Liquefied gas storage tank emergency response method and system

Abstract

The invention discloses a liquefied gas storage tank emergency response method and system. Initial liquefied gas storage tank data are obtained through a sensor array; performing data cleaning on the initial liquid gas storage tank data by using an improved efficiency noise reduction algorithm, establishing a liquid gas storage tank state prediction model, and optimizing hyper-parameters of the liquid gas storage tank state prediction model by using an IWOA improved whale optimization algorithm to obtain a target liquid gas storage tank state prediction model; and inputting the characteristic liquefied gas storage tank data into the target liquefied gas storage tank state prediction model for prediction, judging the risk level of the liquefied gas storage tank according to the liquefied gas state evaluation value, and performing automatic emergency response through the risk level. The accuracy and reliability of liquefied gas storage tank state prediction are obviously improved, and the potential risk of the storage tank can be accurately predicted in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial safety management, and in particular to an emergency response method and system for a liquefied gas storage tank. Background Art

[0002] As critical facilities for storing flammable, explosive, and highly hazardous chemicals, the operational safety of liquefied gas storage tanks is directly linked to the safety of people's lives, property, and social stability. In recent years, with the rapid development of the chemical industry, the scale and number of liquefied gas storage tanks have continued to increase, and safety accidents such as tank leaks and explosions have occurred frequently, causing significant losses to society. Existing liquefied gas storage tank safety monitoring and emergency response methods have many limitations. Traditional emergency response strategies lack precision and automation, rely heavily on manual judgment, and are delayed, making it difficult to implement effective prevention and control measures in the early stages of an accident, leading to the escalation of accidents. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design an emergency response method for a liquefied gas storage tank.

[0004] To achieve the above-mentioned purpose, the technical solution of the present invention is as follows: further, in the above-mentioned liquefied gas storage tank emergency response method, the liquefied gas storage tank emergency response method comprises the following steps:

[0005] Acquire gas concentration data, pressure data, temperature data, and liquid level data of the liquefied gas storage tank through a sensor array, and pre-process the acquired data to obtain initial liquefied gas tank data;

[0006] Using an improved efficiency noise reduction algorithm to perform data cleaning on the initial liquefied gas tank data to obtain characteristic liquefied gas tank data;

[0007] Based on the bidirectional Bi-LSTM neural network to process time series data and the GNN graph neural network to analyze the spatial topological relationship of sensors, a Bi-LSTM-GNN liquefied gas tank status prediction model was established. The hyperparameters of the liquefied gas tank status prediction model were optimized using the improved whale optimization algorithm of IWOA to obtain the target liquefied gas tank status prediction model.

[0008] Inputting the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state evaluation value;

[0009] The risk level of the liquefied gas tank is determined according to the liquefied gas status assessment value, and an automatic emergency response is performed based on the risk level.

[0010] Furthermore, in the above-mentioned liquefied gas storage tank emergency response method, the gas concentration data, pressure data, temperature data, and liquid level data of the liquefied gas storage tank are acquired through the sensor array, and the acquired data are preprocessed to obtain initial liquefied gas tank data, including:

[0011] Acquire sensor data and convert the signals output by different sensors into a unified digital signal format;

[0012] Based on the statistical 3σ principle, outlier detection is performed on the data collected by each sensor, and data points that exceed the mean ±3 times the standard deviation are marked as outliers. The outliers are deleted to obtain the initial liquefied gas tank data.

[0013] Furthermore, in the above-mentioned liquefied gas storage tank emergency response method, the initial liquefied gas tank data is cleaned using an improved efficiency noise reduction algorithm to obtain characteristic liquefied gas tank data, including:

[0014] Obtain the initial liquefied gas tank data and construct a time window of length N with the current data point as the center. The time window includes (N-1) / 2 data points before and after.

[0015] The Pearson correlation coefficient is used to calculate the time correlation between the current data point and other data points in the window to obtain the time correlation weight;

[0016] Obtain the sensor spatial position corresponding to each data point, calculate the spatial correlation based on the spatial position relationship of the sensors, and obtain the spatial correlation weight;

[0017] The temporal correlation weight and the spatial correlation weight are adaptively fused to obtain the target weight of each data point, and the data is subjected to noise reduction processing by weighted averaging to obtain cleaned data.

[0018] Furthermore, in the above-mentioned liquefied gas storage tank emergency response method, the bidirectional Bi-LSTM neural network is used to process time series data, and the GNN graph neural network is used to analyze the spatial topological relationship of sensors to establish a Bi-LSTM-GNN liquefied gas tank status prediction model, including:

[0019] The bidirectional Bi-LSTM neural network consists of two Bi-LSTM layers. The input layer receives the time series of feature liquefied gas tank data. After processing by the Bi-LSTM layer, it outputs the hidden state containing the temporal context information to obtain the temporal features.

