Oil tank leakage monitoring method and system, electronic equipment and storable medium

Abstract

The invention provides an oil tank leakage monitoring method and system, electronic equipment and a storage medium, and belongs to the field of fire safety and intelligent monitoring. The method comprises the following steps: collecting capacitance and liquid level signals related to a monitored oil tank through an arranged measuring device; the capacitance and liquid level signals are converted into corresponding electric information; training and fusing the electrical information by using a neural network to obtain a fused feature map; enhancing the feature graph based on a residual network so as to expand a feature graph set needing to be monitored; extracting low-order and high-order feature maps from the enhanced feature map set based on a feature extraction network; performing feature fusion on the low-order feature map and the high-order feature map by adopting a feature fusion algorithm to generate a multi-attention feature map; and displaying the multi-attention feature map on a man-machine interaction platform through a related program of MATLAB (Matrix Laboratory). According to the invention, the oil leakage on-line monitoring of the oil tank with high measurement precision and small positioning error of the permeation position can be realized.

Description

Technical Field

[0001] This invention belongs to the technical field of fire safety and intelligent monitoring, specifically relating to a method, system, electronic device and storage medium for monitoring oil tank leakage. Background Technology

[0002] With the increasing use of air transport as people's living standards improve, oil depots have become an indispensable part of long-distance transportation. Underground oil depot tanks, due to long-term burial and exposure to external environmental factors, the characteristics of the oil itself, and the materials used in the tanks, are prone to corrosion, leading to oil leaks. If these leaks are not detected in time, they can pollute surrounding groundwater and soil resources. Therefore, the detection of oil leaks in oil depot tanks is crucial.

[0003] The main methods for detecting oil tank leaks in China include: manual measurement, level gauge method, acoustic method, and radar detection method. Manual measurement is affected by factors such as the standardization of personnel operation, oil temperature, and liquid level fluctuations, resulting in low measurement accuracy and large errors. The level gauge method uses infrared or capacitance sensors to measure changes in the liquid level inside the tank to determine if there is a leak, but its measurement accuracy is low. The acoustic method locates the leak area by detecting changes in the propagation characteristics of elastic waves caused by changes in sound wave energy, but the error in determining the leak location is large. The radar detection method emits high-frequency electromagnetic waves to the leak area and uses the received data to create an image for leak location, but the measurement error for leaks at the bottom of the tank is large, and the error in determining the leak location is also large.

[0004] Therefore, how to achieve online monitoring of oil leakage in oil tanks with high measurement accuracy and small leakage location error is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address at least one of the aforementioned technical problems, this invention provides a method, system, electronic device, and storage medium for monitoring oil tank leakage, which enables online monitoring of oil tank leakage with high measurement accuracy and small location error.

[0006] In a first aspect, the invention provides a method for monitoring oil tank leakage, comprising:

[0007] The measured devices are used to collect capacitance and liquid level signals related to the monitored oil tank.

[0008] The capacitance and liquid level signals are converted into corresponding electrical information;

[0009] The electrical information is trained and fused using a neural network to obtain a fused feature map.

[0010] The feature maps are enhanced using residual networks to expand the feature map set to be monitored.

[0011] Low-order and high-order feature maps are extracted from the enhanced feature map set based on the feature extraction network;

[0012] A feature fusion algorithm is used to fuse the low-order and high-order feature maps to generate a multi-attention feature map;

[0013] The multi-attention feature map is displayed on a human-computer interaction platform using relevant MATLAB programs.

[0014] Preferably, the step of acquiring the capacitance and liquid level signals related to the monitored oil tank through the set measuring device specifically includes:

[0015] Based on the characteristics of the substances around the oil tank being monitored in the deep-buried area, a map showing the location and distribution of the electrode plate sensor and the liquid level sensor was developed.

[0016] The capacitance signal between the electrode plate sensors is obtained by combining the sensitive field formed between the monitored outer wall of the oil tank and the electrode plate sensor, and the electrostatic field formed by the electrode plate sensor in the sensitive field.

[0017] The liquid level signal is obtained by the distributed liquid level sensors in the monitoring tank interlayer.

[0018] Preferably, the step of converting the capacitance and liquid level signals into corresponding electrical information specifically includes:

[0019] The capacitance signal is processed by an AC excitation measurement circuit to obtain voltage information. The AC excitation measurement circuit includes an operational amplifier circuit, an amplifier circuit, a demodulation circuit, and a filter circuit.

[0020] A charge amplification circuit is used to convert liquid level information into electrical quantity information;

[0021] The voltage and electrical information are sampled in a timing manner using a hardware description language.

[0022] Preferably, the step of training a neural network and fusing the electrical information to obtain the fused feature map specifically includes:

[0023] The electrical information is preprocessed to obtain preprocessed data;

[0024] The preprocessed data is input and a BP neural network is trained to output the desired feature vector.

[0025] The desired feature vectors are fused using a weighted mechanism to obtain a fused feature vector, and then the feature map is reconstructed through deconvolution.

[0026] Preferably, the step of enhancing the feature map based on the residual network to expand the feature map set to be monitored specifically includes:

[0027] The feature map is sharpened by calculating the gradient difference;

[0028] The direction of data variance variation is captured by calculating the covariance matrix to determine the sampling direction;

[0029] The feature map is explicitly expanded along the sampling direction to form an expanded feature map set.

[0030] Preferably, the step of extracting low-order and high-order feature maps from the enhanced feature map set based on the feature extraction network specifically includes:

[0031] The input feature map set is segmented using a segmentation module;

[0032] The adjustment module adjusts the number of image channels for the feature map set.

