A control method for false alarm of energy storage system fire protection

Abstract

A method for controlling false fire alarms in energy storage systems is disclosed. The method involves acquiring data streams of current, voltage, power, and heat energy from the energy storage system. The fluctuation period, derivative, and probability density function are calculated using Fast Fourier Transform (FFT) to determine the system state and predict related trends. Power is predicted using an autoregressive moving average algorithm and a support vector machine (SVM), and anomalies are identified based on a threshold set according to the probability density function. For heat energy data, derivative calculation and signal analysis are performed, followed by the use of deep convolutional adversarial neural networks (DANNNs) and SVMs to identify the state and issue warnings. A long short-term memory (LSTM) neural network is used to determine the normal range of the monitored data. For abnormal data, cubic exponential smoothing and Gaussian process regression analysis are used to predict trends and physical quantity relationships. Weighted fusion and an adaptive Simpson algorithm are then used to assess potential risks, effectively reducing false fire alarms, accurately identifying fire risks, and ensuring the safety of the energy storage system.

Description

Technical Field

[0001] This invention belongs to the field of fire false alarm control technology for energy storage systems, and specifically relates to a control method for fire false alarms in energy storage systems. Background Technology

[0002] During the operation of energy storage systems, various sensors continuously monitor system operating parameters such as current, voltage, power, and heat energy due to complex internal electrochemical reactions, electrical equipment operation, and environmental factors. This data is crucial for timely detection of potential system faults and prevention of fire accidents. However, a major technical challenge is the complex and variable operating environment of energy storage systems, leading to significant noise interference and uncertainty in the sensor data. These interference factors can easily cause traditional fire alarm systems to misjudge, resulting in false fire alarms.

[0003] On the one hand, normal system fluctuations, such as changes in current and voltage during charging and discharging, and temperature fluctuations during heat dissipation, may be misinterpreted as abnormalities and trigger alarms. This not only interferes with maintenance personnel's judgment of the system's true state and reduces work efficiency, but also may cause maintenance personnel to become complacent about alarms due to frequent false alarms, making it impossible to respond promptly when a real fire hazard occurs. On the other hand, some minor faults may initially exhibit characteristics similar to normal fluctuations. If they cannot be accurately distinguished, the fire protection system may miss the optimal warning opportunity, thereby increasing the risk of fire and posing a serious threat to personnel safety, equipment assets, and the surrounding environment.

[0004] Furthermore, most existing fire alarm technologies for energy storage systems are based on single parameters or simple threshold judgments, lacking the ability to comprehensively analyze multi-source data and accurately model and predict system operating status. They cannot deeply mine and analyze historical and real-time monitoring data, making it difficult to accurately identify true fire risk signals and meet the high reliability and high safety fire early warning requirements of energy storage systems under complex operating conditions.

[0005] Therefore, there is an urgent need to develop a control method that can effectively reduce false fire alarms and accurately identify fire risks in energy storage systems. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a control method for false fire alarms in energy storage systems, which can predict the development trend and severity of fire risks in advance, provide maintenance personnel with sufficient time to formulate response strategies, minimize the possibility of fire accidents and the extent of losses after a fire, ensure the safe and stable operation of energy storage systems, and also effectively protect the surrounding environment and personnel safety.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A method for controlling false alarms in fire protection systems of energy storage systems, comprising the following steps:

[0009] S1. Obtain the real-time data stream of current, voltage and power values ​​transmitted by the sensors of the energy storage system, calculate the fluctuation period of the system operating speed through fast Fourier transform, calculate the second derivative of the fluctuation period, and obtain the probability density function of the data change trend.

[0010] S2. Based on the probability density function of the data change trend, set the quantiles of the normal operating range under different states. Combined with the pre-set empirical upper limit threshold of the normal operating range, if the probability of the data change trend exceeds the upper limit threshold, an alarm mechanism is triggered.

[0011] S3. Collect thermal energy changes from multiple time points, obtain the first derivative distribution of thermal energy data with respect to time, obtain the second derivative curve of temperature change data with respect to time, and construct a set of thermal energy change rate index data.

[0012] S4. Train a support vector machine model based on the data set of thermal energy change rate index, extract the classification boundary between normal state data and abnormal state data, and construct a state recognition mechanism based on the weight ratio of each state.

[0013] S5. A state recognition mechanism based on support vector machine: when the monitoring data of the energy storage system triggers the upper limit of the preset safe operating range threshold, the difference of historical information records is compared to determine the safety status level and then the corresponding warning mode is triggered.

[0014] S6. By training a long short-term memory neural network model, using existing data change trend data as features, extracting the time-series features of data change trend data and monitoring data intensity changes, thereby determining the feature differences between the historical state data set and the current state data set, and outputting the normal operating range of each monitoring data of the system.

[0015] S7. When the monitoring data deviates from the normal operating range after being judged by the long short-term memory neural network, the monitoring data is included in the abnormal database. The intensity of the monitoring data of rare phenomena in the abnormal database is fitted by the triple exponential smoothing method to obtain the probability density distribution of the predicted data change trend of the future time series.

[0016] S8. Calculate the standard deviation of data over a period of time based on the results of rare phenomenon identification, substitute it into the covariance kernel function of Gaussian process for regression analysis, and thus obtain the functional relationship between the predicted data change trend and the frequency of change of physical quantities.

[0017] S9. Based on the relationship between the probability density distribution of the predicted data change trend and the frequency of change of physical quantities, a weighted fusion is performed, and the area integral of the function image is performed using the adaptive Simpson algorithm. Based on the size of the area of ​​the integral result, it is determined whether there is a potential security risk.

[0018] Preferably, the sub-step of step S1 is:

[0019] S1.1 Acquire the current, voltage, and power data streams output by the sensors of the energy storage system;

[0020] S1.2 Calculate the fast Fourier transform of the three parameter data streams of current value, voltage value and power value to obtain frequency domain component signals, and determine the system fluctuation period based on the correspondence between specific frequency components in the frequency domain component signals and fluctuation period;

[0021] S1.3 Calculate the second derivative value based on the obtained fluctuation cycle information to obtain the fluctuation state change rate index;

[0022] S1.4. Calculate the probability density function based on the numerical value of the rate of change of the fluctuation state, and determine the statistical distribution probability of the system operating in each fluctuation range;

[0023] S1.5. Determine the current energy storage state of the energy storage system by using the existing set of relationships between probability density functions and energy storage states;

[0024] S1.6 Using known voltage and current data, combined with the autoregressive moving average algorithm, predict the voltage and current trends over a period of time.

[0025] S1.7. Using the obtained voltage and current trends, input them into the support vector machine to determine the output power value within a preset period after the corresponding energy storage system changes.

[0026] Preferably, the sub-step of step S2 is as follows:

[0027] S2.1 Collect status data to construct a dataset. The dataset includes historical data under multiple different states. Input the status data into the probability density function model to obtain the probability density function corresponding to the data change trend and obtain the probability value.

[0028] S2.2 Using the obtained probability density function, based on the pre-set quantile values, the function is solved in reverse from the quantile values ​​to obtain the normal operating range corresponding to each state, and the corresponding upper and lower threshold values ​​are obtained to construct the range interval of different quantile points corresponding to each state.

[0029] S2.3 Obtain and calculate the state data. Calculate the statistical results of the current data using histogram statistics and use these statistical results to determine the probability of obtaining the current state data.

[0030] S2.4. Compare the probability of the current state data with the probability corresponding to different quantiles of the state, and make a threshold judgment;

[0031] S2.5 If the probability value of the current state is greater than the probability of occurrence corresponding to the preset quantile threshold, an alarm signal is triggered, and the current state data and related time information are recorded for subsequent analysis.

[0032] Preferably, the sub-step of step S3 is as follows:

[0033] S3.1 The time-domain sensor records and acquires the thermal energy data stream of the object at multiple moments, calculates the difference in thermal energy values ​​between adjacent moments, and obtains the first derivative of thermal energy with respect to time.

[0034] S3.2. Divide the obtained thermal energy difference data between adjacent time points by the time difference to obtain temperature change data;

[0035] S3.3. Obtain temperature change data using the difference method to obtain the second derivative curve of temperature with respect to time;

[0036] S3.4 Statistically count the frequency of the second derivative of temperature exceeding the preset range within a preset time period, and construct an index reflecting the dynamic fluctuation amplitude of thermal energy and the rate of change of thermal energy.