[0020] Based on the GNN graph neural network, sensors are regarded as nodes in the graph. The connecting edges between nodes represent the spatial proximity of sensors. The weight of the edge is determined by the distance between sensors. The GNN layer uses the GAT graph attention network to aggregate and update the features of the nodes and extract the spatial correlation features between sensors.

[0021] The time feature and the spatial correlation feature are fused by using a weighted summation method to obtain a fusion feature including the time and space features;

[0022] According to the fusion features, the predicted value of the liquefied gas tank state is output through the fully connected layer and the activation function to obtain the liquefied gas state evaluation value.

[0023] Furthermore, in the above-mentioned liquefied gas storage tank emergency response method, the Whale Optimization Algorithm improved by IWOA is used to optimize the hyperparameters of the liquefied gas tank state prediction model to obtain the target liquefied gas tank state prediction model, including:

[0024] Set the model's hyperparameters, including the number of neurons in the Bi-LSTM layer, the number of convolution kernels in the GNN layer, the learning rate, the training batch size, and the regularization parameter;

[0025] A certain number of whale individuals are randomly generated, each of which represents a set of hyperparameter combinations, and the RMSE root mean square error of the model on the validation set is used as the fitness function;

[0026] In the stage of surrounding the prey, an adaptive weight factor is introduced to dynamically adjust the search step size according to the current number of iterations to improve the local search capability of the algorithm; in the stage of spiral position update, a mutation operation is added;

[0027] Sort whale individuals according to their fitness values, retain excellent individuals and eliminate inferior ones, and generate new individuals through crossover and mutation operations to obtain a new generation of population;

[0028] When the number of iterations reaches the preset maximum value or the fitness value no longer changes significantly, the optimization is stopped and the optimal hyperparameter combination is obtained.

[0029] Furthermore, in the above-mentioned liquefied gas storage tank emergency response method, the step of inputting the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state assessment value includes:

[0030] The characteristic liquefied gas tank data is input into the target liquefied gas tank state prediction model, and the time series data is processed through the Bi-LSTM layer to capture the time dependency of the data;

[0031] The spatial topological relationship of the sensors is analyzed through the GNN layer, spatial correlation features are extracted, and finally the liquefied gas state evaluation value is obtained through the fusion layer and output layer.

[0032] Furthermore, in the above-mentioned liquefied gas storage tank emergency response method, the risk level of the liquefied gas tank is determined according to the liquefied gas state assessment value, and the automatic emergency response is performed according to the risk level, including:

[0033] The risk level is divided into four levels according to the liquefied gas status assessment value, including at least safety level, caution level, warning level and danger level;

[0034] If it is determined to be a dangerous level, a red warning signal will be issued, and the emergency shutdown procedure will be triggered immediately. The air inlet and outlet valves of the storage tank will be closed, and the full-range spray system and ventilation equipment will be started to reduce the temperature and gas concentration in the tank.

[0035] A liquefied gas storage tank emergency response system, comprising the following modules:

[0036] The tank data acquisition module is used to acquire the gas concentration data, pressure data, temperature data and liquid level data of the liquefied gas storage tank through the sensor array, and pre-process the acquired data to obtain the initial liquefied gas tank data;

[0037] a tank data processing module, configured to clean the initial liquefied gas tank data using an improved efficiency noise reduction algorithm to obtain characteristic liquefied gas tank data;

[0038] A prediction model building module is used to process time series data based on a bidirectional Bi-LSTM neural network and analyze the spatial topological relationship of sensors using a GNN graph neural network to establish a Bi-LSTM-GNN liquefied gas tank status prediction model. The hyperparameters of the liquefied gas tank status prediction model are optimized using the IWOA-improved whale optimization algorithm to obtain a target liquefied gas tank status prediction model.

[0039] A tank state assessment module is used to input the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state assessment value;

[0040] The emergency status response module is used to judge the risk level of the liquefied gas tank according to the liquefied gas status evaluation value and perform automatic emergency response based on the risk level.

[0041] Furthermore, in a system for implementing the above-mentioned liquefied gas storage tank emergency response method, the tank data processing module includes the following submodules:

[0042] The acquisition submodule is used to obtain the initial liquefied gas tank data and construct a time window of length N with the current data point as the center. The time window includes (N-1) / 2 data points before and after.

[0043] The time calculation submodule is used to calculate the time correlation between the current data point and other data points in the window using the Pearson correlation coefficient to obtain the time correlation weight;

[0044] The spatial calculation submodule is used to obtain the sensor spatial position corresponding to each data point, calculate the spatial correlation based on the spatial position relationship of the sensors, and obtain the spatial correlation weight;

[0045] The denoising submodule is used to adaptively fuse the temporal correlation weight and the spatial correlation weight to obtain the target weight of each data point, and perform denoising on the data by weighted averaging to obtain cleaned data.

