning of safety situation of a large-capacity battery energy storage power station, belonging to the technical field of energy system safety situation perception. Background Art
[0002] As the penetration rate of renewable energy such as wind power and photovoltaic power continues to increase, the randomness and intermittency of renewable energy have an increasingly greater impact on the stability of the power grid. In order to balance the fluctuations in energy supply, a large number of energy storage power stations are increasingly becoming an effective way to solve this problem. However, the continuous outbreak of safety accidents in energy storage power stations in recent years is a key factor restricting its development. By early warning of safety accidents in energy storage power stations, more effective warning time can be provided for the power stations to take protective measures, which will promote the development of large-scale energy storage power stations.
[0003] The concept of situational awareness originated in the military field. It refers to the recognition and understanding of environmental factors under specific time and space conditions, and the prediction of future development trends. Traditional situational awareness methods are aimed at time series data prediction, and do not consider the prediction of different types of safety accidents and safety accidents with non-time series characteristics, resulting in the inability to effectively warn of safety accidents.
[0004] The invention patent with application number 202311568074.0 and patent name is "An intelligent safety early warning and protection method for energy storage power stations". The disclosed safety early warning and protection method includes the following steps: S1, the early warning and protection device is started to monitor multiple intrinsic parameters of the lithium battery in real time; S2, real-time detection of air particles, trace characteristic gases, and particulate matter generated in various thermal degradation stages of equipment, materials, cables and connectors in the battery compartment for early or very early warning of fire in the battery compartment; S3, real-time monitoring and early warning of the rupture pressure and release pressure of the battery pressure membrane during the lithium battery bulging stage; S4, detection, identification and alarm of AI images of gas, liquid and flames in the battery compartment; S5, the pump group is started to instantly vaporize the pressureless stored perfluorohexanone inhibitor to achieve the effect of absorbing heat and cooling; S6, the heat generated in the oxidation process of battery thermal runaway and explosion is transferred to the inert gas by combining "real-time nitrogen production + pre-storage of high-purity nitrogen". The solution in this patent has an early warning module, which is used to detect in real time the air particles, trace characteristic gases, and particulate matter generated by various thermal degradation stages of equipment, materials, cables, and connectors in the battery compartment for early or very early warning of fires in the battery compartment, and issue warning information based on changes in concentration. This type of solution is mainly aimed at early fire warnings in energy storage power stations, and the type of warning accidents is single, but actual energy storage power stations include four types of safety accidents, such as fire, explosion, chemical risks, and electrical risks. It can be seen from this that there are currently many types of safety accidents in large-capacity energy storage power stations, and many types of signals are collected. It is difficult for general neural networks to provide efficient warnings for multiple types of safety accidents. Summary of the invention
[0005] In order to solve the above problems, the present invention proposes a method and device for early warning of safety situation of a large-capacity battery energy storage power station, which can greatly improve the prediction ability of the power station energy management system for various types of safety accidents.
[0006] The technical solution adopted by the present invention to solve the technical problem is:
[0007] In a first aspect, an embodiment of the present invention provides a method for early warning of safety situation of a large-capacity battery energy storage power station, comprising the following steps:
[0008] Step S1, obtaining multidimensional signal data of a large-capacity battery energy storage power station, and changing the time series of the multidimensional signal from a numerical value to a time series of a state according to the abnormal or normal state of the multidimensional signal, and extracting the state and type of safety accidents in the safety log of the large-capacity battery energy storage power station to construct a data set, wherein the multidimensional signal data includes multidimensional signals of electricity, temperature, sound and gas;
[0009] Step S2, finding multi-dimensional signal data corresponding to different types of security incidents in the security log through an improved random forest method, and using them as key security feature signals;
[0010] Step S3, establish a long short-term memory network key feature prediction model, set the safety accident warning threshold by performing ultra-short-term prediction on the key safety feature signal, and trigger a safety warning of the corresponding accident type when the predicted value of the safety feature exceeds the threshold.