[0033] The extraction module extracts low-order and high-order feature maps from the adjusted feature map set.

[0034] Preferably, the step of using a feature fusion algorithm to fuse the low-order and high-order feature maps to generate a multi-attention feature map specifically includes:

[0035] Upsampling is used to adjust the size of the high-order feature map to be consistent with that of the low-order feature map, and a portion of the low-order feature map is concatenated with the high-order feature map to obtain an initial fused feature map;

[0036] The initial fused feature map size is adjusted to be consistent with the size of the remaining low-order feature maps by upsampling, and the initial fused feature map and the remaining low-order feature maps are then concatenated to obtain the final fused feature map.

[0037] The final fused feature map is processed by a multi-attention mechanism algorithm to generate a multi-attention feature map.

[0038] Secondly, an oil tank leakage monitoring system includes:

[0039] Acquisition module: Used to acquire capacitance and liquid level signals related to the monitored oil tank through the set measuring device;

[0040] The conversion module is used to convert the capacitance and liquid level signals into corresponding electrical information;

[0041] The training module is used to train a neural network and fuse the electrical information to obtain a fused feature map;

[0042] An enhancement module is used to enhance the feature maps based on a residual network to expand the feature map set to be monitored;

[0043] The extraction module is used to extract low-order and high-order feature maps from the enhanced feature map set based on the feature extraction network;

[0044] The fusion module is used to fuse the low-order and high-order feature maps using a feature fusion algorithm to generate a multi-attention feature map;

[0045] The demonstration module is used to display the multi-attention feature map on a human-computer interaction platform using relevant MATLAB programs.

[0046] Preferably, the acquisition module specifically includes:

[0047] The analysis unit is used to develop a map showing the location and distribution of electrode plate sensors and liquid level sensors based on the characteristics of the substances around the oil tank being monitored in the deep burial facility.

[0048] The conversion unit is used to combine the sensitive field formed between the monitored outer wall of the oil tank and the electrode plate sensor, and the electrostatic field formed by the electrode plate sensor in the sensitive field to convert the capacitance signal between the electrode plate sensors.

[0049] The intermittent measurement unit is used to obtain the liquid level signal through the distributed liquid level sensor in the monitoring tank interlayer.

[0050] Preferably, the conversion module specifically includes:

[0051] The excitation unit is used to process the capacitance signal to obtain voltage information using an AC excitation measurement circuit, wherein the AC excitation measurement circuit includes an operational amplifier circuit, an amplifier circuit, a demodulation circuit, and a filter circuit.

[0052] The amplification unit is used to convert liquid level information into electrical quantity information using a charge amplification circuit.

[0053] The timing unit is used to control the timing sampling of the voltage information and the power information through a hardware description language.

[0054] Preferably, the training module specifically includes:

[0055] The preprocessing module is used to preprocess the electrical information to obtain preprocessed data;

[0056] The training unit is used to input the preprocessed data and train the BP neural network to output the desired feature vector;

[0057] The reconstruction unit is used to fuse the desired feature vectors using a weighted mechanism to obtain a fused feature vector, and then reconstructs the feature map through deconvolution operation.

[0058] Preferably, the enhancement module specifically includes:

[0059] A sharpening unit is used to sharpen the feature map by calculating the gradient difference;

[0060] The capture unit is used to capture the direction of change of data variance by calculating the covariance matrix to determine the sampling direction;

[0061] Expansion unit: used to explicitly expand the feature map along the sampling direction to form an expanded feature map set.

[0062] Preferably, the extraction module specifically includes:

[0063] A segmentation unit is used to segment the input feature map set using a segmentation module;

[0064] An adjustment unit is used to adjust the number of image channels of the feature map set through the adjustment module;

[0065] An extraction unit is used to extract low-order and high-order feature maps from the adjusted feature map set based on the extraction module.

[0066] Preferably, the fusion module specifically includes:

[0067] The first stitching unit is used to adjust the size of the high-order feature map to be consistent with the size of the low-order feature map by upsampling, and to stitch a portion of the low-order feature map with the high-order feature map to obtain an initial fused feature map;

[0068] The second splicing unit is used to adjust the size of the initial fused feature map to be consistent with the size of the remaining low-order feature maps by upsampling, and to splice the initial fused feature map with the remaining low-order feature maps to obtain the final fused feature map;

[0069] The pooling unit is used to process the final fused feature map using a multi-attention mechanism algorithm to generate a multi-attention feature map.

[0070] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the oil tank leakage monitoring method as described in the first aspect.

[0071] Fourthly, this application provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the oil tank leakage monitoring method as described in the first aspect.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0073] 1. To address the difficulty of detecting minute leaks in deeply buried oil tanks using existing indirect methods and the inability to pinpoint their location, a dual-channel solution based on distributed electrode sensors and liquid level sensors is adopted to collect data for oil tank leak monitoring, providing data support for monitoring whether oil tanks are leaking, the amount of leakage, and the location of leakage.

[0074] 2. The capacitance signal and liquid level signal are converted into corresponding electrical information by the AC excitation measurement circuit and the charge amplification circuit, respectively, which facilitates the use of subsequent neural network algorithms; and the timing of dual-channel sampling is controlled by the hardware description language, which can meet the requirements of real-time acquisition of electrode and liquid level information and ensure the real-time performance of the acquired data.