[0046] Furthermore, in a system for implementing the above-mentioned liquefied gas storage tank emergency response method, the tank data processing module includes the following submodules:

[0047] A processing submodule, configured to input the characteristic liquefied gas tank data into a target liquefied gas tank state prediction model, process the time series data through a Bi-LSTM layer, and capture the temporal dependency of the data;

[0048] The analysis submodule is used to analyze the spatial topological relationship of sensors through the GNN layer, extract spatial correlation features, and finally obtain the liquefied gas status evaluation value through the fusion layer and output layer.

[0049] Its beneficial effects are that the gas concentration data, pressure data, temperature data and liquid level data of the liquefied gas storage tank are obtained through the sensor array, and the obtained data are preprocessed to obtain initial liquefied gas tank data; the initial liquefied gas tank data is cleaned using an improved efficiency noise reduction algorithm to obtain characteristic liquefied gas tank data; based on the bidirectional Bi-LSTM neural network to process time series data, and using the GNN graph neural network to analyze the spatial topological relationship of the sensor, a Bi-LSTM-GNN liquefied gas tank state prediction model is established, and the hyperparameters of the liquefied gas tank state prediction model are optimized using the IWOA improved whale optimization algorithm to obtain a target liquefied gas tank state prediction model; the characteristic liquefied gas tank data is input into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state evaluation value; the risk level of the liquefied gas tank is judged according to the liquefied gas state evaluation value, and an automatic emergency response is performed according to the risk level. 1. The accuracy and reliability of the liquefied gas storage tank state prediction are significantly improved. Furthermore, IWOA's improved whale optimization algorithm was used to optimize model hyperparameters, further enhancing the model's performance and generalization capabilities, enabling accurate and proactive prediction of potential tank risks. Second, it achieved intelligent and precise emergency response. Compared to traditional emergency response methods that rely on manual experience, this method is more responsive and enables timely and effective measures to be taken at the earliest stages of an accident, minimizing the probability and severity of an accident and ensuring the safety of both personnel and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0051] Figure 1 This is a schematic diagram of a first embodiment of a liquefied gas storage tank emergency response method according to an embodiment of the present invention;

[0052] Figure 2 Schematic diagram of a second embodiment of a liquefied gas storage tank emergency response method according to an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of a first embodiment of a liquefied gas storage tank emergency response system in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0056] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, a liquefied gas storage tank emergency response method includes the following steps:

[0057] Step 101: Acquire gas concentration data, pressure data, temperature data, and liquid level data of a liquefied gas storage tank through a sensor array, and pre-process the acquired data to obtain initial liquefied gas tank data;

[0058] Specifically, in this embodiment, sensor data is obtained and the signals output by different sensors are converted into a unified digital signal format;

[0059] Based on the statistical 3σ principle, outlier detection is performed on the data collected by each sensor, and data points that exceed the mean ±3 times the standard deviation are marked as outliers. The outliers are deleted to obtain the initial liquefied gas tank data.

[0060] Specifically,

[0061] 1. Sensor Array Deployment

[0062] A sensor array is deployed in a grid-like distribution at key locations on the top, wall, and bottom of the liquefied gas tank. The array consists of the following types of sensors:

[0063] Gas Concentration Sensor: A high-precision infrared gas concentration sensor (NDIR sensor) is used. It has high sensitivity and specificity for the main components of liquefied gas (propane, butane, etc.), and can monitor the gas concentration in the tank environment and inside the tank in real time. Monitoring points are set every 5 meters on the tank roof and wall to ensure that there are no blind spots.

[0064] Pressure Sensor: A high-temperature, high-pressure piezoresistive pressure sensor is installed at the tank's air inlet, outlet, and center, collecting real-time pressure data. The sensor's range is determined by the tank's design pressure, with an accuracy of 0.1% FS.

[0065] Temperature Sensors: Platinum resistance temperature sensors (Pt100) are used, featuring high measurement accuracy and excellent stability. They are evenly spaced around the outside of the tank wall, with a temperature monitoring point set every 10 meters. Sensors are also installed in both the liquid and gas phases within the tank to obtain accurate temperature distribution data.

[0066] Liquid Level Sensor: A radar level sensor operates stably in harsh environments, unaffected by changes in medium density, temperature, and pressure. Installed on the top of the tank, it transmits radar waves vertically downward and calculates the liquid level by measuring the echo time.

[0067] 2. Data Collection

[0068] Each sensor transmits data to the data acquisition system via wired or wireless communication. The data acquisition frequency is set to once per second to meet the needs of real-time monitoring and subsequent data analysis. The acquisition system timestamps the raw data to ensure the time series integrity of the data.