[0037] S3.5. Perform time-domain signal analysis on the acquired thermal energy values ​​and calculate the Fourier transform of their frequency components; for discrete thermal energy signal time series... , Given the sequence length, its discrete Fourier transform The expression is:

[0038] ;

[0039] in, Corresponding to different discrete frequency points, and = 0, 1, 2,……, N - 1;

[0040] The imaginary unit satisfies , Let pi be the mathematical constant; the discrete Fourier transform of this value yields the spectral components of the thermal signal at different discrete frequency points; then, these spectral components are analyzed to determine the proportion of high-frequency components. If the high-frequency components exceed a predetermined proportion threshold, the next step is executed:

[0041] Within a set of second-order derivative values ​​corresponding to temperature changes at different times, the Loop Outlier Detection (LOF) algorithm is used to detect and identify moments that deviate from the rest of the data. It is a time series of thermal energy signals. It is the distance between moments;

[0042] If it is a continuous thermal energy signal time series Its Fourier transform The expression is:

[0043]

[0044] Similarly, through the The analysis determines the proportion of high-frequency components, and the subsequent processing is the same as for the discrete case; It is a continuous frequency;

[0045] S3.6 If an error is detected at any point, proceed to the next step:

[0046] Thermal energy change data is imported into a trained deep convolutional adversarial neural network model. The model uses an encoder to map the thermal energy features into a hidden space vector, and a discriminator distinguishes the probability that the hidden space vector is real data.

[0047] Preferably, the sub-step of step S4 is as follows:

[0048] S4.1. Based on the time series of collected thermal energy data, obtain a set of thermal energy change rate data;

[0049] S4.2 Normalize all data in the thermal energy change rate dataset to obtain a normalized dataset, and then set normal state data labels and abnormal state data labels according to preset conditions.

[0050] S4.3 Obtain labeled normal and abnormal state data sets from the normalized dataset, train the support vector machine model, and obtain the optimal support vector machine classification model;

[0051] S4.4 Using all the labeled normal and abnormal state data, obtain the classification boundary hyperplane equation of the support vector machine classification model. , Represents the rate of change of thermal energy. It is a weight vector. It is a bias term;

[0052] S4.5. By statistically analyzing the occurrence frequency of normal and abnormal states of the labeled data, and obtaining the ratio of the two frequencies, the state weight ratio value is determined.

[0053] S4.6. Based on the support vector machine classification model and the state weight ratio, the state of the real-time collected thermal energy change rate data is determined. If the number of thermal energy change rates is greater than the classification boundary value and the number of abnormal thermal energy states is greater than the state weight ratio, then an early warning state is obtained.

[0054] S4.7. Collect thermal energy status information corresponding to the thermal energy change rate data in real time, extract related equipment type, environmental information and process parameter data from the database for data fusion, and obtain equipment operation status and maintenance information based on the above data.

[0055] Preferably, the sub-step of step S5 is as follows:

[0056] S5.1 The support vector machine obtains real-time energy storage system operation data from the acquisition equipment, performs preprocessing operations, and generates monitoring data.

[0057] S5.2 When the monitoring data reaches the upper limit of the preset safe operating range threshold, the safety module is triggered, and the operating data is transmitted to the historical database to form an information record.

[0058] S5.3 Obtain all information records from the historical database as comparison objects, calculate the degree of difference between the comparison objects and the current operating data;

[0059] S5.4. Based on the pre-set level rules corresponding to the degree of difference, determine which security level the current monitoring data belongs to and generate a judgment result;

[0060] S5.5. Based on the judgment result, obtain one of the various pre-configured warning modes and use the pre-established support vector machine for state recognition;

[0061] S5.6, State recognition output state signal set, which contains multiple state signals;

[0062] S5.7 If any status signal exceeds the preset safety range, activate the corresponding warning mode and issue an alarm prompt;

[0063] S5.8 The system continuously acquires status signals. If it finds that the safety range is constantly out of control, it continuously issues warnings and repeatedly executes support vector machines to perform status recognition operations and update the comparison objects.

[0064] Preferably, the sub-step of step S6 is as follows:

[0065] S6.1. Obtain historical data from each type of monitoring point, split the data, and use the first historical dataset as the training set, the second historical dataset as the test set, and the third historical dataset as the validation set.

[0066] S6.2. Perform data cleaning, data denoising, data transformation and feature extraction on the training set data. The features include trend, intensity and fluctuation range, and construct feature vectors to obtain the initial training set.

[0067] S6.3. Based on the initial training set data, a feature difference extraction model is constructed using the Long Short-Term Memory Neural Network algorithm. The LSTM model is defined with 2 hidden layers, 64 neurons, and a learning rate of 0.01. The objective function of the model is set as the mean squared error function, and the number of iterations is 500 to obtain the first training model.

[0068] S6.4 Using validation set data, input the first training model to obtain the first predicted value, and obtain the accuracy by using the historical normal range of the validation set historical data;

[0069] S6.5 If the model accuracy is lower than the threshold, modify the network structure parameters, reset the learning rate to 0.005 and the number of hidden layers to 3, update the first training model to the second training model; return to step S6.4;

[0070] If the model accuracy reaches the threshold, the first final model is obtained;

[0071] Input the real-time monitoring data to retrieve the data from the current time. N Minute-by-minute data is used for data preprocessing to obtain real-time trend data and intensity data;

[0072] Input real-time trend data and intensity data, calculate feature differences through the first final model, compare with normal range values, and determine whether the value of each monitoring point is normal or not.

[0073] Preferably, the sub-step of step S7 is as follows:

[0074] S7.1 Obtain the abnormal monitoring data stream after it has been determined by the Long Short-Term Memory Network, and determine that the current abnormal data stream has exceeded the preset normal data range;

[0075] S7.2 Determine the time point when the current abnormal data occurred based on the monitoring duration of the data points in the abnormal monitoring data stream;

[0076] S7.3 Obtain the intensity value of the data points in the current abnormal data stream, and use the intensity corresponding to all historical abnormal monitoring data points to obtain the optimal value of the three smoothing coefficients in the triple exponential smoothing method through the least squares method.

[0077] S7.4. Based on the triple exponential smoothing method and the smoothing coefficient, fit the intensity of all abnormal monitoring data points in the time period after the current time by using the intensity of the abnormal monitoring data points that have occurred at the current time.

[0078] S7.5. Based on the intensity values ​​corresponding to all fitted anomaly monitoring data points, calculate the probability density distribution function of the intensity corresponding to all anomaly monitoring data points. The calculation formula is as follows:

[0079] P(x)=(1 / (σ√(2π)))e^(-((x - μ)^2) / (2σ^2)),

[0080] Where P(x) represents the probability density of data point x, μ represents the mean of the set of data points, σ ​​represents the standard deviation of the set of data points, e represents the base of the natural logarithm, and π represents pi.

[0081] S7.6 Obtain the probability density of the intensity corresponding to all calculated abnormal monitoring data points, and determine the extreme value range of abnormal data intensity by using a preset threshold method;

[0082] S7.7 Based on the extreme value range of abnormal data intensity and the current abnormal data monitoring time point, obtain the changing trend of this abnormal data in the future.

[0083] Preferably, the sub-step of step S8 is as follows:

[0084] S8.1 After obtaining the time series of historical data, identify rare events in the time series and obtain the standard deviation of the time series based on this result, and record the standard deviation as the first standard deviation;

[0085] S8.2 After obtaining the first standard deviation and the time series of the corresponding time interval, perform data preprocessing, obtain the second time series by median filtering, obtain the time of each data point in the second time series and label it as the second time point;

[0086] S8.3. Using the second time series and the corresponding second time point, take the data point as the center, and calculate the standard deviation of the three data points before and after it. This standard deviation is then used as the standard deviation of the neighborhood of each data point at the second time point. The standard deviations of all neighborhoods are then combined with the first standard deviation and substituted into the covariance kernel function of the Gaussian process. Among them and This represents data at different points in time within a time series. As the independent variable, As the dependent variable,

[0087] ,

[0088] It is the standard deviation. It is the characteristic length of time;

[0089] S8.4. Based on the covariance kernel function, a Gaussian process regression model is constructed, and regression analysis is performed on the processed second time series to obtain an initial function model between the time series change and the frequency of physical quantity change. This model is constructed based on the regression analysis and the second time series.

[0090] S8.5. Based on the initial function model, the processed time series is segmented. By setting a time length, a third time point and multiple third time series are obtained. The standard deviation and frequency of change of each third time series are calculated to generate a frequency feature vector group.

[0091] S8.6. Based on the initial function model, perform predictive analysis, input the elements in the frequency feature vector group into the initial function model respectively, obtain the fourth time series and the corresponding fourth time point, and calculate the difference between the fourth time point series and the third time point series to obtain the frequency error.

[0092] Preferably, the sub-step of step S9 is as follows:

[0093] S9.1. Based on multiple related attribute data in the equipment operation time sequence data stream, establish a matrix of signals in the acquisition equipment. A row or column in this matrix is ​​called a signal vector. The signal amplitude value at each acquisition time constitutes a vector component. Perform a discrete Fourier transform on a vector to obtain information data representing the frequency of change of physical quantity.