[0075] 3. Based on the monitoring principle of combining electrodes and liquid level, a neural network is used to fuse capacitance data and liquid level data, and the data is trained and learned based on the neural network to output a feature map with the desired feature vector, so as to further accurately determine the micro-leakage status of the oil tank wall.

[0076] 4. Use residual networks to extract semantic information from images, estimate the covariance matrix of oil leakage features, and sample from the zero-mean multivariate normal distribution of covariance. Apply it to the features of training samples in this class to expand the dataset and solve the problem of insufficient oil leakage monitoring data.

[0077] 5. Introduce a convolutional pyramid structure into the feature extraction network to expand the receptive field of the convolutional kernel without increasing the computational load, thereby enhancing the information extraction capability of the feature extraction network.

[0078] 6. To avoid the loss of spatial information during convolution, low-order and high-order feature maps are fused. Furthermore, to avoid the loss of semantic information in the channel direction and space of the feature maps during fusion, the fused feature maps are processed through a multi-attention mechanism. This ensures that online monitoring of oil tank leakage with high measurement accuracy and small leakage location error can be achieved. Attached Figure Description

[0079] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0080] Figure 1 This is a flowchart of the oil tank leakage monitoring method provided in Embodiment 1 of the present invention.

[0081] Figure 2This is a circuit diagram of the AC excitation measurement circuit of Embodiment 1 of the present invention.

[0082] Figure 3 This is the residual network structure of Embodiment 1 of the present invention.

[0083] Figure 4 This is a schematic diagram of the improved feature fusion process in Embodiment 1 of the present invention.

[0084] Figure 5 This is a structural block diagram of an oil tank leakage monitoring system corresponding to the method in Embodiment 1, provided in Embodiment 2 of the present invention.

[0085] Figure 6 This is a schematic diagram of the hardware structure of the computer provided in Embodiment 3 of the present invention.

[0086] Explanation of reference numerals in the attached figures:

[0087] 10-Acquisition module, 11-Analysis unit, 12-Conversion unit, 13-Intermittent measurement unit.

[0088] 20 - Conversion module, 21 - Excitation unit, 22 - Amplification unit, 23 - Timing unit.

[0089] 30 - Training module, 31 - Preprocessing module, 32 - Training unit, 33 - Reconstruction unit.

[0090] 40-Enhancement module, 41-Sharpening unit, 42-Capture unit, 43-Expansion unit.

[0091] 50 - Extraction module, 51 - Segmentation unit, 52 - Adjustment unit, 53 - Extraction unit.

[0092] 60 - Fusion module, 61 - First splicing unit, 62 - Second splicing unit, 63 - Pooling unit.

[0093] 70 - Display Module.

[0094] 80 - Bus, 81 - Processor, 82 - Memory, 83 - Communication interface. Detailed Implementation

[0095] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0096] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., may be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this disclosure.

[0097] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0098] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the actual situation.

[0099] Oil tanks are the primary means of storing oil in oil depots. Some oil tanks, due to long-term underground burial and exposure to various factors, are prone to damage, leading to oil and gas leaks. Common methods for detecting oil tank leaks in China include manual measurement, level gauge methods, acoustic methods, and radar detection methods. The advantages and disadvantages of existing leak monitoring methods are shown in the table below. While each of these methods has its own advantages, they also have limitations. They are easily affected by external factors, leading to inaccurate measurement of oil tank leaks, large errors in locating leak positions, and the inability to achieve online leak monitoring. This application is based on these limitations.

[0100]

[0101] Example 1

[0102] Specifically, Figure 1 The diagram shown is a flowchart of the oil tank leakage monitoring method provided in this embodiment.

[0103] like Figure 1 As shown, the oil tank leakage monitoring method of this embodiment includes the following steps:

[0104] S101 collects capacitance and liquid level signals related to the monitored oil tank through a set measuring device.

[0105] Specifically, addressing the problems of insensitivity to minor leaks, low measurement accuracy, poor real-time performance, and inability to locate leaks in oil tank leakage monitoring, this embodiment proposes a solution combining electrode sensors and liquid level sensors to achieve online measurement and location of oil tank leaks, based on existing technologies. Specifically, a dual-channel scheme based on distributed electrode sensors and liquid level sensors is used to collect data for oil tank leakage monitoring, providing data support for monitoring whether leaks have occurred, the amount of leakage, and the location of the leakage. It should be noted that the oil tanks described in this embodiment are all double-layered tanks.

[0106] Furthermore, the specific steps of step S101 include:

[0107] S1011, based on the characteristics of the substances around the oil tank monitored in the deep burial, develop a map showing the location and distribution of the electrode plate sensor and the liquid level sensor.

[0108] Specifically, the monitoring mechanism of this application for the monitored oil tank is based on detecting changes in the response of sensitive areas. That is, when leakage occurs, the content of surrounding substances changes slightly, and the leakage monitoring device acquires the data of changes in sensitive areas. In this embodiment, an array of distributed electrode sensors is designed, and the current single-point liquid level sensor is extended to a distributed liquid level sensor. Combining the two arrangement methods, the function of monitoring and locating minor leaks in buried double-walled oil tanks is realized. In specific implementation, the liquid level sensor is installed in the interlayer between the inner and outer tanks of the double-walled oil tank, and the array electrode sensors are installed directly below the oil tanks. If the liquid level sensor detects a change in liquid level and it is oil, the inner tank may be damaged and leaking; if it detects a change in liquid level and it is water, the outer tank may be leaking.

[0109] S1012, the capacitance signal between the electrode plate sensors is obtained by combining the sensitive field formed between the monitored outer wall of the oil tank and the electrode plate sensor, and the electrostatic field formed by the electrode plate sensor in the sensitive field.