[0011] As a possible implementation of this embodiment, step S1 includes:
[0012] Step S11, collecting various safety accident related data of large-capacity battery energy storage power stations, and performing data cleaning, missing value processing and outlier detection processing on various safety accident related data, wherein the various safety accident related data include multi-dimensional signals of electricity, temperature, sound and gas during the accident period and safety accident logs;
[0013] Step S12, collecting M-dimensional signal data of the large-capacity battery energy storage power station, wherein the M-dimensional signal data at least includes multi-dimensional signals of electricity, temperature, sound and gas, and setting a reasonable normal operating range of each dimensional signal data to determine whether the M-dimensional signal data is in an abnormal state, and using this as sample data;
[0014] Step S13, the multi-dimensional signal state is set to:
[0015]
[0016] Where: D kn is the rated value of the kth type of data, Dkn is the actual value of the k-th data, R k is the normal operating interval of the kth type of data. When S k (D kn ,D kt ) is 1, indicating that the data is in an abnormal state;
[0017] Similarly, according to the safety accident log of the large-capacity battery energy storage power station, its safety status is divided into accident and normal states, where the safety state can be represented as 0 and the accident state as 1, thereby constructing a data set containing M-dimensional signal state and safety accident state.
[0018] As a possible implementation of this embodiment, step S2 includes:
[0019] Step S21, taking various types of safety accident data in the data set as category labels and multi-dimensional signal state data as features;
[0020] Step S22, independently extracting m samples from the original samples of the data set, and extracting a certain number of new samples with replacement based on the original samples to form a multi-dimensional decision tree, wherein sample features are randomly selected to decompose corresponding nodes of the decision tree;
[0021] In step S23, the abnormal probabilities of various types of safety accidents and signals of different dimensions in the random sample data are calculated respectively, and feature selection is performed on different types of safety accidents based on randomly selected sample features, and used as key safety feature signals.
[0022] As a possible implementation of this embodiment, the feature selection for different types of safety accidents based on randomly selected sample features includes:
[0023] Calculate the Gini impurity of a randomly selected sample feature:
[0024]
[0025] Where: For a c-type safety accident, randomly select the j-th m-type feature of the Gini impurity, is the probability of randomly selecting the jth m-type feature anomaly for a c-type safety accident; N is the safety accident state of the energy storage power station and there are only two states, 1 is the normal state and 2 is the accident state;
[0026] Calculate the Gini impurity without the features of the randomly selected sample:
[0027]
[0028] Where: For a c-type safety accident, randomly select the Gini impurity without the j-th m-type feature, The probability of randomly selecting anomalies without the jth m-type feature for a c-type safety accident;
[0029] The expression for the reduction in Gini impurity for randomly selected sample features is calculated as:
[0030]
[0031] Where: The reduction in Gini impurity of randomly selecting the jth m-class feature for a c-class safety accident;
[0032] Calculate the maximum reduction in Gini impurity for a randomly selected sample feature:
[0033]
[0034] Where: ΔG c For a c-type safety accident, select the maximum Gini impurity reduction from a total of H m-type features;
[0035] Safe features are selected based on the maximum reduction in Gini impurity of randomly selected sample features.
[0036] As a possible implementation of this embodiment, step S3 includes:
[0037] Step S31, normalizing the time series data with security features:
[0038]
[0039] in: represents the normalized variable value at time l, x l represents the variable value at time l, x max and x min Respectively represent the maximum and minimum values in the time series of each variable;
[0040] Step S32, using the sigmoid activation function and the Adam optimizer to construct a long short-term memory network key feature prediction model, wherein the first layer is the input layer, the second layer is the long short-term memory network layer, the third layer is the Dropout layer, the fourth layer is the fully connected layer, and the last layer is the output layer;
[0041] Step S33: input the multidimensional time series after data preprocessing into the long short-term memory network key feature prediction model to perform ultra-short-term security feature prediction to obtain the prediction result sequence
[0042] Step S34, setting a safety accident warning threshold, when the predicted value of the safety feature exceeds the safety threshold, triggering a safety warning of the corresponding accident type.
[0043] As a possible implementation of this embodiment, step S33 includes:
[0044] Step S331, for time series data, select the input time step as j and the output time step as k, that is, use j×m historical data to predict the future k data and construct a data set: Where j is generated through demand autocorrelation analysis, k is determined according to actual demand, and m is the input data dimension, which is generated through multivariate correlation analysis; for time t, O x ={X t-j ,X t-j+1 ,…,X t}, …, S x ={X t-j ,X t-j+1 ,…,X t}, T y ={X t+1 ,X t+2 ,…,X t+k}, where O x ,P x ,…,S x is the input time series used for prediction, T y is the real value time series of the output;
[0045] Step S332: Build multiple security feature prediction models, predict each type of security feature, and obtain a prediction result sequence.