[0094] S9.2 Obtain the magnitude data of the vector after discrete Fourier transform and determine the information entropy value of each data according to the Shannon information entropy calculation method. Use the maximum entropy among all data at multiple acquisition times as the measure of risk weight, and then obtain the value of each vector after fusion in each column or row of the matrix.

[0095] S9.3 Perform principal component analysis and eigenvalue decomposition on the matrix to obtain the importance weights of the main vector groups of the signal represented by different eigenvalues ​​arranged in descending order. After all eigenvalues ​​are weighted proportionally, the weights are accumulated to obtain the normalization factor of the weighted eigenvector group.

[0096] S9.4. Collect and obtain the probability density function that represents the probability of the predicted data change trend. The input of the probability density function is the normalization factor of the weight feature vector group obtained in the previous step. After the output of the probability density function is weighted and calculated with the weight, if there is a distribution shift of the probability density function, the gradient descent method is used to iteratively and adaptively correct the parameters in the probability density function in response to the shift phenomenon.

[0097] S9.5. By obtaining the data parameters of the predicted probability density function in the previous step, the upper and lower limits of the integration interval in the probability density function are determined, and the integral function body based on the adaptive Simpson algorithm is obtained.

[0098] S9.6 Determine the relationship between the set precision value and the pre-set termination threshold when using the integration method to calculate the area of ​​the function. If the precision value is greater than the threshold, obtain the value of half of the current precision, use this value as the adaptive value of the Simpson algorithm, and calculate the area of ​​the sub-interval of the function obtained in this step. The areas of multiple sub-intervals obtained in this step are combined to obtain the integration result of this step.

[0099] S9.7 Based on all the above-calculated results, determine the specific location and equipment attribute information corresponding to different locations of the equipment represented by each data result. The signal matrices of different locations of the equipment at different times constitute a tensor. Use the tensor to calculate the risk value of different components and locations of the equipment in a certain time series for judgment. Determine the value and the risk threshold. If the risk value is greater than or equal to the risk threshold, it is determined that the corresponding component has potential risk.

[0100] The present invention can achieve the following beneficial effects:

[0101] 1. This invention performs comprehensive and in-depth analysis and processing of multi-source data collected by sensors in the energy storage system. Utilizing complex algorithms such as the Fast Fourier Transform, it accurately calculates the fluctuation period of the system's operating speed and related indicators, effectively distinguishing between normal operating fluctuations and abnormal danger signals. Compared to traditional methods based on simple threshold judgments, this significantly reduces the probability of false alarms caused by normal system fluctuations, thus improving the accuracy of fire alarms.

[0102] 2. Based on the probability density function of data change trends, this invention sets quantiles and empirical upper limit thresholds for the normal operating range under different states, as well as detailed analysis and processing of thermal energy data. This allows for more accurate detection of potential system anomalies, avoiding unnecessary alarms triggered by data interference or misjudgments. It ensures that the fire alarm system only issues an alarm when there is a real fire risk, reducing the workload of maintenance personnel in investigating and handling false alarms, and improving the efficiency of the entire energy storage system's operation and maintenance management.

[0103] 3. This invention utilizes a support vector machine model trained on a dataset of thermal energy change rate indicators. This model accurately extracts the classification boundary between normal and abnormal state data. Based on a robust state recognition mechanism constructed using state weight ratios, it can determine the operating status of the energy storage system in real time and with precision. This not only promptly detects current anomalies but also predicts potential future system failure risks based on historical data and current state trends. This provides strong support for preventative maintenance, effectively extending the lifespan of the energy storage system, reducing equipment failures and downtime, and improving the overall reliability and availability of the energy storage system.

[0104] 4. This invention extracts and analyzes the temporal characteristics of data change trends and monitoring data intensity changes using a long short-term memory neural network model. This allows for in-depth analysis of the characteristic differences between historical and current state data sets, thereby accurately outputting the normal operating range of each monitoring data point in the system. This helps to detect subtle abnormal trends in monitoring data in advance, enabling timely adjustments or repairs before a fault occurs, further improving the stability and safety of the energy storage system. It also provides data support for optimized system operation, achieving performance optimization and improvement of the energy storage system.

[0105] 5. This invention employs a cubic exponential smoothing method to fit the monitoring data intensity of rare phenomena in anomaly databases, obtaining the probability density distribution of predicted data change trends for future time series. Furthermore, it substitutes the data standard deviation into the covariance kernel function of a Gaussian process for regression analysis to obtain the functional relationship between predicted data change trends and the frequency of changes in physical quantities. Based on these results, weighted fusion and adaptive Simpson algorithm integration are performed to determine potential safety risks. This multi-dimensional, multi-level risk assessment method comprehensively considers various operating parameters and historical data characteristics of the energy storage system, providing a comprehensive and in-depth assessment of the potential fire risks the system may face. Attached Figure Description

[0106] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0107] Figure 1 This is a logic flowchart of the present invention. Detailed Implementation

[0108] Preferred solutions include Figure 1 As shown, a control method for false fire alarms in an energy storage system includes the following steps:

[0109] S1. Acquire the real-time data streams of current, voltage, and power values ​​transmitted by the sensors of the energy storage system. Calculate the fluctuation period of the system's operating speed using a Fast Fourier Transform (FFT). Calculate the second derivative of the fluctuation period to obtain the probability density function of the data change trend; specifically:

[0110] S1.1 Acquire the current, voltage, and power data streams output by the sensors of the energy storage system;

[0111] In practical operation, it is necessary to ensure the accuracy and stability of the sensors. Their sampling frequency should be reasonably set according to the characteristics of the energy storage system and the requirements for real-time data. For example, for some fast-response energy storage systems, the sampling frequency may need to reach thousands of times per second to accurately capture minute changes in current, voltage, and power. These sensors should be evenly distributed in key parts of the energy storage system, such as battery module connection points and power conversion circuits, to comprehensively reflect the system's operating status.

[0112] S1.2 Calculate the fast Fourier transform of the three parameter data streams of current value, voltage value and power value to obtain the frequency domain component signal. Determine the system fluctuation period based on the correspondence between specific frequency components in the frequency domain component signal and the fluctuation period (for example, according to the frequency and period conversion rules determined by empirical formulas or industry standards).

[0113] The implementation of the Fast Fourier Transform (FFT) algorithm requires efficient computing resources, such as dedicated digital signal processors (DSPs) or field-programmable gate arrays (FPGAs) with powerful computing capabilities. For different types of energy storage systems, the correspondence between specific frequency components and fluctuation periods may differ, necessitating extensive experimentation and data analysis to determine accurate conversion rules. For example, in lithium-ion battery energy storage systems, the internal chemical reactions and electrical characteristics determine that certain frequency ranges are closely related to the battery's charge / discharge state and health. By analyzing these frequency components, the system's fluctuation period can be accurately determined, thereby understanding the system's operational stability.

[0114] S1.3 Calculate the second derivative value based on the obtained fluctuation cycle information to obtain the fluctuation state change rate index;

[0115] This step not only reflects the current fluctuation state of the system but also provides important basis for subsequent predictions of the system's changing trends. The calculation of the second derivative requires precise numerical methods to avoid misjudging the system state due to computational errors. Simultaneously, to improve computational efficiency, appropriate preprocessing of the data can be performed, such as data smoothing to remove noise interference; however, care must be taken not to over-smooth the data and lose important fluctuation information.

[0116] S1.4. Calculate the probability density function based on the numerical value of the rate of change of the fluctuation state, and determine the statistical distribution probability of the system operating in each fluctuation range;

[0117] When calculating the probability density function, methods such as kernel density estimation can be used to construct a probability distribution model of the system's operating state based on historical data and the current rate of change of fluctuation states. Analyzing this model provides a clear understanding of the probability of the system occurring in different fluctuation ranges, thus providing a quantitative basis for determining whether the system is operating normally. For example, if the probability of the system occurring in a certain abnormal fluctuation range exceeds a set threshold, it may indicate a potential risk of system failure.

[0118] S1.5. Determine the current energy storage state of the energy storage system by using the existing set of relationships between probability density functions and energy storage states;

[0119] This set of relationships was established through the analysis and research of a large amount of historical data, encompassing the probability density function characteristics of energy storage systems in different energy storage states. For example, the fluctuation patterns of current, voltage, and power, as well as the corresponding probability density functions, differ in different states such as charging, discharging, and resting. By comparing the currently calculated probability density function with this set of relationships, the specific energy storage state of the energy storage system can be accurately determined.

[0120] S1.6 Using known voltage and current data, combined with an autoregressive moving average algorithm (where the parameters of the autoregressive moving average algorithm are set as follows: autoregression order p = 3, moving average order q = 2, and the range of these parameters is determined based on preliminary analysis of historical data and industry experience to effectively handle this type of data), predict the voltage and current trends for a period of time to come.