[0110] Specifically, when the array electrodes are excited, a sensitive field is formed in the region between the outer wall of the oil tank and the inner wall of the electrodes, from which the electrical characteristic distribution of the sensitive field can be established. The sensitive field region can be approximated as an electrostatic field model, in which there are no free charges, i.e., the divergence of the electric flux density is always zero. During the measurement process, when the electrodes are excited sequentially, the electrodes form an electrostatic field within the sensitive field. The capacitance value between any two electrode plates is calculated using the following formula: ;

[0112] In the formula, i is the excitation electrode, j is the measurement electrode; C ij Let V be the capacitance between electrode pair ij, V be the voltage difference between electrode pair ij, ε(x,y) be the dielectric constant distribution function in the sensitive field, ψ(x,y) be the surface area of ​​the measuring electrode, and Q(ψ) be the capacitance between electrode pair ij. jWhen an excitation is applied to electrode i, the charge φ on electrode j is measured. i (x,y) is the electric potential distribution function.

[0113] When monitoring leaks in buried oil tanks, excitation signals are applied sequentially to each electrode, and the capacitance values ​​of each electrode pair are measured sequentially. When a leak occurs, the medium within the sensitive field becomes a mixture of the original medium and the leaking oil, causing changes in the parameters of the sensitive field medium, which in turn alters the electric field distribution and the capacitance values ​​of each electrode pair. By using the aforementioned formula to measure the capacitance value between the two electrode plates, data support can be provided for subsequent determinations of whether a leak has occurred in the oil tank, the amount of leakage, and the location of the leak.

[0114] S1013, the liquid level signal is obtained by the distributed liquid level sensor in the monitoring tank interlayer.

[0115] Specifically, the liquid level sensor used in this embodiment utilizes infrared optics for measurement. Based on the different refractive indices and reflectivities of light depending on the presence or absence of a liquid surface, the sensor distinguishes liquid level and surface information using light, which is then converted into an electrical signal output by a signal processing circuit. When the sensor identifies water, leakage may occur in the outer tank; when the sensor identifies oil, leakage may occur in the inner tank. Expanding a single sensor into a distributed liquid level sensor system allows for the acquisition of a large amount of data, resulting in classification characteristics of oil tank leakage.

[0116] S102, convert the capacitor and liquid level signals into corresponding electrical information.

[0117] Specifically, to achieve online monitoring of oil tank leakage, the signals detected by the sensors need to be converted into stable electrical parameter information. This application employs two types of sensors, thus requiring corresponding signal processing modules for conversion. This embodiment uses an AC excitation measurement circuit and a charge amplification circuit to convert the capacitance signal and liquid level signal into corresponding electrical information, facilitating the subsequent use of neural network algorithms. Furthermore, the timing of dual-channel sampling is controlled via a hardware description language, ensuring real-time acquisition of electrode and liquid level information and guaranteeing the real-time performance of the collected data.

[0118] Furthermore, the specific steps of step S102 include:

[0119] S1021 uses an AC excitation measurement circuit to process the capacitance signal to obtain voltage information.

[0120] The AC excitation measurement circuit includes an operational amplifier circuit, an amplifier circuit, a demodulation circuit, and a filter circuit.

[0121] Specifically, such as Figure 2 The diagram shown is of the AC excitation measurement circuit in this embodiment. Based on... Figure 2 The capacitor under test, consisting of the excitation terminal EX and the detection terminal RE, is measured at the capacitor C.x An excitation signal V with frequency w, amplitude A, and phase α is applied to the device. i = Asin(wt+α), through the operational amplifier circuit and the AC amplifier circuit, the amplification factor is A0. According to circuit theory analysis, the output voltage is obtained as follows:

[0122] ;

[0123] Due to the capacitor C under test x Included in V 01 In the signal, therefore for V 01 Demodulation is performed. During demodulation, V is used... 01 Reference signal V at the same frequency ref After multiplying A1sin(wt+β), the output of the demodulation circuit is:

[0124] ;

[0125] When α=β, the signal V 02 After passing through a low-pass filter, we get:

[0126] ;

[0127] Based on the above calculations, a direct correlation between the capacitance C being measured and the AC excitation circuit can be obtained. x DC output V DC .

[0128] S1022 uses a charge amplifier circuit to convert liquid level information into electrical quantity information.

[0129] Specifically, the level sensor detects changes in the liquid level between the inner and outer tank layers. Since the induced electrical signal received by the sensor is relatively weak, a charge amplification circuit is used to amplify the minute signal to ensure it is within the full-scale range of the acquisition system. In this embodiment, the level sensor, based on the different refractive indices of light relative to the liquid level or surface, transmits the detected information optically, then processes it into an electrical signal output. The sensitivity is adjusted by a sliding rheostat, and the signal is amplified by a chip to achieve the photoelectric conversion process.

[0130] S1023, the voltage information and the power information are sampled in a timing manner by controlling the hardware description language.

[0131] Specifically, the FPGA-controlled electrode switching circuit sequentially selects the excitation, detection, and grounding functions. Data from the ADC acquisition module is converted and processed by the FPGA, and the information is transmitted to the host computer at a matching communication rate. Using an FPGA as a microcontroller for digital circuit and logic program design provides good stability and reliability for the logic control of the electrode switching module. By controlling an external ADC and using parallel transmission, the acquisition and processing of data from buried oil tank leakage monitoring can be achieved. The data acquisition circuit in this embodiment mainly consists of an FPGA and an AD chip. The FPGA is controlled by logic timing; an unreasonable clock design can easily cause logic errors and increase debugging difficulty. Verilog HDL is used as the hardware description language to implement the control timing for electrode switching and ADC dual-channel sampling, which can meet the requirement of real-time acquisition of array electrode and liquid level information.