[0046] As a possible implementation of this embodiment, in step S34, the judgment formula for the safety accident warning is:
[0047]
[0048] Where: is the predicted state of the corresponding feature of the c-th type of safety accident, is the characteristic prediction value corresponding to the c-th type of safety accident, is the characteristic safety threshold corresponding to the c-type safety accident; when It is predicted that the energy storage power station will have a type C safety accident, otherwise the energy storage power station is in a safe state.
[0049] In a second aspect, an embodiment of the present invention provides a device for early warning of safety situation of a large-capacity battery energy storage power station, comprising:
[0050] The signal status and safety accident status extraction module is used to obtain the multi-dimensional signal data of the large-capacity battery energy storage power station, and convert the time series of the multi-dimensional signal from the numerical value to the time series of the state according to the abnormal or normal state of the multi-dimensional signal, and at the same time extract the safety accident state and type in the safety log of the large-capacity battery energy storage power station to construct a data set. The multi-dimensional signal data includes electrical, temperature, acoustic and gas multi-dimensional signals;
[0051] The key safety feature signal screening module is used to find the multi-dimensional signal data corresponding to different types of safety accidents in the safety log through the improved random forest method and use it as the key safety feature signal;
[0052] The safety accident warning module is used to establish a long short-term memory network key feature prediction model. By performing ultra-short-term prediction on key safety feature signals, the safety accident warning threshold is set. When the predicted value of the safety feature exceeds the threshold, a safety warning of the corresponding accident type is triggered.
[0053] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of any of the above-mentioned methods for safety situation warning of large-capacity battery energy storage power stations.
[0054] In a fourth aspect, an embodiment of the present invention provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods for early warning of safety situation of a large-capacity battery energy storage power station are executed.
[0055] The beneficial effects of the technical solution of the embodiment of the present invention are as follows:
[0056] A method for early warning of safety situation of a large-capacity battery energy storage power station according to the technical solution of an embodiment of the present invention finds out key features corresponding to safety accidents through autocorrelation analysis and multivariate correlation analysis of safety features, and uses a long short-term memory network model to predict future ultra-short-term safety features; since different types of safety accidents correspond to different safety features, the present invention constructs a safety accident early warning system and sets different safety thresholds respectively. When the predicted value of the safety feature exceeds the threshold, the corresponding safety accident early warning is implemented. The present invention greatly improves the prediction ability of the power station energy management system for multiple types of safety accidents by establishing a safety situation early warning system based on feature selection.
[0057] The present invention selects key characteristic signals that can reflect the safety status of the energy storage power station from the sampled multi-dimensional signals, and matches the characteristics with the types of safety accidents. Since the key characteristic signals are time series data, the problem that it is difficult to use traditional methods to predict safety accidents when they are binary discrete events is solved, and a new method and idea is provided for safety accident prediction.
[0058] The present invention constructs a safety feature prediction model, performs ultra-short-term prediction of key safety feature signals, sets a safety feature warning threshold, and triggers a safety warning of the corresponding accident type when the predicted value of the safety feature exceeds the threshold. In this way, the predicted value of the time series can be converted into warnings of different types of safety accidents, reserving more time for the energy storage power station to take preventive measures in advance, thereby improving the safe operation level of the power station.
[0059] The device for early warning of the safety situation of a large-capacity battery energy storage power station according to the technical solution of an embodiment of the present invention has the same beneficial effects as the method for early warning of the safety situation of a large-capacity battery energy storage power station according to the technical solution of an embodiment of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 It is a flow chart of a method for early warning of safety situation of a large-capacity battery energy storage power station according to an exemplary embodiment;
[0061] Figure 2 It is a structural schematic diagram of a device for early warning of safety situation of a large-capacity battery energy storage power station according to an exemplary embodiment;
[0062] Figure 3 It is a schematic diagram of a method for classifying safety incidents and key safety features according to an exemplary embodiment. DETAILED DESCRIPTION
[0063] In order to more clearly illustrate the technical features of the solution of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.