[0121] S1.7. The obtained voltage and current trends are input into the support vector machine (the support vector machine uses a radial basis kernel function, and the kernel function parameter gamma is selected in the range of [0.1, 10] through cross-validation. The penalty parameter C is selected in the range of [1, 100] to balance the model's fitting ability and generalization ability) to determine the output power value within the preset period after the corresponding energy storage system changes.

[0122] Radial basis function kernels can effectively handle nonlinear relationships. By selecting gamma and C parameters within a given range using cross-validation, the most suitable model configuration for the current data can be found. During cross-validation, the dataset is divided into training, validation, and test sets. Parameters are continuously adjusted to optimize model performance on the validation set, and then the model is finally evaluated on the test set. The resulting support vector machine model can accurately predict the output power value within a preset period based on the input voltage and current trends, providing crucial information for energy management and control of energy storage systems.

[0123] S2. Based on the probability density function of the data change trend, set the quantiles of the normal operating range under different states. Combined with the pre-set empirical upper limit threshold of the normal operating range, if the probability of the data change trend exceeds the upper limit threshold, an alarm mechanism is triggered; specifically:

[0124] S2.1 Collect status data (status data includes temperature, pressure and ambient humidity data of key components of the energy storage system) to build a dataset. The dataset includes historical data under multiple different states. Input the status data into the probability density function model to obtain the probability density function corresponding to the data change trend and obtain the probability value.

[0125] When collecting status data, high-precision sensors must be used to ensure accuracy. For example, the accuracy of temperature sensors should be ±0.1℃, pressure sensors should be within ±0.01MPa, and the accuracy of ambient humidity sensors should also be controlled within a certain error range. These sensors should be installed near key components of the energy storage system, such as battery modules, heat dissipation devices, and electrical connections, to monitor the system's operating status in a timely and accurate manner. For different types of energy storage systems, the key components and corresponding status data may differ, requiring targeted data collection and analysis based on specific circumstances. By inputting a large amount of historical status data into the probability density function model, the relationship between the data change trends of the system under different states and the probability density function can be established, thus providing a basis for setting the subsequent normal operating range.

[0126] S2.2 Using the obtained probability density function, based on the pre-set quantile values ​​(for example, setting the 95th quantile as the basis for judging the normal operating range, which can be determined based on a large amount of historical data statistical analysis and the safety requirements of the energy storage system), the quantile values ​​are solved in reverse to obtain the normal operating range corresponding to each state, and the corresponding upper and lower thresholds are obtained, and the range interval of different quantiles corresponding to each state is constructed.

[0127] When determining the quantile value, factors such as the reliability requirements of the energy storage system, the probability of failure, and the false alarm rate need to be comprehensively considered. For some energy storage systems with extremely high safety requirements, a higher quantile value, such as 99%, may be chosen to ensure that the system can be judged as operating normally in the vast majority of cases. For some systems that are more sensitive to cost and efficiency, the quantile value can be appropriately reduced, but the basic safety of the system must still be guaranteed. By solving the probability density function in reverse, the specific values ​​of the normal operating range in each state can be obtained. These values ​​will serve as important criteria for judging whether the system is operating normally.

[0128] S2.3 Obtain and calculate the state data. Calculate the statistical results of the current data using histogram statistics and use these statistical results to determine the probability of obtaining the current state data.

[0129] Histogram statistics are an intuitive and effective data analysis method. They divide data into intervals and count the quantity or frequency of data within each interval. In this step, based on a pre-defined interval width, the current data is categorized and statistically analyzed to obtain the distribution of data across each interval, thereby calculating the probability of the current data. For example, if the current temperature data has a high frequency within the middle interval of the normal operating range, its probability is higher; conversely, if the data is close to or exceeds the boundary interval of the normal operating range, its probability will decrease accordingly.

[0130] S2.4. Compare the probability of the current state data with the probability corresponding to different quantiles of the state, and make a threshold judgment;

[0131] This step is crucial in determining whether the system is operating normally. The calculated probability of the current state data is compared with the probabilities corresponding to different pre-defined quantiles. If the current probability exceeds the upper threshold, it indicates that the system may be in an abnormal state, requiring further analysis and processing. When determining the threshold, the volatility and uncertainty of the data must be considered to avoid misjudgments due to random factors. Filtering algorithms or statistical tests can be used to further verify and confirm the data.

[0132] S2.5 If the probability value of the current state is greater than the probability of occurrence corresponding to the preset quantile threshold, an alarm signal is triggered, and the current state data and related time information are recorded for subsequent analysis.

[0133] Once an alarm signal is triggered, relevant information should be promptly transmitted to the monitoring center or the operator's terminal device to facilitate appropriate action. Recorded status data and time information should include detailed sensor data, data acquisition time, system operating status, etc. This information will provide crucial data for subsequent troubleshooting and system optimization. For example, analyzing historical alarm data can help identify patterns and causes of system failures, allowing for targeted improvements to enhance system reliability and stability.

[0134] S3. Collect thermal energy changes from multiple time points in the data stream, obtain the first derivative distribution of the thermal energy data with respect to time, obtain the second derivative curve of the temperature change data with respect to time, and construct a data set of thermal energy change rate index; specifically:

[0135] S3.1 The time-domain sensor records and acquires the thermal energy data stream of the object at multiple moments, calculates the difference in thermal energy values ​​between adjacent moments, and obtains the first derivative of thermal energy with respect to time.

[0136] The sampling frequency of the time-domain sensor should be set appropriately based on the rate of thermal energy change and the required data accuracy. Generally, for energy storage systems with rapid thermal energy changes, the sampling frequency should be higher, such as more than 10 times per second. When calculating the difference in thermal energy values ​​between adjacent time points, attention should be paid to the accuracy and stability of the data to avoid deviations in the calculation results due to sensor errors or data transmission interference. Methods such as averaging multiple measurements or data filtering can be used to improve the accuracy of the calculation.

[0137] S3.2. Divide the obtained thermal energy difference data between adjacent time points by the time difference to obtain temperature change data;

[0138] The time difference should remain constant and is typically determined based on the sampling frequency. For example, if the sampling frequency is 5 times per second, then the time difference is 0.2 seconds. By dividing the thermal energy difference by the time difference, the change in thermal energy per unit time, i.e., the temperature change data, can be obtained. These data reflect the rate of temperature change of the energy storage system at different times and are important indicators for judging the thermal stability of the system.

[0139] S3.3. Obtain temperature change data using the difference method to obtain the second derivative curve of temperature with respect to time;

[0140] Finite difference calculation is a commonly used numerical calculation method for approximating the derivative of a function. In this step, by performing a difference operation on the temperature change data, the second derivative curve of temperature with respect to time can be obtained. This curve reflects the acceleration of temperature change, that is, the rate of temperature change. The shape and characteristics of the second derivative curve can provide important clues for judging whether there are abnormalities such as thermal runaway in the system. For example, if the second derivative curve shows a sharp rise or large fluctuations, it may indicate problems such as local overheating or uneven heat transfer within the system.

[0141] S3.4 Statistically count the frequency of the second derivative of temperature exceeding the preset range within a preset time period (the preset range is determined by statistical analysis of historical data on the rate of temperature change during normal operation of the energy storage system, for example, set to ±0.5 degrees Celsius per second), and construct an index reflecting the dynamic fluctuation amplitude of thermal energy and the rate of change of thermal energy.

[0142] When calculating the frequency of occurrences where the second derivative of temperature exceeds a preset range, the length of the preset time should be determined based on the characteristics of the energy storage system and the timeliness requirements for fault detection. For example, for some fast-responding energy storage systems, the preset time can be set to a shorter period, such as 10 seconds; while for some slower-changing systems, the preset time can be appropriately extended, such as 1 minute. By statistically analyzing the frequency of occurrences, an index reflecting the dynamic fluctuation amplitude of thermal energy can be constructed. This index can intuitively reflect the drastic degree of thermal energy change in the system over a period of time, providing a basis for subsequent anomaly detection and risk assessment.

[0143] S3.5. Perform time-domain signal analysis on the acquired thermal energy values ​​and calculate the Fourier transform of their frequency components; for discrete thermal energy signal time series... , Given the sequence length, its discrete Fourier transform The expression is:

[0144] ;

[0145] in, Corresponding to different discrete frequency points, and = 0, 1, 2,……, N - 1;

[0146] The imaginary unit satisfies , Let π be the mathematical constant. The discrete Fourier transform yields the spectral components of the thermal signal at different discrete frequency points. These spectral components are then analyzed to determine the proportion of high-frequency components. If the high-frequency components exceed a predetermined threshold (determined by comparing the high-frequency components of the thermal signal under normal and fault conditions of the energy storage system, for example, set to 30%), then the next step is executed:

[0147] In a set constructed from the second derivative values ​​corresponding to temperature changes at different times, the Loop Outlier Detection (LOF) algorithm (with the parameter minPts set to 5, determined based on the local density distribution of data points to effectively identify outliers) is used to detect and identify moments deviating from other data. It is a time series of thermal energy signals. It is the distance between moments;

[0148] If it is a continuous thermal energy signal time series Its Fourier transform The expression is:

[0149]

[0150] Similarly, through the The analysis determines the proportion of high-frequency components, and the subsequent processing is the same as for the discrete case; It is a continuous frequency.