[0069] 3. Preprocessing:

[0070] Format conversion: Converts the output signals (voltage and current) of different sensors into a unified digital signal format for easy subsequent processing. For example, the 4-20mA current signal of a pressure sensor is converted into a digital quantity through an analog-to-digital converter.

[0071] Initial Outlier Detection: Data collected by each sensor is tested for outliers using the statistically based 3σ principle. Data points exceeding ±3 standard deviations from the mean are flagged as suspected outliers. Furthermore, based on the sensor's physical characteristics and the tank's historical operating data, appropriate thresholds are set to exclude or flag data that clearly falls outside the normal range.

[0072] Step 102: Using an improved efficiency noise reduction algorithm to clean the initial liquefied gas tank data to obtain characteristic liquefied gas tank data;

[0073] Specifically, in this embodiment, the initial liquefied gas tank data is obtained, and a time window of length N is constructed with the current data point as the center. The time window includes (N-1) / 2 data points before and after.

[0074] The Pearson correlation coefficient is used to calculate the time correlation between the current data point and other data points in the window to obtain the time correlation weight;

[0075] Obtain the sensor spatial position corresponding to each data point, calculate the spatial correlation based on the spatial position relationship of the sensors, and obtain the spatial correlation weight;

[0076] The temporal correlation weight and the spatial correlation weight are adaptively fused to obtain the target weight of each data point. The data is denoised by weighted averaging to obtain the cleaned data.

[0077] Specifically,

[0078] 1. Improved efficiency noise reduction algorithm:

[0079] Based on the characteristics of liquefied gas tank data, we improve traditional noise reduction algorithms (wavelet denoising and Kalman filtering) and propose an efficient noise reduction algorithm based on adaptive weighting. The core idea of ​​this algorithm is to dynamically adjust the weight of each data point based on the temporal correlation of the data and the spatial correlation between sensors to effectively remove noise interference.

[0080] The specific steps are as follows:

[0081] Construct a data window: With the current data point as the center, construct a time window of length N, which includes (N-1) / 2 data points before and after (N is an odd number).

[0082] Calculate the time correlation weight: Use the Pearson correlation coefficient to calculate the time correlation between the current data point and other data points in the window. The higher the correlation, the greater the weight.

[0083] Calculate spatial correlation weights: Based on the spatial relationships of the sensors, a spatial correlation matrix is ​​constructed between them. For each data point, the spatial correlation weight is calculated by combining the data from the sensor where the point is located with the data from adjacent sensors.

[0084] Adaptive weight fusion: Adaptively fuse the temporal correlation weights and spatial correlation weights to obtain the final weight for each data point. Denoise the data using weighted averaging to obtain cleaned data.

[0085] 2. Feature extraction:

[0086] Perform multi-dimensional feature extraction on the cleaned data to obtain feature data that can reflect the status of the liquefied gas storage tank. Specific features include:

[0087] Time domain features: statistical features such as mean, variance, maximum, minimum, peak-to-peak value, skewness, and kurtosis, which describe the distribution and changes of data in the time domain.

[0088] Frequency domain features: Convert time domain data into frequency domain data through fast Fourier transform (FFT), extract features such as main frequency, energy spectrum density, and power spectrum, and reflect the frequency components and energy distribution of the data.

[0089] Time series features: The autocorrelation function (ACF) and partial autocorrelation function (PACF) are used to extract the time correlation characteristics of the data. At the same time, the trend term (linear trend, quadratic trend) and fluctuation term within the sliding window are calculated to describe the long-term trend and short-term fluctuation of the data.

[0090] Spatial correlation features: Based on the spatial topological relationship of the sensors, the concentration gradient, pressure gradient, temperature gradient, and liquid level gradient between adjacent sensors are calculated to reflect the spatial correlation between sensors and the spatial distribution differences of parameters in the tank.

[0091] Step 103: Process the time series data based on a bidirectional Bi-LSTM neural network, analyze the spatial topological relationship of the sensors using a GNN graph neural network, establish a Bi-LSTM-GNN liquefied gas tank state prediction model, and optimize the hyperparameters of the liquefied gas tank state prediction model using the IWOA-improved whale optimization algorithm to obtain a target liquefied gas tank state prediction model.

[0092] Specifically, the bidirectional Bi-LSTM neural network in this embodiment includes two Bi-LSTM layers. The input layer receives the time series of feature liquefied gas tank data, and after processing by the Bi-LSTM layer, it outputs a hidden state containing time context information to obtain time features.

[0093] Based on the GNN graph neural network, sensors are regarded as nodes in the graph. The connecting edges between nodes represent the spatial proximity of sensors. The weight of the edge is determined by the distance between sensors. The GNN layer uses the GAT graph attention network to aggregate and update the features of the nodes and extract the spatial correlation features between sensors.