[0064] like Figure 1 As shown, a method for early warning of safety situation of a large-capacity battery energy storage power station provided by an embodiment of the present invention includes the following steps:
[0065] Step S1, obtaining multidimensional signal data of a large-capacity battery energy storage power station, and changing the time series of the multidimensional signal from a numerical value to a time series of a state according to the abnormal or normal state of the multidimensional signal, and extracting the state and type of safety accidents in the safety log of the large-capacity battery energy storage power station to construct a data set, wherein the multidimensional signal data includes multidimensional signals of electricity, temperature, sound and gas;
[0066] Step S2, finding multi-dimensional signal data corresponding to different types of security incidents in the security log through an improved random forest method, and using them as key security feature signals;
[0067] Step S3, establish a long short-term memory network key feature prediction model, set the safety accident warning threshold by performing ultra-short-term prediction on the key safety feature signal, and trigger a safety warning of the corresponding accident type when the predicted value of the safety feature exceeds the threshold.
[0068] As a possible implementation of this embodiment, step S1 includes:
[0069] Step S11, collecting various safety accident related data of large-capacity battery energy storage power stations, and performing data cleaning, missing value processing and outlier detection processing on various safety accident related data, wherein the various safety accident related data include multi-dimensional signals of electricity, temperature, sound and gas during the accident period and safety accident logs;
[0070] Step S12, collecting M-dimensional signal data of the large-capacity battery energy storage power station, wherein the M-dimensional signal data includes at least electrical, temperature, acoustic and gas multi-dimensional signals, and setting a reasonable normal operating range for each dimensional signal data to determine whether the M-dimensional signal data is abnormal (i.e., whether it exceeds the normal operating range), and using this as sample data;
[0071] Step S13, the multi-dimensional signal state is set to:
[0072]
[0073] Where: D kn is the rated value of the kth type of data, D kn is the actual value of the k-th data, R k is the normal operating interval of the kth type of data. When S k (D kn ,D kt ) is 1, indicating that the data is in an abnormal state;
[0074] Similarly, according to the safety accident log of the large-capacity battery energy storage power station, its safety status is divided into accident and normal states, where the safety state can be represented as 0 and the accident state as 1, thereby constructing a data set containing M-dimensional signal state and safety accident state.
[0075] As a possible implementation of this embodiment, step S2 includes:
[0076] Step S21, taking various types of safety accident data in the data set as category labels and multi-dimensional signal state data as features;
[0077] Step S22, independently extracting m samples from the original samples of the data set, and extracting a certain number of new samples with replacement based on the original samples to form a multi-dimensional decision tree, wherein sample features are randomly selected to decompose corresponding nodes of the decision tree;
[0078] In step S23, the abnormal probabilities of various types of safety accidents and signals of different dimensions in the random sample data are calculated respectively, and feature selection is performed on different types of safety accidents based on randomly selected sample features, and used as key safety feature signals.
[0079] In order to select features for different types of safety incidents and obtain the most appropriate feature set, it is necessary to calculate the probability of misclassification. As a possible implementation of this embodiment, the feature selection for different types of safety incidents based on randomly selected sample features includes:
[0080] Calculate the Gini impurity of a randomly selected sample feature:
[0081]
[0082] Where: For a c-type safety accident, randomly select the j-th m-type feature of the Gini impurity, is the probability of randomly selecting the jth m-type feature anomaly for a c-type safety accident; N is the safety accident state of the energy storage power station and there are only two states, 1 is the normal state and 2 is the accident state;
[0083] Calculate the Gini impurity without the features of the randomly selected sample:
[0084]
[0085] Where: For a c-type safety accident, randomly select the Gini impurity without the j-th m-type feature, The probability of randomly selecting anomalies without the jth m-type feature for a c-type safety accident;
[0086] The expression for the reduction in Gini impurity for randomly selected sample features is calculated as:
[0087]
[0088] Where: The reduction in Gini impurity of randomly selecting the jth m-class feature for a c-class safety accident;
[0089] Calculate the maximum reduction in Gini impurity for a randomly selected sample feature:
[0090]
[0091] Where: ΔG cFor a c-type safety accident, select the maximum Gini impurity reduction from a total of H m-type features;
[0092] Safe features are selected based on the maximum reduction in Gini impurity of randomly selected sample features.