[0151] S3.6 If an error is detected at any point, proceed to the next step:

[0152] Thermal energy change data was imported into a trained deep convolutional adversarial neural network (DCNN) model (the DCNN model structure consists of 3 convolutional layers with kernel sizes of 3x3, 5x5, and 3x3, and the number of kernels in each layer being 16, 32, and 64, respectively, using the ReLU activation function; 2 fully connected layers with 128 and 64 neurons, respectively; the learning rate during model training was set to 0.001; the Adam optimization algorithm was used; the training iterations were 1000; and the model structure and parameters were obtained through training and optimization based on a large amount of sample data). The model uses an encoder to map thermal energy features into a hidden space vector, and a discriminator distinguishes the probability that the hidden space vector is real data.

[0153] Deep convolutional adversarial neural networks (DANs) possess powerful feature extraction and classification capabilities for processing image, signal, and other data. In this application, by inputting thermal energy change data into a trained model, the model's encoder automatically extracts deep-level features from the thermal energy data and maps them into hidden space vectors. These hidden space vectors can more effectively represent the intrinsic characteristics of the thermal energy data, facilitating subsequent analysis and processing. The discriminator further judges the authenticity and abnormality of the data by distinguishing the probability that the hidden space vectors represent real data. During model training, a large number of normal and abnormal thermal energy data samples are required. By continuously adjusting the model's parameters, such as the weights and biases of the convolutional kernels, the model can accurately identify different types of thermal energy data. Simultaneously, to avoid overfitting, data augmentation and regularization techniques can be employed to improve the model's generalization ability.

[0154] S4. Train a support vector machine model based on the data set of thermal energy change rate index, extract the classification boundary between normal and abnormal state data, and construct a state recognition mechanism based on the weight ratio of each state; specifically:

[0155] S4.1. Based on the time series of collected thermal energy data, obtain a set of thermal energy change rate data;

[0156] When collecting thermal energy data, it is crucial to ensure data integrity and continuity, avoiding data loss or interruption. Data storage and backup technologies can be employed to regularly save and back up thermal energy data, ensuring access to complete time-series data when needed. Simultaneously, rigorous data quality checks and screening are essential to remove outliers and noisy data, thereby improving data reliability.

[0157] S4.2 Normalize all data in the thermal energy change rate dataset to obtain a normalized dataset, and then set normal state data labels and abnormal state data labels according to preset conditions.

[0158] Normalization can map data of different magnitudes to the same range, such as normalizing data to the interval [0,1] or [-1, 1], which can improve the efficiency and accuracy of model training. When setting data labels, it is necessary to determine which data belongs to the normal state and which belongs to the abnormal state based on the actual operating conditions of the energy storage system and fault diagnosis standards. Labeling rules can be determined by combining expert experience, historical fault data, and relevant industry standards.

[0159] S4.3 Obtain labeled normal and abnormal state data sets from the normalized dataset, train the support vector machine model, and obtain the optimal support vector machine classification model;

[0160] When training a support vector machine (SVM) model, it is crucial to select the appropriate kernel function and adjust the model parameters. As mentioned earlier, a radial basis function (RBF) kernel is used, and the gamma and C parameters are selected within a given range using cross-validation to optimize model performance. Simultaneously, the dataset should be divided into training, validation, and test sets. During training, the model's performance on the validation set is continuously evaluated, and the model parameters are adjusted based on the evaluation results until the model achieves satisfactory performance metrics on the validation set.

[0161] S4.4 Using all the labeled normal and abnormal state data, obtain the classification boundary hyperplane equation of the support vector machine classification model. , Represents the rate of change of thermal energy. It is a weight vector. It is a bias term;

[0162] S4.5. By statistically analyzing the occurrence frequency of normal and abnormal states of the labeled data, and obtaining the ratio of the two frequencies, the state weight ratio value is determined.

[0163] S4.6. Based on the support vector machine classification model and the state weight ratio, the state of the real-time collected thermal energy change rate data is determined. If the number of thermal energy change rates greater than the classification boundary value and the number of abnormal thermal energy states (by setting a sliding time window, the number of abnormal states within the window is counted, and the window size is set to 10 sampling periods according to the time characteristics of thermal energy change and system response requirements) is greater than the state weight ratio, then an early warning state is obtained.

[0164] S4.7. Collect thermal energy status information corresponding to the thermal energy change rate data in real time, extract related equipment type, environmental information and process parameter data from the database for data fusion, and obtain equipment operation status and maintenance information based on the above data.

[0165] S5. A state recognition mechanism based on support vector machines: when the monitoring data of the energy storage system triggers the upper limit of the preset safe operating range threshold, the difference between historical information records is compared to determine the safety state level, and then the corresponding warning mode is triggered; specifically:

[0166] S5.1 The support vector machine obtains real-time energy storage system operation data from the acquisition equipment, performs preprocessing operations, and generates monitoring data.

[0167] The raw data collected by the acquisition equipment may contain noise, missing data, or inconsistent data formats. Preprocessing operations include data cleaning to remove obvious errors and outliers; data completion to fill in missing data using methods such as interpolation; and data format conversion to unify different types of data into a format suitable for subsequent analysis. After these preprocessing steps, the obtained monitoring data can more accurately reflect the actual operating status of the energy storage system, providing a reliable data foundation for subsequent status identification and judgment.

[0168] S5.2 When the monitoring data reaches the upper limit of the preset safe operating range threshold, the safety module is triggered, and the operating data is transmitted to the historical database to form an information record.

[0169] The triggering mechanism of the safety module should have high sensitivity and a low false alarm rate. The setting of the upper threshold needs to comprehensively consider the design parameters of the energy storage system, safety standards, and a large amount of experimental and actual operation data. Once triggered, the operational data should be promptly stored in the historical database for subsequent analysis and comparison. The historical database should have good data storage and management functions, enabling rapid data retrieval and access, and ensuring data integrity and security to prevent data loss or tampering.

[0170] S5.3 Obtain all information records from the historical database as comparison objects, calculate the degree of difference between the comparison objects and the current operating data (the degree of difference is calculated using the Euclidean distance formula, taking into account the differences in each data dimension).

[0171] The Euclidean distance formula effectively measures the distance between two data points in a multidimensional space. In this application, the current operating data and the data in the historical database are considered as points in a multidimensional space. By calculating the Euclidean distance between them, we can intuitively understand the degree of deviation between the current data and the historical data. For example, if the current data differs significantly from the historical data in multiple dimensions such as current, voltage, and temperature, the calculated Euclidean distance will be large, indicating that the system may be in an abnormal operating state.

[0172] S5.4. Based on the pre-set grading rules corresponding to the degree of difference (the grading rules are divided into three levels according to the degree of difference: slight difference, moderate difference, and severe difference, corresponding to different safety levels. These rules are determined through simulation of a large amount of historical data and analysis of actual accident cases), determine which safety level the current monitoring data belongs to and generate a judgment result.

[0173] S5.5. Based on the judgment result, obtain one of the various pre-configured warning modes, and use the pre-built support vector machine to perform state recognition (the support vector machine is used here to further refine the analysis of the current state. The support vector machine parameters are consistent with the model parameters trained in S4. By reclassifying and judging the current data, the accuracy of state recognition is improved).

[0174] Different security levels correspond to different warning modes, which can include various forms such as audible alarms, visual alarms, and SMS notifications to maintenance personnel. Simultaneously with issuing a warning, a further detailed analysis of the current state is performed using a Support Vector Machine (SVM) to confirm any system anomalies. Since the SVM has already been trained in S4, its parameters possess a certain degree of reliability and validity. Using a model with the same parameters to classify and judge the current data again can improve the accuracy of state identification and avoid misjudgments or omissions due to the limitations of a single judgment method.

[0175] S5.6, State recognition output state signal set, which contains multiple state signals;

[0176] These status signals reflect the operational status of the energy storage system in various aspects, such as battery health, electrical connection status, and thermal management status. Comprehensive analysis of these status signals provides a more complete understanding of the overall system operation, offering more information for further fault diagnosis and handling.

[0177] S5.7 If any status signal exceeds the preset safety range, activate the corresponding warning mode and issue an alarm prompt;

[0178] For each status signal, a corresponding safety range is set. When a status signal exceeds this range, the corresponding warning mode is immediately activated. For example, if the battery health status signal shows a sharp drop in battery capacity or an abnormal increase in internal resistance, exceeding the preset safety range, a warning mode specifically for battery failure will be triggered to remind maintenance personnel to check and maintain the battery in a timely manner.