[0094] The time feature and the spatial correlation feature are fused by weighted summation to obtain the fusion feature containing the time and space features;

[0095] According to the fusion features, the predicted value of the liquefied gas tank status is output through the fully connected layer and activation function to obtain the liquefied gas status evaluation value.

[0096] Set the model's hyperparameters, including the number of neurons in the Bi-LSTM layer, the number of convolution kernels in the GNN layer, the learning rate, the training batch size, and the regularization parameter;

[0097] A certain number of whale individuals are randomly generated, each of which represents a set of hyperparameter combinations, and the RMSE root mean square error of the model on the validation set is used as the fitness function;

[0098] In the stage of surrounding the prey, an adaptive weight factor is introduced to dynamically adjust the search step size according to the current number of iterations to improve the local search capability of the algorithm; in the stage of spiral position update, a mutation operation is added;

[0099] Sort whale individuals according to their fitness values, retain excellent individuals and eliminate inferior ones, and generate new individuals through crossover and mutation operations to obtain a new generation of population;

[0100] When the number of iterations reaches the preset maximum value or the fitness value no longer changes significantly, the optimization is stopped and the optimal hyperparameter combination is obtained.

[0101] Specifically,

[0102] 1. Bi-LSTM-GNN model architecture:

[0103] Bi-LSTM neural network: This neural network is used to process time series data and capture long-term dependencies. The network consists of two Bi-LSTM layers, with the number of neurons in each layer determined by the length of the time series and the feature dimension of the data. The input layer receives the time series data of the featured liquefied gas tanks. After processing by the Bi-LSTM layer, the output is a hidden state containing temporal context information.

[0104] GNN (Graph Neural Network): Used to analyze the spatial topological relationships of sensors. Sensors are viewed as nodes in a graph. The edges connecting nodes represent the spatial proximity of sensors, and the edge weights are determined by the distance or correlation between sensors. The GNN layer uses a graph convolutional network (GCN) or a graph attention network (GAT) to aggregate and update node features and extract spatial correlation features between sensors.

[0105] Fusion layer: The temporal features output by Bi-LSTM and the spatial features output by GNN are fused. Methods such as splicing and weighted summation can be used to obtain a comprehensive representation that includes temporal and spatial features.

[0106] Output layer: Based on the fused features, the predicted value of the liquefied gas tank status, that is, the liquefied gas status evaluation value, is output through the fully connected layer and activation function (linear function).

[0107] 2. IWOA’s improved whale optimization algorithm:

[0108] To address the problem that the traditional Whale Optimization Algorithm (WOA) is prone to falling into local optimality, an improved WOA (IWOA) is introduced to optimize the hyperparameters of the Bi-LSTM-GNN model. Hyperparameters include the number of neurons in the Bi-LSTM layer, the number of convolution kernels in the GNN layer, the learning rate, the training batch size, and the regularization parameter.

[0109] The optimization process is as follows:

[0110] Initialize the population: Randomly generate a certain number of whale individuals, each of which represents a set of hyperparameter combinations.

[0111] Definition of fitness function: The root mean square error (RMSE) of the model on the validation set is used as the fitness function. The smaller the RMSE, the higher the fitness.

[0112] Predation behavior simulation: This includes three behaviors: encircling the prey, spiral position update, and random search. During the encircling phase, an adaptive weight factor is introduced to dynamically adjust the search step size based on the current iteration count, improving the algorithm's local search capabilities. During the spiral position update phase, mutation operations are added to prevent the algorithm from falling into local optima.

[0113] Population update: Sort whale individuals according to their fitness values, retain excellent individuals, eliminate inferior individuals, and generate new individuals through crossover and mutation operations to form a new generation of population.

[0114] Termination condition: When the number of iterations reaches the preset maximum value or the fitness value no longer changes significantly, the optimization is stopped to obtain the optimal hyperparameter combination for building the target liquefied gas tank state prediction model.

[0115] Step 104: Input the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model to perform prediction and obtain a liquefied gas state evaluation value;

[0116] Specifically, in this embodiment, the characteristic liquefied gas tank data is input into the target liquefied gas tank state prediction model, and the time series data is processed through the Bi-LSTM layer to capture the time dependency of the data;

[0117] The spatial topological relationship of the sensors is analyzed through the GNN layer, spatial correlation features are extracted, and finally the liquefied gas state evaluation value is obtained through the fusion layer and output layer.