[0132] S103, a neural network is used to train and fuse the electrical information to obtain a fused feature map.

[0133] Specifically, the information fusion mechanism of this application is as follows: Abnormal information generated by the environment to be monitored is converted into relevant information sources. Through bidirectional data preprocessing, an artificial neural network is used for training and learning to identify the relevant patterns of data types, and the relevant information is globally fused. In this embodiment, based on the monitoring principle combining electrodes and liquid level, a BP neural network is used to fuse capacitance data and liquid level data. The data is then trained and learned based on the neural network to output a feature map with a desired feature vector, further accurately determining the state of minor leaks in the tank wall.

[0134] Furthermore, the specific steps of step S103 include:

[0135] S1031, The electrical information is preprocessed to obtain preprocessed data.

[0136] Specifically, the data preprocessing steps in this embodiment are as follows: ①. Data segmentation: Segmenting is performed using a sliding window to divide the long time series into fixed-length segments; ②. Normalization: Data is normalized using Z-score standardization; ③. Data format conversion: Data is converted to a format suitable for neural network input (e.g., tensor).

[0137] S1032, The preprocessed data is input and a BP neural network is trained to output the desired feature vector.

[0138] Specifically, by combining the acquired data with BP neural network learning, the leakage of the inner / outer tank can be more accurately identified, and the leakage area can be determined, thereby timely stopping the leakage of buried oil tanks and reducing pollution. The BP neural network algorithm flow used in this embodiment is as follows: ①. Network initialization settings, selecting the error function, accuracy, and number of training iterations; ②. Obtaining multiple sets of data through the buried oil tank leakage monitoring system, selecting input samples and network output samples; ③. Calculating the input of each neuron in the hidden layer, and the output of the neurons after computation. The functional relationship between the input vector and output vector of the BP neural network structure in this embodiment is as follows:

[0139] ;

[0140] In the formula: y k For the k-th output, w 2 ij Let w be the weights from hidden layer i to output layer j, θ(·) be the activation function, and w be the weights from hidden layer i to output layer j. 1 ij Let b be the weights from input layer i to hidden layer j. j This represents the bias value of neuron j in the hidden layer.

[0141] After initialization, the BP neural network in this embodiment continuously changes the connection weights between networks under the condition of input training set samples. Taking the minimum mean square error as the standard, the partial derivatives are calculated. Under the condition of steepest descent, the weight values ​​of each layer are corrected and the global error is calculated, so that the output of the network continuously approaches the expected output. When the error reaches the required accuracy, the algorithm stops.

[0142] S1033, the desired feature vectors are fused using a weighted mechanism to obtain a fused feature vector, and then the feature map is reconstructed through deconvolution operation.

[0143] Specifically, the weight fusion mechanism in this embodiment is a fast normalization fusion, which uses ReLU to ensure the weights are non-negative and ε to prevent division by zero, avoiding the overhead of Softmax, resulting in fast and stable training. For example, a PyTorch implementation code example:

[0144] Suppose we have two input features f1 and f2

[0145] import torch

[0146] import torch.nn as nn

[0147] epsilon=1e-4

[0148] w=nn.Parameter(torch.ones(2)) # Learnable weights

[0149] relu = nn.ReLU()

[0150] def fast_norm_fusion(f1, f2):

[0151] weight = relu(w)

[0152] weight / =(torch.sum(weight,dim=0)+ epsilon)

[0153] return weight[0]*f1 + weight[1]*f2

[0154] Furthermore, in deep learning frameworks, deconvolution can be implemented using corresponding API functions. For example, in PyTorch, nn.ConvTranspose2d can be used to create a deconvolutional layer, as illustrated in the following code example:

[0155] import torch

[0156] import torch.nn as nn

[0157] class TransposedConvolution(nn.Module):

[0158] def_init__(self, in_channels, out_channels, kernel_size, stride=1,

[0159] padding=0, output_padding=0, bias=True):

[0160] super(TransposedConvolution, self)._init__()

[0161] self.conv_transpose=nn.ConvTranspose2d(in_channels=in_channels,

[0162] out_channels=out_channels,

[0163] kernel_size=kernel_size,stride=stride,

[0164] padding = padding,

[0165] output_padding=output_padding,bias=bias)

[0166] def forward(self, x):

[0167] return self.conv_transpose(x)

[0168] S104, Enhance the feature map based on the residual network to expand the feature map set to be monitored.

[0169] Specifically, addressing the incompleteness of feature image datasets, this paper proposes a dataset construction method based on semantic enhancement. A semantically enhanced data network based on residual networks is constructed to build the detection dataset. This embodiment utilizes residual networks to extract semantic information from images, estimates the covariance matrix of oil leakage features, and samples from a zero-mean multivariate normal distribution of covariance. This sample is then applied to the features of training samples in this class to expand the dataset and solve the problem of insufficient oil leakage monitoring data. The main body of the oil leakage data enhancement network in this embodiment adopts a residual network structure. Residual networks can mine deeper semantic information from images to perform data enhancement, such as… Figure 3 The residual network structure is shown.

[0170] Furthermore, the specific steps of step S104 include:

[0171] S1041, The feature map is sharpened by calculating the gradient difference.