[0179] S5.8 The system continuously acquires status signals. If it finds that the safety range is constantly out of control, it continuously issues warnings and repeatedly executes support vector machines to perform status recognition operations and update the comparison objects.

[0180] During the operation of an energy storage system, continuous monitoring of status signals is crucial to ensuring system safety. If a certain status signal remains out of range, it indicates a persistent safety hazard, requiring continuous warnings until the problem is resolved. Simultaneously, to adapt to changes in system operating status, a support vector machine is periodically executed for status identification, and the comparison objects are updated—that is, the latest historical data is compared and analyzed with the current data. This ensures the accuracy and timeliness of status identification and judgment.

[0181] S6. By training a long short-term memory neural network model, using existing data change trend data as features, extracting the temporal characteristics of data change trend data and monitoring data intensity changes, thereby determining the feature differences between the historical state data set and the current state data set, and outputting the normal operating range of each monitoring data point of the system; specifically:

[0182] S6.1. Obtain historical data from each type of monitoring point, split the data, and use the first historical dataset as the training set, the second historical dataset as the test set, and the third historical dataset as the validation set.

[0183] During data acquisition, it is crucial to ensure the integrity and accuracy of data from all types of monitoring points, covering all key components and parameters of the energy storage system, such as the voltage, current, and temperature of battery modules, and the efficiency and power factor of power conversion devices. Data splitting should employ a reasonable ratio, for example, a common practice of 60% for training, 20% for testing, and 20% for validation. This approach fully utilizes historical data for model training and evaluation while ensuring the model possesses good generalization capabilities.

[0184] S6.2. Perform data cleaning, data denoising, data transformation and feature extraction on the training set data. The features include trend, intensity and fluctuation range, and construct feature vectors to obtain the initial training set.

[0185] Data cleaning and denoising can employ filtering algorithms, outlier detection and correction, and other methods to remove noise and erroneous data, improving data quality. Data transformation can include logarithmic transformation, standardization, and other operations to make the data more suitable for model training. Feature extraction is a crucial step; by analyzing features such as trends, intensity, and fluctuation range, the inherent patterns in the data can be captured more effectively. For example, for voltage data, trends can reflect changes in the battery's charging and discharging state, intensity can indicate the magnitude of the voltage, and fluctuation range can indicate voltage stability. These features are combined into feature vectors to provide input data for a long short-term memory neural network model.

[0186] S6.3. Based on the initial training set data, a feature difference extraction model is constructed using the Long Short-Term Memory Neural Network algorithm. The LSTM model is defined with 2 hidden layers (determined based on the temporal complexity of the data and model performance testing), 64 neurons (adjusted through multiple experiments to ensure effective capture of data features), and a learning rate of 0.01 (determined based on the trade-off between model convergence speed and accuracy). The objective function of the model is set as the mean squared error function, and the number of iterations is 500 to obtain the first training model.

[0187] S6.4 Using validation set data, input the first training model to obtain the first predicted value, and obtain the accuracy by using the historical normal range of the validation set historical data;

[0188] The validation set data is input into the first training model to obtain the corresponding predicted values. These predicted values ​​are then compared with the historical normal range of the validation set to calculate the proportion of correct predictions, i.e., the accuracy. If the accuracy is low, it indicates that the model may have problems such as overfitting or underfitting, requiring further adjustment of model parameters or improvement of model structure.

[0189] S6.5 If the model accuracy is lower than the threshold (the threshold is set to 90%, determined according to the system's requirements for data accuracy), then modify the network structure parameters, reset the learning rate to 0.005 and the number of hidden layers to 3, update the first training model to the second training model; return to step S6.4;

[0190] If the model accuracy reaches the threshold, the first final model is obtained;

[0191] Input the real-time monitoring data to retrieve the data from the current time. N Minute data ( NBased on the timeliness of the data and the system response requirements (set to 10 minutes), data preprocessing is performed to obtain real-time trend data and intensity data;

[0192] Input real-time trend data and intensity data, calculate feature differences through the first final model, compare with normal range values, and determine whether the value of each monitoring point is normal or not.

[0193] S7. When monitoring data deviates from the normal operating range after being processed by a long short-term memory neural network, it is included in an anomaly database. A triple exponential smoothing method is used to fit the intensity of monitoring data for rare phenomena in the anomaly database, yielding the probability density distribution of the predicted trend of future time series data. Specifically:

[0194] S7.1 Obtain the abnormal monitoring data stream after it has been determined by the Long Short-Term Memory Network, and determine that the current abnormal data stream has exceeded the preset normal data range;

[0195] After the Long Short-Term Memory (LSTM) neural network identifies anomalies in the monitoring data, these abnormal data are promptly collected to form an anomaly monitoring data stream. Simultaneously, it's crucial to clarify the specific circumstances under which these data deviate from the preset normal data range—for example, whether a single parameter value is excessively high or low, or whether multiple parameters are simultaneously abnormal. This is essential for subsequent analysis and processing.

[0196] S7.2 Determine the time point when the current abnormal data occurred based on the monitoring duration of the data points in the abnormal monitoring data stream;

[0197] By recording the monitoring duration of data points, it's possible to determine when an anomaly first appears, which helps in analyzing its development process and possible causes. For example, if anomalous data suddenly appears after a specific time period, it may be related to changes in the system's operating mode or external environmental interference during that period.

[0198] S7.3 Obtain the intensity values ​​of data points in the current abnormal data stream. Using the intensity values ​​corresponding to all historical abnormal monitoring data points, obtain the optimal values ​​of the three smoothing coefficients in the triple exponential smoothing method through the least squares method (the method is determined based on the distribution characteristics and fitting effect of historical data).

[0199] S7.4. Based on the triple exponential smoothing method and the smoothing coefficient, fit the intensity of all abnormal monitoring data points in the time period after the current time by using the intensity of the abnormal monitoring data points that have occurred at the current time.

[0200] S7.5. Based on the intensity values ​​corresponding to all fitted anomaly monitoring data points, calculate the probability density distribution function of the intensity corresponding to all anomaly monitoring data points. The calculation formula is as follows:

[0201] P(x)=(1 / (σ√(2π)))e^(-((x - μ)^2) / (2σ^2)),

[0202] Where P(x) represents the probability density of data point x, μ represents the mean of the set of data points, σ ​​represents the standard deviation of the set of data points, e represents the base of the natural logarithm, and π represents pi.

[0203] S7.6 Obtain the probability density of the intensity corresponding to all the calculated abnormal monitoring data points, and determine the extreme value range of abnormal data intensity by using the preset threshold method (the preset threshold is determined according to the risk assessment of the abnormal data intensity and the system security requirements, for example, set to 0.05).

[0204] S7.7 Based on the extreme value range of abnormal data intensity and the current abnormal data monitoring time point, obtain the changing trend of this abnormal data in the future.

[0205] S8. Based on the results of rare phenomenon identification, calculate the standard deviation of the data over a period of time, and substitute it into the covariance kernel function of the Gaussian process for regression analysis to obtain the functional relationship between the predicted data change trend and the frequency of change of physical quantities; specifically:

[0206] S8.1 After obtaining the time series of historical data, identify rare events in the time series and obtain the standard deviation of the time series based on this result, and record the standard deviation as the first standard deviation;

[0207] S8.2 After obtaining the first standard deviation and the time series of the corresponding time interval, perform data preprocessing, obtain the second time series by median filtering, obtain the time of each data point in the second time series and label it as the second time point;

[0208] When analyzing historical time series data, appropriate anomaly detection algorithms are used to identify rare events, such as anomaly detection based on statistical models and outlier detection based on machine learning. Once a rare event is identified, the standard deviation of the data within that time period is calculated. This standard deviation reflects the dispersion of the data under that specific condition, and it is recorded as the first standard deviation for subsequent Gaussian process analysis.

[0209] S8.3. Using the second time series and the corresponding second time point, take three data points before and after the data point (the neighborhood range is determined based on the local correlation and noise characteristics of the data) and calculate their standard deviations as the standard deviations of the neighborhood of each data point at the second time point. Then, combine the standard deviations of all neighborhoods with the first standard deviation and substitute them into the covariance kernel function of the Gaussian process. Among them and This represents data at different points in time within a time series. As the independent variable, As the dependent variable,

[0210] ,

[0211] It is the standard deviation. It is the characteristic length of time;

[0212] S8.4. Based on the covariance kernel function, a Gaussian process regression model is constructed. Regression analysis is performed on the processed second time series to obtain an initial function model between the time series change and the frequency of physical quantity change. This model is constructed based on regression analysis and the second time series.

[0213] S8.5. Based on the initial function model, the processed time series is segmented. By setting a time length (e.g., 5 minutes), the third time point and multiple third time series are obtained. The standard deviation and frequency of change of each third time series are calculated to generate a frequency feature vector group.