[0118] Specifically, the feature-extracted LPG tank data is fed into the target LPG tank status prediction model. The model first processes the time series data using a Bi-LSTM layer to capture temporal dependencies. The GNN layer then analyzes the spatial topology of the sensors and extracts spatial correlation features. Finally, the fusion and output layers generate an LPG status assessment. The assessment range is set between 0 and 100, with higher values ​​indicating a more unstable LPG tank status and a higher risk.

[0119] Step 105: Determine the risk level of the liquefied gas tank based on the liquefied gas status assessment value, and perform automatic emergency response based on the risk level.

[0120] Specifically, in this embodiment, the risk level is divided into four levels according to the liquefied gas state assessment value, including at least a safety level, a caution level, a warning level, and a danger level;

[0121] If it is determined to be a dangerous level, a red warning signal will be issued, and the emergency shutdown procedure will be triggered immediately. The air inlet and outlet valves of the storage tank will be closed, and the full-range spray system and ventilation equipment will be started to reduce the temperature and gas concentration in the tank.

[0122] Specifically,

[0123] 1. Risk level classification:

[0124] According to the liquefied gas status assessment value, the risk level is divided into four levels:

[0125] Safety level (0-25): The tank is in a stable state, all parameters are within the normal range, and there are no safety hazards.

[0126] Attention level (26-50): Some parameters fluctuate abnormally, posing certain safety risks and requiring enhanced monitoring.

[0127] Warning level (51-75): Multiple parameters are outside the normal range, the safety risk is high, and appropriate prevention and control measures must be taken immediately.

[0128] Danger level (76-100): Parameters are seriously abnormal, and there are major safety hazards such as explosion and leakage. An emergency response must be initiated.

[0129] 2. Automatic Emergency Response Strategy

[0130] Safety level: The system operates normally, generates monitoring reports regularly, and records tank status data.

[0131] Attention level: A yellow warning signal is issued, and relevant management personnel are notified via text messages, emails, etc., prompting them to strengthen on-site inspections and check the working status of sensors and tank operations.

[0132] Warning Level: An orange warning signal is issued, automatically initiating cooling and pressure reduction measures for the tank. These measures include activating the spray system to lower the tank temperature and adjusting the air intake and output volumes to stabilize the tank pressure. Furthermore, the data collection frequency is increased to five times per second to track parameter changes in real time.

[0133] Danger level: A red warning signal is issued, immediately triggering the emergency shutdown procedure, closing the air inlet and outlet valves of the storage tank to cut off the gas source; starting the full range of spray systems and ventilation equipment to reduce the temperature and gas concentration in the tank; at the same time, sending an alarm message to the relevant emergency management department and starting the on-site personnel evacuation plan.

[0134] Its beneficial effects are as follows: the gas concentration data, pressure data, temperature data and liquid level data of the liquefied gas storage tank are acquired through the sensor array, the acquired data are preprocessed to obtain the initial liquefied gas tank data; the initial liquefied gas tank data are cleaned using the improved efficiency noise reduction algorithm to obtain the characteristic liquefied gas tank data; the time series data are processed based on the bidirectional Bi-LSTM neural network, and the spatial topological relationship of the sensors is analyzed using the GNN graph neural network to establish a Bi-LSTM-GNN liquefied gas tank state prediction model; the hyperparameters of the liquefied gas tank state prediction model are optimized using the whale optimization algorithm improved by IWOA to obtain the target liquefied gas tank state prediction model; the characteristic liquefied gas tank data are input into the target liquefied gas tank state prediction model for prediction to obtain the liquefied gas state evaluation value; the risk level of the liquefied gas tank is judged according to the liquefied gas state evaluation value, and an automatic emergency response is performed according to the risk level. 1. The accuracy and reliability of the liquefied gas storage tank state prediction are significantly improved. Furthermore, IWOA's improved whale optimization algorithm was used to optimize model hyperparameters, further enhancing the model's performance and generalization capabilities, enabling accurate and proactive prediction of potential tank risks. Second, it achieved intelligent and precise emergency response. Compared to traditional emergency response methods that rely on manual experience, this method is more responsive and enables timely and effective measures to be taken at the earliest stages of an accident, minimizing the probability and severity of an accident and ensuring the safety of both personnel and property.

[0135] See also Figure 2 In a liquefied gas storage tank emergency response method, the initial liquefied gas tank data is cleaned using an improved efficiency noise reduction algorithm to obtain characteristic liquefied gas tank data, including the following steps:

[0136] Step 201: Acquire initial liquefied gas tank data, and construct a time window of length N with the current data point as the center. The time window includes (N-1) / 2 data points before and after.

[0137] Step 202: Calculate the time correlation between the current data point and other data points in the window using the Pearson correlation coefficient to obtain a time correlation weight;

[0138] Step 203: Obtain the sensor spatial position corresponding to each data point, calculate the spatial correlation based on the spatial position relationship of the sensors, and obtain the spatial correlation weight;

[0139] Step 204: Adaptively fuse the time correlation weight and the space correlation weight to obtain the target weight of each data point, and perform noise reduction on the data by weighted averaging to obtain cleaned data.