[0093] As a possible implementation of this embodiment, step S3 includes:
[0094] Step S31, normalizing the time series data with security features:
[0095]
[0096] in: represents the normalized variable value at time l, x l represents the variable value at time l, x max and x min Respectively represent the maximum and minimum values in the time series of each variable;
[0097] Step S32, using the sigmoid activation function and the Adam optimizer to construct a long short-term memory network key feature prediction model, wherein the first layer is the input layer, the second layer is the long short-term memory network layer, the third layer is the Dropout layer, the fourth layer is the fully connected layer, and the last layer is the output layer;
[0098] Step S33: input the multidimensional time series after data preprocessing into the long short-term memory network key feature prediction model to perform ultra-short-term security feature prediction to obtain the prediction result sequence
[0099] Step S34, setting a safety accident warning threshold, when the predicted value of the safety feature exceeds the safety threshold, triggering a safety warning of the corresponding accident type.
[0100] As a possible implementation of this embodiment, step S33 includes:
[0101] Step S331, for time series data, select the input time step as j and the output time step as k, that is, use j×m historical data to predict the future k data and construct a data set: Where j is generated through demand autocorrelation analysis, k is determined according to actual demand, and m is the input data dimension, which is generated through multivariate correlation analysis; for time t, O x ={X t-j ,X t-j+1 ,…,X t}, …, S x ={X t-j ,X t-j+1 ,…,X t}, Ty ={X t+1 ,X t+2 ,…,X t+k}, where O x ,P x ,…,S x is the input time series used for prediction, T y is the real value time series of the output;
[0102] Step S332: Build multiple security feature prediction models, predict each type of security feature, and obtain a prediction result sequence.
[0103] As a possible implementation of this embodiment, in step S34, the judgment formula for the safety accident warning is:
[0104]
[0105] Where: is the predicted state of the corresponding feature of the c-th type of safety accident, is the characteristic prediction value corresponding to the c-th type of safety accident, is the characteristic safety threshold corresponding to the c-type safety accident; when It is predicted that the energy storage power station will have a type C safety accident, otherwise the energy storage power station is in a safe state.
[0106] like Figure 2 As shown, an embodiment of the present invention provides a device for early warning of safety situation of a large-capacity battery energy storage power station, comprising:
[0107] The signal status and safety accident status extraction module is used to obtain the multi-dimensional signal data of the large-capacity battery energy storage power station, and convert the time series of the multi-dimensional signal from the numerical value to the time series of the state according to the abnormal or normal state of the multi-dimensional signal, and at the same time extract the safety accident state and type in the safety log of the large-capacity battery energy storage power station to construct a data set. The multi-dimensional signal data includes electrical, temperature, acoustic and gas multi-dimensional signals;
[0108] The key safety feature signal screening module is used to find the multi-dimensional signal data corresponding to different types of safety accidents in the safety log through the improved random forest method and use it as the key safety feature signal;
[0109] The safety accident warning module is used to establish a long short-term memory network key feature prediction model. By performing ultra-short-term prediction on key safety feature signals, the safety accident warning threshold is set. When the predicted value of the safety feature exceeds the threshold, a safety warning of the corresponding accident type is triggered.
[0110] like Figure 3As shown in the figure, the operating states of various types of signals collected from a certain type of safety accident are taken as data sets. By randomly classifying the data sets, different classification data sets can be obtained. Here, it is assumed that there are H sub-data sets in total. The Gini impurity reduction of each classification is calculated, and the maximum Gini impurity reduction is obtained by comparison. The corresponding classification is the determined classification, and the features in the classification are the key safety features, thereby screening out the key safety features corresponding to different types of safety accidents.
[0111] The present invention is mainly divided into two parts for the early warning of the safety situation of large-capacity battery energy storage power stations. The first part: through the analysis of the autocorrelation of safety features, the multivariate correlation analysis, and the use of the long short-term memory network model to predict the future ultra-short-term safety features; the second part: to build a safety accident early warning system. Since different types of safety accidents correspond to different safety features, different safety thresholds are set respectively. When the predicted value of the safety feature exceeds the threshold, the corresponding safety accident early warning is realized.