[0172] Specifically, while data augmentation networks can quickly and efficiently augment data, the generated images often have low resolution. Image sharpening primarily enhances image contours, making them more prominent and addressing the resolution issue. This embodiment's image sharpening model calculates gradient differences to identify and sharpen image edge information. Specifically, it uses the Sobel algorithm to calculate pixel gradients. The Sobel algorithm defines two matrices S... x and S y The former is used to calculate the pixel gradient matrix G in the x-direction of the image. x The latter is used to calculate the pixel gradient matrix G in the y-direction of the image. y The gradient calculation formula is as follows:

[0173] ;

[0174] Furthermore, using S x and S y G is obtained by performing convolution calculations on each pixel in the R, G, and B channels of the image. x and G y Then the gradient value G of that pixel can be obtained.t If the gradient is greater than a certain threshold T, then the point (x, y) is considered an edge point. After obtaining the edge points, the edge points are sharpened and enhanced. The gradient sharpening formula is as follows:

[0175] ;

[0176] In the formula, f(x,y) represents the value of the image at that edge point, and S c Indicates the sharpening intensity.

[0177] S1042, the direction of change of data variance is captured by calculating the covariance matrix to determine the sampling direction.

[0178] Specifically, by sharpening the feature map through image sharpening, the data augmentation network can find features of the same type but different shapes, clarifying which directions the network should sample from, and avoiding sampling directions that are meaningless. In this embodiment, instead of randomly and evenly sampling in all directions, the covariance matrix of the data is estimated first to capture the direction of variance variation of this type of data. Then, directions are sampled from a zero-mean multivariate normal distribution with estimated covariance, and these are applied to the features of the training samples in this class to expand the dataset.

[0179] S1043, The feature map is explicitly expanded along the sampling direction to form an expanded feature map set.

[0180] Specifically, let A-dimensional vector a i =[a i1 ,…,a iA ] T In order to obtain enhanced a i The semantic direction, from the zero-mean multivariate normal distribution (0, ∑ yi A random sample vector in ) where ∑ yi It is based on class y i The class conditional covariance matrix of the feature estimates for all samples is calculated. In this embodiment, to achieve semantic data augmentation, each a... i Explicitly expand M times to form the expanded feature set {(a i 1 ,y i ), ..., (a i M ,y i )} N i=1 The size is M×N, where a i k It is sample x i The network is then trained by minimizing the cross-entropy loss and obtaining the k-th augmented matrix of the enhanced features.

[0181] S105, extract low-order and high-order feature maps from the enhanced feature map set based on the feature extraction network.

[0182] Specifically, most current feature extraction networks are designed and built on top of CNNs. Further existing designs use the Swin Transformer as the backbone feature extraction network, enhancing the network's feature extraction capabilities by introducing spatial information fusion. The feature extraction network backbone in this embodiment is also based on the Swin Transformer, introducing a dilated convolutional pyramid structure. On top of this, depthwise separable convolutions are introduced, forming a depthwise separable dilated convolutional pyramid structure with lower computational cost. Under the condition that the kernel size remains unchanged (i.e., the receptive field is fixed), feature map acquisition can be completed with lower computational resources. This expands the receptive field of the convolutional kernels without increasing computational cost, greatly reducing computational resource consumption while enhancing the information extraction capabilities of the feature extraction network.

[0183] Furthermore, the specific steps of step S105 include:

[0184] S1051, The input feature map set is segmented using a segmentation module.

[0185] Specifically, in the feature extraction network of this embodiment, the input feature map is segmented by a segmentation module (such as the Patch Partition layer) into windows of size patch_size × patch_size, and then flattened in the channel direction; for example, if the patch_size is 4, that is, each patch will contain 16 pixels. If the initial input image has three channels: R, G, and B, the flattened value of each patch is 48.

[0186] S1052, the adjustment module adjusts the number of image channels of the feature map set.

[0187] Specifically, in the feature extraction network of this embodiment, the adjustment module includes an Embedding layer and a Linear layer. The Embedding layer maps discrete numbers (such as indices in a vocabulary) to continuous dense vectors, directly transforming the input numbers into vectors of a specified dimension without constructing one-hot encoding. The Linear layer maps the input to the output through a linear transformation; it typically receives one-hot encoded input and transforms it into a low-dimensional dense vector using a weight matrix. Specifically, in this embodiment, the Embedding layer directly processes the numeric indices, saving the overhead of storing and computing one-hot vectors, and is used as the first layer of the model. The Linear layer requires one-hot vectors as input and is suitable for subsequent layers of the model.

[0188] S1053, Based on the extraction module, extract low-order and high-order feature maps from the adjusted feature map set.

[0189] Specifically, in the feature extraction network of this embodiment, the adjusted feature map enters the Swing Transformer Block layer for feature map extraction. Specifically, the Swing Transformer Block layer sets the window to be diagonally spaced, so that the middle window contains some information from the other windows. Through continuous sliding operations, the patches within the window fully exchange information. After each Swing Transformer Block layer, a Patch Merging operation is performed to adjust the image size. Feature extraction is completed after each Swing Transformer Block and Patch Merging operation.

[0190] S106, A feature fusion algorithm is used to fuse the low-order and high-order feature maps to generate a multi-attention feature map.

[0191] Specifically, a multi-attention mechanism combining channel attention and spatial attention is introduced to further enhance the feature extraction capability of deep neural networks and improve the detection accuracy of the leakage detection model. In this embodiment, to avoid the loss of spatial information during convolution, low-order and high-order feature maps are fused; furthermore, to avoid the loss of semantic information in the channel direction and space of the feature maps during fusion, the fused feature maps are processed using a multi-attention mechanism.