[0140] See also Figure 3In a liquefied gas storage tank emergency response system, the liquefied gas storage tank emergency response system includes the following modules:

[0141] The tank data acquisition module is used to acquire the gas concentration data, pressure data, temperature data and liquid level data of the liquefied gas storage tank through the sensor array, and pre-process the acquired data to obtain the initial liquefied gas tank data;

[0142] The tank data processing module is used to clean the initial liquefied gas tank data using an improved efficiency noise reduction algorithm to obtain characteristic liquefied gas tank data;

[0143] The prediction model building module is used to process time series data based on a bidirectional Bi-LSTM neural network and analyze the spatial topological relationship of sensors using a GNN graph neural network. This module then establishes a Bi-LSTM-GNN liquefied gas tank status prediction model. This module then uses the improved whale optimization algorithm (IWOA) to optimize the hyperparameters of the liquefied gas tank status prediction model and obtain the target liquefied gas tank status prediction model.

[0144] The tank state assessment module is used to input the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain the liquefied gas state assessment value;

[0145] The emergency status response module is used to judge the risk level of the liquefied gas tank according to the liquefied gas status assessment value and perform automatic emergency response based on the risk level.

[0146] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A liquefied gas storage tank emergency response method, characterized in that: The liquefied gas storage tank emergency response method comprises the following steps: Acquire gas concentration data, pressure data, temperature data, and liquid level data of the liquefied gas storage tank through a sensor array, and pre-process the acquired data to obtain initial liquefied gas tank data; Using an improved efficiency noise reduction algorithm to perform data cleaning on the initial liquefied gas tank data to obtain characteristic liquefied gas tank data; Based on the bidirectional Bi-LSTM neural network to process time series data and the GNN graph neural network to analyze the spatial topological relationship of sensors, a Bi-LSTM-GNN liquefied gas tank status prediction model was established. The hyperparameters of the liquefied gas tank status prediction model were optimized using the improved whale optimization algorithm of IWOA to obtain the target liquefied gas tank status prediction model. Inputting the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state evaluation value; The risk level of the liquefied gas tank is determined according to the liquefied gas status assessment value, and an automatic emergency response is performed based on the risk level.

2. A liquefied gas storage tank emergency response method according to claim 1, characterized in that: The method of acquiring gas concentration data, pressure data, temperature data, and liquid level data of the liquefied gas storage tank through the sensor array and preprocessing the acquired data to obtain initial liquefied gas tank data includes: Acquire sensor data and convert the signals output by different sensors into a unified digital signal format; Based on the statistical 3σ principle, outlier detection is performed on the data collected by each sensor, and data points that exceed the mean ±3 times the standard deviation are marked as outliers. The outliers are deleted to obtain the initial liquefied gas tank data.

3. The liquefied gas storage tank emergency response method according to claim 1, characterized in that: The improved efficiency noise reduction algorithm is used to clean the initial liquefied gas tank data to obtain characteristic liquefied gas tank data, including: Obtain the initial liquefied gas tank data and construct a time window of length N with the current data point as the center. The time window includes (N-1) / 2 data points before and after. The Pearson correlation coefficient is used to calculate the time correlation between the current data point and other data points in the window to obtain the time correlation weight; Obtain the sensor spatial position corresponding to each data point, calculate the spatial correlation based on the spatial position relationship of the sensors, and obtain the spatial correlation weight; The temporal correlation weight and the spatial correlation weight are adaptively fused to obtain the target weight of each data point, and the data is subjected to noise reduction processing by weighted averaging to obtain cleaned data.

4. A liquefied gas storage tank emergency response method according to claim 1, characterized in that: The bidirectional Bi-LSTM neural network is used to process time series data, and the GNN graph neural network is used to analyze the spatial topological relationship of sensors to establish a Bi-LSTM-GNN liquefied gas tank status prediction model, including: The bidirectional Bi-LSTM neural network consists of two Bi-LSTM layers. The input layer receives the time series of feature liquefied gas tank data. After processing by the Bi-LSTM layer, it outputs the hidden state containing the temporal context information to obtain the temporal features. Based on the GNN graph neural network, sensors are regarded as nodes in the graph. The connecting edges between nodes represent the spatial proximity of sensors. The weight of the edge is determined by the distance between sensors. The GNN layer uses the GAT graph attention network to aggregate and update the features of the nodes and extract the spatial correlation features between sensors. The time feature and the spatial correlation feature are fused by weighted summation to obtain a fusion feature including the time and space features; According to the fusion features, the predicted value of the liquefied gas tank state is output through the fully connected layer and the activation function to obtain the liquefied gas state evaluation value.