[0214] S8.6. Based on the initial function model, perform predictive analysis, input the elements in the frequency feature vector group into the initial function model respectively, obtain the fourth time series and the corresponding fourth time, and calculate the difference between the fourth time series and the third time series to obtain the frequency error.

[0215] S8.7 When the frequency error is greater than the set threshold (the threshold is determined according to the system's requirements for prediction accuracy and the sensitivity of the physical quantity to the frequency of change, for example, it is set to 0.1Hz), more time series data are collected and the processing steps from step one to step six are repeated. The initial function model is updated with the new time series. If the frequency error is less than the set threshold, the future time series is predicted according to the initial function model.

[0216] S9. Based on the relationship between the probability density distribution of the predicted data's changing trend and the frequency of change of physical quantities, a weighted fusion is performed. An adaptive Simpson algorithm is then used to perform area integration of the function's image, and the size of the integrated area is used to determine if there are any potential security risks. Specifically:

[0217] S9.1. Based on multiple related attribute data in the equipment operation time sequence data stream, establish a matrix of signals in the acquisition equipment. A row or column in this matrix is ​​called a signal vector. The signal amplitude value at each acquisition time constitutes a vector component. Perform a discrete Fourier transform on a vector to obtain information data representing the frequency of change of physical quantity.

[0218] S9.2 Obtain the magnitude data of the vector after discrete Fourier transform and determine the information entropy value of each data according to the Shannon information entropy calculation method. Use the maximum entropy among all data at multiple acquisition times as the measure of risk weight, and then obtain the value of each vector after fusion in each column or row of the matrix.

[0219] S9.3 Perform principal component analysis and eigenvalue decomposition on the matrix to obtain the importance weights of the main vector groups of the signal represented by different eigenvalues ​​arranged in descending order. After all eigenvalues ​​are weighted proportionally, the weights are accumulated to obtain the normalization factor of the weighted eigenvector group.

[0220] S9.4. Collect and obtain the probability density function that represents the probability of the predicted data change trend. The input of the probability density function is the normalization factor of the weight feature vector group obtained in the previous step. After the output of the probability density function is weighted and calculated, if there is a distribution shift of the probability density function, then the gradient descent method (the method is determined according to the mathematical characteristics of the probability density function and the optimization objective) is used to iteratively and adaptively correct the parameters in the probability density function.

[0221] S9.5. By obtaining the data parameters of the predicted probability density function in the previous step, the upper and lower limits of the integration interval in the probability density function are determined, and the integral function body based on the adaptive Simpson algorithm is obtained.

[0222] S9.6 Determine the relationship between the set precision value (initial precision value is set to 0.01, which can be adjusted according to the accuracy requirements of computing resources and risk assessment) and the pre-set termination threshold when using the integration method to calculate the area of ​​the function. If the precision value is greater than the threshold, obtain the value of half of the current precision, use this value as the adaptive value of the Simpson algorithm, and calculate the area of ​​the sub-interval of the function obtained in this step. The areas of multiple sub-intervals obtained in this step are combined to obtain the integration result of this step.

[0223] S9.7 Based on all the above-calculated results, determine the specific location and equipment attribute information corresponding to different locations of the equipment represented by each data result. The signal matrices of different locations of the equipment at different times constitute a tensor. Use the tensor to calculate the risk value of different components and locations of the equipment in a certain time series for judgment. Determine the value and the risk threshold. If the risk value is greater than or equal to the risk threshold, it is determined that the corresponding component has potential risk.

[0224] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A control method for false fire alarms in an energy storage system, characterized in that... Includes the following steps: S1. Obtain the real-time data stream of current, voltage and power values ​​transmitted by the sensors of the energy storage system, calculate the fluctuation period of the system operating speed through fast Fourier transform, calculate the second derivative of the fluctuation period, and obtain the probability density function of the data change trend. S2. Based on the probability density function of the data change trend, set the quantiles of the normal operating range under different states. Combined with the pre-set empirical upper limit threshold of the normal operating range, if the probability of the data change trend exceeds the upper limit threshold, an alarm mechanism is triggered. S3. Collect thermal energy changes from multiple time points, obtain the first derivative distribution of thermal energy data with respect to time, obtain the second derivative curve of temperature change data with respect to time, and construct a set of thermal energy change rate index data. S4. Train a support vector machine model based on the data set of thermal energy change rate index, extract the classification boundary between normal state data and abnormal state data, and construct a state recognition mechanism based on the weight ratio of each state. S5. A state recognition mechanism based on support vector machine: when the monitoring data of the energy storage system triggers the upper limit of the preset safe operating range threshold, the difference of historical information records is compared to determine the safety status level and then the corresponding warning mode is triggered. S6. By training a long short-term memory neural network model, using existing data change trend data as features, extracting the time-series features of data change trend data and monitoring data intensity changes, thereby determining the feature differences between the historical state data set and the current state data set, and outputting the normal operating range of each monitoring data of the system. S7. When the monitoring data deviates from the normal operating range after being judged by the long short-term memory neural network, the monitoring data is included in the abnormal database. The intensity of the monitoring data of rare phenomena in the abnormal database is fitted by the triple exponential smoothing method to obtain the probability density distribution of the predicted data change trend of the future time series. S8. Calculate the standard deviation of data over a period of time based on the results of rare phenomenon identification, substitute it into the covariance kernel function of Gaussian process for regression analysis, and thus obtain the functional relationship between the predicted data change trend and the frequency of change of physical quantity. S9. Based on the relationship between the probability density distribution of the predicted data change trend and the frequency of change of physical quantities, a weighted fusion is performed, and the area integral of the function image is performed using the adaptive Simpson algorithm. Based on the size of the area of ​​the integral result, it is determined whether there is a potential security risk.

2. The control method for false fire alarms in an energy storage system according to claim 1, characterized in that: The sub-steps of step S1 are: S1.1 Acquire the current, voltage, and power data streams output by the sensors of the energy storage system; S1.2 Calculate the fast Fourier transform of the three parameter data streams of current value, voltage value and power value to obtain frequency domain component signals, and determine the system fluctuation period based on the correspondence between specific frequency components in the frequency domain component signals and fluctuation period; S1.3 Calculate the second derivative value based on the obtained fluctuation cycle information to obtain the fluctuation state change rate index; S1.

4. Calculate the probability density function based on the numerical value of the rate of change of the fluctuation state, and determine the statistical distribution probability of the system operating in each fluctuation range; S1.

5. Determine the current energy storage state of the energy storage system by using the existing set of relationships between probability density functions and energy storage states; S1.6 Using known voltage and current data, combined with the autoregressive moving average algorithm, predict the voltage and current trends over a period of time. S1.

7. Using the obtained voltage and current trends, input them into the support vector machine to determine the output power value within a preset period after the corresponding energy storage system changes.

3. The control method for false fire alarms in an energy storage system according to claim 1, characterized in that: The sub-steps of step S2 are: S2.1 Collect status data to construct a dataset. The dataset includes historical data under multiple different states. Input the status data into the probability density function model to obtain the probability density function corresponding to the data change trend and obtain the probability value. S2.2 Using the obtained probability density function, based on the pre-set quantile values, the function is solved in reverse from the quantile values ​​to obtain the normal operating range corresponding to each state, and the corresponding upper and lower threshold values ​​are obtained to construct the range interval of different quantile points corresponding to each state. S2.3 Obtain and calculate the state data. Calculate the statistical results of the current data using histogram statistics and use these statistical results to determine the probability of obtaining the current state data. S2.

4. Compare the probability of the current state data with the probability corresponding to different quantiles of the state, and make a threshold judgment; S2.5 If the probability value of the current state is greater than the probability of occurrence corresponding to the preset quantile threshold, an alarm signal is triggered, and the current state data and related time information are recorded for subsequent analysis.

4. The control method for false alarms in a fire protection system according to claim 1, characterized in that: The sub-steps of step S3 are as follows: S3.1 The time-domain sensor records and acquires the thermal energy data stream of the object at multiple moments, calculates the difference in thermal energy values ​​between adjacent moments, and obtains the first derivative of thermal energy with respect to time. S3.

2. Divide the obtained thermal energy difference data between adjacent time points by the time difference to obtain temperature change data; S3.

3. Obtain temperature change data using the difference method to obtain the second derivative curve of temperature with respect to time; S3.4 Statistically count the frequency of the second derivative of temperature exceeding the preset range within a preset time period, and construct an index reflecting the dynamic fluctuation amplitude of thermal energy and the rate of change of thermal energy. S3.