[0192] Furthermore, the specific steps of step S106 include:

[0193] S1061, upsampling is used to adjust the size of the high-order feature map to be consistent with the size of the low-order feature map, and a portion of the low-order feature map is spliced ​​with the high-order feature map to obtain an initial fused feature map.

[0194] S1062, the initial fused feature map size is adjusted to be consistent with the remaining low-order feature map size by upsampling, and the initial fused feature map and the remaining low-order feature map are spliced ​​together to obtain the final fused feature map.

[0195] Specifically, the purpose of the above two steps is to achieve the fusion of multiple features. Specifically, such as... Figure 4The improved feature fusion process is shown below: Based on the fusion of low-order feature F1 and high-order feature F3, the fusion of low-order feature F2 and the fused feature map (F1, F3) is added. First, after inputting the high-order feature map F3, it is upsampled to make its size consistent with that of the low-order feature F1. Second, after upsampling, the low-order feature map F1 and the high-order feature map F3 are concatenated. Then, the fused feature map (F1, F3) is upsampled to make its size consistent with that of the low-order feature map F2. Finally, the low-order feature map F2 and the feature map (F1, F3) are concatenated to obtain the final fused feature map.

[0196] S1063, The final fused feature map is processed by a multi-attention mechanism algorithm to generate a multi-attention feature map.

[0197] Specifically, the multi-attention mechanism algorithm used in this embodiment employs a hybrid attention mechanism, which includes channel attention and spatial attention. Figure 4 As shown, the input feature map first passes through a channel attention module: ①. The channel attention module performs global average pooling and global max pooling operations on each channel of the input image feature map, calculates the maximum and average eigenvalues ​​for each channel, and generates two vectors containing the number of channels, representing the global maximum and average features for each channel, respectively; ②. The feature vectors after global max pooling and average pooling are input into a shared fully connected layer for learning the attention weights for each channel; ③. To ensure that the attention weights are between 0 and 1, a sigmoid activation function is applied to generate the channel attention weights, thereby compressing the feature map spatially.

[0198] Furthermore, the spatial attention module takes the channel attention feature map as input. After average pooling and max pooling, the channel attention feature map generates feature maps at different spatial locations. Then, after one convolutional layer, the number of channels is adjusted to 1, and then a multi-attention feature map is generated according to the Sigmoid function.

[0199] S107, The multi-attention feature map is displayed on the human-computer interaction platform using relevant MATLAB programs.

[0200] Specifically, MATLAB, used in this embodiment, is a high-level matrix / array language that includes control statements, functions, data structures, input and output, and object-oriented programming features. Users can synchronize input statements with command execution in the command window, or they can write a larger, more complex application first and then run them together. In the specific implementation, the multi-attention feature map is displayed on a human-computer interaction platform through relevant MATLAB programs, enabling intuitive, real-time, and dynamic online monitoring of oil tank leakage.

[0201] In summary: First, a dual-channel scheme based on distributed electrode sensors and liquid level sensors is used to acquire data for oil tank leakage monitoring. AC correlation circuits convert capacitance and liquid level signals into corresponding electrical information, enabling real-time acquisition of electrode and liquid level information. Second, a neural network is used to fuse capacitance and liquid level data, and the data is trained using the neural network to output a feature map with desired feature vectors. Next, a residual network is used to extract semantic information from the images, expanding the dataset by using features from training samples of oil leakage characteristics, thus addressing the problem of insufficient oil leakage monitoring data. Then, a convolutional pyramid structure is introduced into the feature extraction network to expand the receptive field of the convolutional kernel without increasing computational cost. Finally, feature fusion of low-order and high-order feature maps avoids the loss of spatial information during convolution, and the fused feature map is processed using a multi-attention mechanism to prevent the loss of semantic information in the channel direction and space during fusion, ensuring high measurement accuracy and small leakage location error in online oil tank leakage monitoring.

[0202] Example 2

[0203] This embodiment provides a structural block diagram of a system corresponding to the method described in Embodiment 1. Figure 5 This is a structural block diagram of the oil tank leakage monitoring system according to this embodiment, as follows: Figure 5 As shown, the system includes:

[0204] Acquisition module 10: Used to acquire capacitance and liquid level signals related to the monitored oil tank through a set measuring device.

[0205] The conversion module 20 is used to convert the capacitance and liquid level signals into corresponding electrical information.

[0206] Training module 30 is used to train and fuse the electrical information using a neural network to obtain a fused feature map.

[0207] Enhancement module 40 is used to enhance the feature map based on the residual network to expand the feature map set to be monitored.

[0208] Extraction module 50 is used to extract low-order and high-order feature maps from the enhanced feature map set based on the feature extraction network.

[0209] The fusion module 60 is used to perform feature fusion of the low-order and high-order feature maps using a feature fusion algorithm to generate a multi-attention feature map.

[0210] The display module 70 is used to display the multi-attention feature map on the human-computer interaction platform through relevant MATLAB programs.

[0211] Furthermore, the acquisition module 10 specifically includes:

[0212] Analysis unit 11 is used to formulate the location and distribution map of electrode plate sensor and liquid level sensor based on the characteristics of the substances around the oil tank monitored in the deep burial.

[0213] The conversion unit 12 is used to combine the sensitive field formed between the monitored oil tank outer wall and the electrode plate sensor, and the electrostatic field formed by the electrode plate sensor in the sensitive field to convert the capacitance signal between the electrode plate sensors.