5. The liquefied gas storage tank emergency response method according to claim 1, characterized in that: The Whale Optimization Algorithm improved by IWOA is used to optimize the hyperparameters of the liquefied gas tank state prediction model to obtain a target liquefied gas tank state prediction model, including: Set the model's hyperparameters, including the number of neurons in the Bi-LSTM layer, the number of convolution kernels in the GNN layer, the learning rate, the training batch size, and the regularization parameter; A certain number of whale individuals are randomly generated, each of which represents a set of hyperparameter combinations, and the RMSE root mean square error of the model on the validation set is used as the fitness function; In the stage of surrounding the prey, an adaptive weight factor is introduced to dynamically adjust the search step size according to the current number of iterations to improve the local search capability of the algorithm; in the stage of spiral position update, a mutation operation is added; Sort whale individuals according to their fitness values, retain excellent individuals and eliminate inferior ones, and generate new individuals through crossover and mutation operations to obtain a new generation of population; When the number of iterations reaches the preset maximum value or the fitness value no longer changes significantly, the optimization is stopped and the optimal hyperparameter combination is obtained.

6. The liquefied gas storage tank emergency response method according to claim 1, characterized in that: The step of inputting the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state evaluation value includes: The characteristic liquefied gas tank data is input into the target liquefied gas tank state prediction model, and the time series data is processed through the Bi-LSTM layer to capture the time dependency of the data; The spatial topological relationship of the sensors is analyzed through the GNN layer, spatial correlation features are extracted, and finally the liquefied gas status evaluation value is obtained through the fusion layer and output layer.

7. The liquefied gas storage tank emergency response method according to claim 1, characterized in that: The risk level of the liquefied gas tank is determined based on the liquefied gas status assessment value, and the automatic emergency response is performed based on the risk level, including: The risk level is divided into four levels according to the liquefied gas status assessment value, including at least safety level, caution level, warning level and danger level; If it is determined to be a dangerous level, a red warning signal will be issued, and the emergency shutdown procedure will be triggered immediately. The air inlet and outlet valves of the storage tank will be closed, and the full-range spray system and ventilation equipment will be started to reduce the temperature and gas concentration in the tank.

8. A liquefied gas storage tank emergency response system, characterized in that: The liquefied gas storage tank emergency response system includes the following modules: The tank data acquisition module is used to acquire the gas concentration data, pressure data, temperature data and liquid level data of the liquefied gas storage tank through the sensor array, and pre-process the acquired data to obtain the initial liquefied gas tank data; a tank data processing module, configured to clean the initial liquefied gas tank data using an improved efficiency noise reduction algorithm to obtain characteristic liquefied gas tank data; A prediction model building module is used to process time series data based on a bidirectional Bi-LSTM neural network and analyze the spatial topological relationship of sensors using a GNN graph neural network to establish a Bi-LSTM-GNN liquefied gas tank status prediction model. The hyperparameters of the liquefied gas tank status prediction model are optimized using the IWOA-improved whale optimization algorithm to obtain a target liquefied gas tank status prediction model. A tank state assessment module is used to input the characteristic liquefied gas tank data into the target liquefied gas tank state prediction model for prediction to obtain a liquefied gas state assessment value; The emergency status response module is used to judge the risk level of the liquefied gas tank according to the liquefied gas status evaluation value and perform automatic emergency response based on the risk level.

9. The liquefied gas storage tank emergency response system according to claim 8, characterized in that: The tank data processing module includes the following submodules: The acquisition submodule is used to obtain the initial liquefied gas tank data and construct a time window of length N with the current data point as the center. The time window includes (N-1) / 2 data points before and after. The time calculation submodule is used to calculate the time correlation between the current data point and other data points in the window using the Pearson correlation coefficient to obtain the time correlation weight; The spatial calculation submodule is used to obtain the sensor spatial position corresponding to each data point, calculate the spatial correlation based on the spatial position relationship of the sensors, and obtain the spatial correlation weight; The denoising submodule is used to adaptively fuse the temporal correlation weight and the spatial correlation weight to obtain the target weight of each data point, and perform denoising on the data by weighted averaging to obtain cleaned data.

10. The liquefied gas storage tank emergency response system according to claim 8, characterized in that: The tank data processing module includes the following submodules: A processing submodule, configured to input the characteristic liquefied gas tank data into a target liquefied gas tank state prediction model, process the time series data through a Bi-LSTM layer, and capture the temporal dependency of the data; The analysis submodule is used to analyze the spatial topological relationship of sensors through the GNN layer, extract spatial correlation features, and finally obtain the liquefied gas status evaluation value through the fusion layer and output layer.

Images