5. Perform time-domain signal analysis on the acquired thermal energy values ​​and calculate the Fourier transform of their frequency components; for discrete thermal energy signal time series... Given the sequence length, its discrete Fourier transform The expression is: ； in, Corresponding to different discrete frequency points, and = 0, 1, 2,……, N - 1; The imaginary unit satisfies , Let π be the mathematical constant; the discrete Fourier transform can be used to obtain the spectral components of the thermal signal at different discrete frequency points; then these spectral components are analyzed to determine the proportion of high-frequency components; if the high-frequency components exceed a predetermined proportion threshold, the next step is executed: Within a set of second-order derivative values ​​corresponding to temperature changes at different times, the Loop Outlier Detection (LOF) algorithm is used to detect and identify moments that deviate from the rest of the data. It is a time series of thermal energy signals. It is the distance between moments; If it is a continuous thermal energy signal time series Its Fourier transform The expression is: Similarly, through the The analysis determines the proportion of high-frequency components, and the subsequent processing is the same as for the discrete case; It is a continuous frequency; S3.6 If an error is detected at any point, proceed to the next step: Thermal energy change data is imported into a trained deep convolutional adversarial neural network model. The model uses an encoder to map the thermal energy features into a hidden space vector, and a discriminator distinguishes the probability that the hidden space vector is real data.

5. The control method for false alarms in a fire protection system according to claim 1, characterized in that: The sub-steps of step S4 are as follows: S4.

1. Based on the time series of collected thermal energy data, obtain a set of thermal energy change rate data; S4.2 Normalize all data in the thermal energy change rate dataset to obtain a normalized dataset, and then set normal state data labels and abnormal state data labels according to preset conditions. S4.3 Obtain labeled normal and abnormal state data sets from the normalized dataset, train the support vector machine model, and obtain the optimal support vector machine classification model; S4.4 Using all the labeled normal and abnormal state data, obtain the classification boundary hyperplane equation of the support vector machine classification model. , Represents the rate of change of thermal energy. It is a weight vector. It is a bias term; S4.

5. By statistically analyzing the occurrence frequency of normal and abnormal states of the labeled data, and obtaining the ratio of the two frequencies, the state weight ratio value is determined. S4.

6. Based on the support vector machine classification model and the state weight ratio, the state of the real-time collected thermal energy change rate data is determined. If the number of thermal energy change rates is greater than the classification boundary value and the number of abnormal thermal energy states is greater than the state weight ratio, then an early warning state is obtained. S4.

7. Collect thermal energy status information corresponding to the thermal energy change rate data in real time, extract related equipment type, environmental information and process parameter data from the database for data fusion, and obtain equipment operation status and maintenance information based on the above data.

6. The control method for false alarms in a fire protection system according to claim 1, characterized in that: The sub-steps of step S5 are as follows: S5.1 The support vector machine obtains real-time energy storage system operation data from the acquisition equipment, performs preprocessing operations, and generates monitoring data. S5.2 When the monitoring data reaches the upper limit of the preset safe operating range threshold, the safety module is triggered, and the operating data is transmitted to the historical database to form an information record. S5.3 Obtain all information records from the historical database as comparison objects, calculate the degree of difference between the comparison objects and the current operating data; S5.

4. Based on the pre-set level rules corresponding to the degree of difference, determine which security level the current monitoring data belongs to and generate a judgment result; S5.

5. Based on the judgment result, obtain one of the various pre-configured warning modes and use the pre-established support vector machine for state recognition; S5.6, State recognition output state signal set, which contains multiple state signals; S5.7 If any status signal exceeds the preset safety range, activate the corresponding warning mode and issue an alarm prompt; S5.8 The system continuously acquires status signals. If it finds that the safety range is constantly out of control, it continuously issues warnings and repeatedly executes support vector machines to perform status recognition operations and update the comparison objects.

7. The control method for false fire alarms in an energy storage system according to claim 1, characterized in that: The sub-steps of step S6 are as follows: S6.

1. Obtain historical data from each type of monitoring point, split the data, and use the first historical dataset as the training set, the second historical dataset as the test set, and the third historical dataset as the validation set. S6.

2. Perform data cleaning, data denoising, data transformation and feature extraction on the training set data. The features include trend, intensity and fluctuation range, and construct feature vectors to obtain the initial training set. S6.

3. Based on the initial training set data, a feature difference extraction model is constructed using the Long Short-Term Memory Neural Network algorithm. The LSTM model is defined with 2 hidden layers, 64 neurons, and a learning rate of 0.

01. The objective function of the model is set as the mean squared error function, and the number of iterations is 500 to obtain the first training model. S6.4 Using validation set data, input the first training model to obtain the first predicted value, and obtain the accuracy by using the historical normal range of the validation set historical data; S6.5 If the model accuracy is lower than the threshold, modify the network structure parameters, reset the learning rate to 0.005 and the number of hidden layers to 3, and update the first training model to the second training model. Return to step S6.4; If the model accuracy reaches the threshold, the first final model is obtained; Input the real-time monitoring data to retrieve the data from the current time. N Minute-by-minute data is used for data preprocessing to obtain real-time trend data and intensity data; Input real-time trend data and intensity data, calculate feature differences through the first final model, compare with normal range values, and determine whether the value of each monitoring point is normal or not.

8. The control method for false fire alarms in an energy storage system according to claim 1, characterized in that: The sub-steps of step S8 are as follows: S8.1 After obtaining the time series of historical data, identify rare events in the time series and obtain the standard deviation of the time series based on this result, and record the standard deviation as the first standard deviation; S8.2 After obtaining the first standard deviation and the time series of the corresponding time interval, perform data preprocessing, obtain the second time series by median filtering, obtain the time of each data point in the second time series and label it as the second time point; S8.

3. Using the second time series and the corresponding second time point, take the data point as the center, and calculate the standard deviation of the three data points before and after it. This standard deviation is then used as the standard deviation of the neighborhood of each data point at the second time point. The standard deviations of all neighborhoods are then combined with the first standard deviation and substituted into the covariance kernel function of the Gaussian process. Among them and This represents data at different points in time within a time series. As the independent variable, As the dependent variable, ， It is the standard deviation. It is the characteristic length of time; S8.

4. Based on the covariance kernel function, a Gaussian process regression model is constructed, and regression analysis is performed on the processed second time series to obtain an initial function model between the time series change and the frequency of physical quantity change. This model is constructed based on the regression analysis and the second time series. S8.

5. Based on the initial function model, the processed time series is segmented. By setting a time length, a third time point and multiple third time series are obtained. The standard deviation and frequency of change of each third time series are calculated to generate a frequency feature vector group. S8.

6. Based on the initial function model, perform predictive analysis, input the elements in the frequency feature vector group into the initial function model respectively, obtain the fourth time series and the corresponding fourth time point, and calculate the difference between the fourth time point series and the third time point series to obtain the frequency error.

9. The control method for false fire alarms in an energy storage system according to claim 1, characterized in that: The sub-steps of step S9 are as follows: S9.

1. Based on multiple related attribute data in the equipment operation time sequence data stream, establish a matrix of signals in the acquisition equipment. A row or column in this matrix is ​​called a signal vector. The signal amplitude value at each acquisition time constitutes a vector component. Perform a discrete Fourier transform on a vector to obtain information data representing the frequency of change of physical quantity. S9.2 Obtain the magnitude data of the vector after discrete Fourier transform and determine the information entropy value of each data according to the Shannon information entropy calculation method. Use the maximum entropy among all data at multiple acquisition times as the measure of risk weight, and then obtain the value of each vector after fusion in each column or row of the matrix. S9.3 Perform principal component analysis and eigenvalue decomposition on the matrix to obtain the importance weights of the main vector groups of the signal represented by different eigenvalues ​​arranged in descending order. After all eigenvalues ​​are weighted proportionally, the weights are accumulated to obtain the normalization factor of the weighted eigenvector group. S9.

4. Collect and obtain the probability density function that represents the probability of the predicted data change trend. The input of the probability density function is the normalization factor of the weight feature vector group obtained in the previous step. After the output of the probability density function is weighted and calculated with the weight, if there is a distribution shift of the probability density function, the gradient descent method is used to iteratively and adaptively correct the parameters in the probability density function in response to the shift phenomenon. S9.

5. By obtaining the data parameters of the predicted probability density function in the previous step, the upper and lower limits of the integration interval in the probability density function are determined, and the integral function body based on the adaptive Simpson algorithm is obtained. S9.6 Determine the relationship between the set precision value and the pre-set termination threshold when using the integration method to calculate the area of ​​the function. If the precision value is greater than the threshold, obtain the value of half of the current precision, use this value as the adaptive value of the Simpson algorithm, and calculate the area of ​​the sub-interval of the function obtained in this step. The areas of multiple sub-intervals obtained in this step are combined to obtain the integration result of this step. S9.7 Based on all the above-calculated results, determine the specific location and equipment attribute information corresponding to different locations of the equipment represented by each data result. The signal matrices of different locations of the equipment at different times constitute a tensor. Use the tensor to calculate the risk value of different components and locations of the equipment in a certain time series for judgment. Determine the value and the risk threshold. If the risk value is greater than or equal to the risk threshold, it is determined that the corresponding component has potential risk.

Images