[0112] Taking a certain energy storage power station as an example, the specific process of the present invention for performing safety situation warning of a large-scale energy storage power station is as follows.
[0113] Step 1: Extract signal status and safety accident status.
[0114] 1) Data collection. Mainly collect multi-dimensional signals such as battery voltage, current, stress, H2 volume fraction, VOC and CO volume fraction at the battery stack level and energy storage box level, temperature, and safety logs of the energy storage power station EMS.
[0115] 2) Sample extraction. There are four types of safety accidents in energy storage power stations: fire, explosion, chemical risk, and electrical risk. The safety accident information of the power station in the safety log is marked as the above four categories, and the state of the multidimensional signal in step 1 is divided into abnormal state and normal state.
[0116] Step 2: Screening of key safety feature signals.
[0117] 1) Probability calculation: Collect and count the multi-dimensional signal states within one hour before all safety accidents occur, and calculate the probability of abnormal states of each dimensional signal corresponding to different types of accidents.
[0118] 2) Construct decision trees. Based on randomly selected original samples (bootstrap samples), multiple decision trees are constructed for four types of safety accidents, including fire, explosion, chemical risk, and electrical risk in energy storage power plants.
[0119] 3) Randomly extract variables. Here, m=1 features are randomly selected without repetition for the sake of simplicity.
[0120] 4) Calculate the Gini impurity of each feature (corresponding to each dimension of the sampled data).
[0121] 5) Calculate the Gini impurity without this feature.
[0122] 6) Calculate the Gini impurity reduction for each feature separately.
[0123] 7) Critical safety features are screened by selecting the maximum Gini impurity reduction. In this example, the most critical feature of fire accidents is temperature; the most critical feature of explosion accidents is H2 volume fraction; the most critical feature of chemical risks is VOC volume fraction; and the most critical feature of electrical risks is charge and discharge current.
[0124] The present invention selects key characteristic signals that can reflect the safety status of the energy storage power station from the sampled multi-dimensional signals, and matches the characteristics with the types of safety accidents. Since the key characteristic signals are time series data, the problem that it is difficult to use traditional methods to predict safety accidents when they are binary discrete events is solved, and a new method and idea is provided for safety accident prediction.
[0125] Step 3: Safety accident warning.
[0126] 1) Security feature preprocessing: The above data is a time series, the empty values in the data are filled with 0, and the time series data is normalized. The formula is:
[0127]
[0128] in: represents the normalized variable value at time l, x l represents the variable value at time l, x max and x min Respectively represent the maximum and minimum values in the time series of each variable.
[0129] 2) Establish a long short-term memory network key feature prediction model. The first layer of the model is the Input layer, the input shape is a 24×12 matrix; the second layer is the long short-term memory network layer, the output dimension is set to 8; the third layer is the Dropout layer, the ratio of neurons to be disconnected is set to 0.4, the fourth layer is the Dense fully connected layer, and the last layer is the Output prediction result output layer, using the sigmoid activation function The model uses the Adam optimizer. The final output shape is a 1×4 matrix.
[0130] 3) By constructing prediction models for four key characteristics, namely temperature, H2 volume fraction, VOC volume fraction and charge and discharge current, ultra-short-term prediction values of key characteristics can be achieved.
[0131] 4) Safety warning. Set the thresholds of temperature, H2 volume fraction, VOC volume fraction and charge and discharge current respectively. When the predicted value exceeds the threshold, four types of safety accidents such as fire, explosion, chemical risk and electrical risk will be warned in advance.
[0132] The present invention constructs a safety feature prediction model, performs ultra-short-term prediction of key safety feature signals, sets a safety feature warning threshold, and triggers a safety warning of the corresponding accident type when the predicted value of the safety feature exceeds the threshold. In this way, the predicted value of the time series can be converted into warnings of different types of safety accidents, reserving more time for the energy storage power station to take preventive measures in advance, thereby improving the safe operation level of the power station.
[0133] The present invention constructs a classification method for safety accidents and key safety features of large-scale energy storage power stations, and finds out the key safety features in various signals such as electricity, temperature, sound and gas corresponding to different types of safety accidents through the maximum Gini impurity reduction; the present invention provides a safety situation warning method, which predicts key feature signals and sets a safety accident warning threshold. When the predicted value of the feature signal exceeds the threshold, a safety warning of the corresponding fault type is triggered.
[0134] A computer device provided in an embodiment of the present invention includes a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the device is running, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of any of the above-mentioned methods for safety situation warning of large-capacity battery energy storage power stations.
[0135] Specifically, the above-mentioned memory and processor can be general-purpose memory and processor, which are not specifically limited here. When the processor runs the computer program stored in the memory, the above-mentioned method for safety situation warning of a large-capacity battery energy storage power station can be executed.
[0136] Corresponding to the method for starting the above-mentioned application, an embodiment of the present invention further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for providing a safety situation warning for any large-capacity battery energy storage power station are executed.
[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for early warning of safety situation of a large-capacity battery energy storage power station, characterized in that: The steps include: Step S1, obtaining multidimensional signal data of a large-capacity battery energy storage power station, and changing the time series of the multidimensional signal from a numerical value to a time series of a state according to the abnormal or normal state of the multidimensional signal, and extracting the state and type of safety accidents in the safety log of the large-capacity battery energy storage power station to construct a data set, wherein the multidimensional signal data includes multidimensional signals of electricity, temperature, sound and gas; Step S2, finding multi-dimensional signal data corresponding to different types of security incidents in the security log through an improved random forest method, and using them as key security feature signals; Step S3, establish a long short-term memory network key feature prediction model, set the safety accident warning threshold by performing ultra-short-term prediction on the key safety feature signal, and trigger a safety warning of the corresponding accident type when the predicted value of the safety feature exceeds the threshold.
2. The method for early warning of safety situation of a large-capacity battery energy storage power station according to claim 1 is characterized in that: The step S1 comprises: Step S11, collecting various safety accident related data of large-capacity battery energy storage power stations, and performing data cleaning, missing value processing and outlier detection processing on various safety accident related data, wherein the various safety accident related data include multi-dimensional signals of electricity, temperature, sound and gas during the accident period and safety accident logs; Step S12, collecting M-dimensional signal data of the large-capacity battery energy storage power station, wherein the M-dimensional signal data at least includes multi-dimensional signals of electricity, temperature, sound and gas, and setting a reasonable normal operating range of each dimensional signal data to determine whether the M-dimensional signal data is in an abnormal state, and using this as sample data; Step S13, the multi-dimensional signal state is set to: Where: D kn is the rated value of the kth type of data, D kn is the actual value of the k-th data, R k is the normal operating interval of the kth type of data. When S k (D kn ,D kt ) is 1, indicating that the data is in an abnormal state; Similarly, according to the safety accident log of the large-capacity battery energy storage power station, its safety status is divided into accident and normal states, where the safety state can be represented as 0 and the accident state as 1, thereby constructing a data set containing M-dimensional signal state and safety accident state.
3. The method for early warning of safety situation of a large-capacity battery energy storage power station according to claim 1 is characterized in that: The step S2 comprises: Step S21, taking various types of safety accident data in the data set as category labels and multi-dimensional signal state data as features; Step S22, independently extracting m samples from the original samples of the data set, and extracting a certain number of new samples with replacement based on the original samples to form a multi-dimensional decision tree, wherein sample features are randomly selected to decompose corresponding nodes of the decision tree; In step S23, the abnormal probabilities of various types of safety accidents and signals of different dimensions in the random sample data are calculated respectively, and feature selection is performed on different types of safety accidents based on randomly selected sample features, and used as key safety feature signals.
4. The method for early warning of safety situation of a large-capacity battery energy storage power station according to claim 3 is characterized in that: The feature selection of different types of safety accidents based on randomly selected sample features includes: Calculate the Gini impurity of a randomly selected sample feature: Where: For a c-type safety accident, randomly select the j-th m-type feature of the Gini impurity, is the probability of randomly selecting the jth m-type feature anomaly for a c-type safety accident; N is the safety accident state of the energy storage power station and there are only two states, 1 is the normal state and 2 is the accident state; Calculate the Gini impurity without the features of the randomly selected sample: Where: For a c-type safety accident, randomly select the Gini impurity without the j-th m-type feature, The probability of randomly selecting anomalies without the jth m-type feature for a c-type safety accident; The expression for the reduction in Gini impurity for randomly selected sample features is calculated as: Where: The reduction in Gini impurity of randomly selecting the jth m-class feature for a c-class safety accident; Calculate the maximum reduction in Gini impurity for a randomly selected sample feature: Where: ΔG c For a c-type safety accident, select the maximum Gini impurity reduction from a total of H m-type features; Safe features are selected based on the maximum reduction in Gini impurity of randomly selected sample features.
5. The method for early warning of safety situation of a large-capacity battery energy storage power station according to any one of claims 1 to 4, characterized in that: The step S3 comprises: Step S31, normalizing the time series data with security features: in: represents the normalized variable value at time l, x l represents the variable value at time l, x max and x min Respectively represent the maximum and minimum values in the time series of each variable; Step S32, using the sigmoid activation function and the Adam optimizer to construct a long short-term memory network key feature prediction model, wherein the first layer is the input layer, the second layer is the long short-term memory network layer, the third layer is the Dropout layer, the fourth layer is the fully connected layer, and the last layer is the output layer; Step S33: input the multidimensional time series after data preprocessing into the long short-term memory network key feature prediction model to perform ultra-short-term security feature prediction to obtain the prediction result sequence Step S34, setting a safety accident warning threshold, when the predicted value of the safety feature exceeds the safety threshold, triggering a safety warning of the corresponding accident type.
6. The method for early warning of safety situation of a large-capacity battery energy storage power station according to claim 5 is characterized in that: The step S33 comprises: Step S331, for time series data, select the input time step as j and the output time step as k, that is, use j×m historical data to predict the future k data and construct a data set: Where j is generated through demand autocorrelation analysis, k is determined according to actual demand, and m is the input data dimension, which is generated through multivariate correlation analysis; for time t, O x ={X t-j ,X t-j+1 ,…,X t }, …, S x ={X t-j ,X t-j+1 ,…,X t }, T y ={X t+1 ,X t+2 ,…,X t+k }, where O x ,P x ,…,S x is the input time series used for prediction, T y is the real value time series of the output; Step S332: Build multiple security feature prediction models, predict each type of security feature, and obtain a prediction result sequence.
7. The method for early warning of safety situation of a large-capacity battery energy storage power station according to claim 6 is characterized in that: In step S34, the judgment formula for safety accident warning is: Where: is the predicted state of the corresponding feature of the c-th type of safety accident, is the characteristic prediction value corresponding to the c-type safety accident, is the characteristic safety threshold corresponding to the c-th type of safety accident; when It is predicted that the energy storage power station will have a type C safety accident, otherwise the energy storage power station is in a safe state.
8. A device for early warning of safety situation of a large-capacity battery energy storage power station, characterized in that: include: The signal status and safety accident status extraction module is used to obtain the multi-dimensional signal data of the large-capacity battery energy storage power station, and convert the time series of the multi-dimensional signal from the numerical value to the time series of the state according to the abnormal or normal state of the multi-dimensional signal, and at the same time extract the safety accident state and type in the safety log of the large-capacity battery energy storage power station to construct a data set. The multi-dimensional signal data includes electrical, temperature, acoustic and gas multi-dimensional signals; The key safety feature signal screening module is used to find the multi-dimensional signal data corresponding to different types of safety accidents in the safety log through the improved random forest method and use it as the key safety feature signal; The safety accident warning module is used to establish a long short-term memory network key feature prediction model. By performing ultra-short-term prediction on key safety feature signals, the safety accident warning threshold is set. When the predicted value of the safety feature exceeds the threshold, a safety warning of the corresponding accident type is triggered.
9. A computer device, characterized in that: It includes a processor, a memory and a bus, the memory stores machine-readable instructions executable by the processor, when the computer device is running, the processor and the memory communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the method for safety situation warning of a large-capacity battery energy storage power station as described in any one of claims 1-7.
10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for early warning of safety situation of a large-capacity battery energy storage power station as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Intelligent safety early warning protection method for energy storage power station
CN117437738A
Large-scale energy storage power station operation unit fault early warning method and system
CN110535159A
Demand situation awareness method under power Internet of Things architecture
CN113673846A
Cited By
Energy storage power station early warning method and device based on microparticle fault detection
CN120498006A