[0214] Intermittent measurement unit 13 is used to obtain liquid level signals through the distributed liquid level sensor of the monitoring tank interlayer.

[0215] Furthermore, the conversion module 20 specifically includes:

[0216] The excitation unit 21 is used to process the capacitance signal to obtain voltage information using an AC excitation measurement circuit, wherein the AC excitation measurement circuit includes an operational amplifier circuit, an amplifier circuit, a demodulation circuit, and a filter circuit.

[0217] Amplification unit 22 is used to convert liquid level information into electrical quantity information using a charge amplification circuit.

[0218] The timing unit 23 is used to control the timing sampling of the voltage information and the power information through a hardware description language.

[0219] Furthermore, the training module 30 specifically includes:

[0220] The preprocessing module 31 is used to perform data preprocessing on the electrical information to obtain preprocessed data.

[0221] Training unit 32 is used to input the preprocessed data and train the BP neural network to output the desired feature vector.

[0222] Reconstruction unit 33 is used to fuse the desired feature vectors using a weighted mechanism to obtain a fused feature vector, and then reconstruct the feature map through deconvolution operation.

[0223] Furthermore, the enhancement module 40 specifically includes:

[0224] The sharpening unit 41 is used to sharpen the feature map by calculating the gradient difference.

[0225] The capture unit 42 is used to capture the direction of change of data variance by calculating the covariance matrix to determine the sampling direction.

[0226] The expansion unit 43 is used to explicitly expand the feature map along the sampling direction to form an expanded feature map set.

[0227] Furthermore, the extraction module 50 specifically includes:

[0228] The segmentation unit 51 is used to segment the input feature map set using the segmentation module.

[0229] The adjustment unit 52 is used to adjust the number of image channels of the feature map set through the adjustment module.

[0230] Extraction unit 53 is used to extract low-order and high-order feature maps from the adjusted feature map set based on the extraction module.

[0231] Furthermore, the fusion module 60 specifically includes:

[0232] The first splicing unit 61 is used to adjust the size of the high-order feature map to be consistent with the size of the low-order feature map by upsampling, and to splice part of the low-order feature map with the high-order feature map to obtain an initial fused feature map.

[0233] The second splicing unit 62 is used to adjust the size of the initial fused feature map to be consistent with the size of the remaining low-order feature map by upsampling, and to splice the initial fused feature map with the remaining low-order feature map to obtain the final fused feature map.

[0234] Pooling unit 63 is used to process the final fused feature map using a multi-attention mechanism algorithm to generate a multi-attention feature map.

[0235] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0236] Example 3

[0237] Combination Figure 1 The described method for monitoring oil tank leaks can be implemented using a computer. Figure 6 This is a schematic diagram of the hardware structure of a computer according to this embodiment.

[0238] The computer may include a processor 81 and a memory 82 storing computer program instructions.

[0239] Specifically, the processor 81 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement this application.

[0240] The memory 82 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include removable or non-removable (or fixed) media. Where appropriate, the memory 82 may be internal or external to a data processing device. In a particular embodiment, the memory 82 is non-volatile memory. In a particular embodiment, the memory 82 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0241] The memory 82 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 81.

[0242] The processor 81 reads and executes the computer program instructions stored in the memory 82 to implement the oil tank leakage monitoring method of Embodiment 1 above.

[0243] In some embodiments, the computer may further include a communication interface 83 and a bus 80. For example, Figure 6 As shown, the processor 81, memory 82, and communication interface 83 are connected through bus 80 and complete communication with each other.

[0244] The communication interface 83 is used to enable communication between the various modules, devices, units, and / or equipment in this application. The communication interface 83 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0245] Bus 80 includes hardware, software, or both, that couples computer components together. Bus 80 includes, but is not limited to, at least one of the following: Data Bus, Address Bus, Control Bus, Expansion Bus, and Local Bus. For example, and not as a limitation, bus 80 may include Accelerated Graphics Port (AGP) or other graphics buses, Extended Industry Standard Architecture (EISA) buses, Front Side Bus (FSB), HyperTransport (HT) interconnects, Industry Standard Architecture (ISA) buses, InfiniBand interconnects, Low Pin Count (LPC) buses, memory buses, MicroChannel Architecture (MCA) buses, Peripheral Component Interconnect (PCI) buses, PCI-Express (PCI-X) buses, Serial Advanced Technology Attachment (SATA) buses, Video Electronics Standards Association (VES) local buses, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 80 may include one or more buses. Although this application describes and illustrates a specific bus, this application considers any suitable bus or interconnection.

[0246] The computer can access the oil tank leakage monitoring system and execute the oil tank leakage monitoring method of Example 1.

[0247] In addition, in conjunction with the oil tank leakage monitoring method in Embodiment 1 above, this application can provide a storage medium for implementation. This storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement the oil tank leakage monitoring method of Embodiment 1 above.

[0248] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0249] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring oil tank leakage, characterized in that, include: The capacitance and liquid level signals of the monitored oil tank are collected using the set measuring device; The capacitance and liquid level signals are converted into corresponding electrical information; The electrical information is trained and fused using a neural network to obtain a fused feature map. The feature maps are enhanced using residual networks to expand the feature map set to be monitored. Low-order and high-order feature maps are extracted from the enhanced feature map set based on the feature extraction network; A feature fusion algorithm is used to fuse the low-order and high-order feature maps to generate a multi-attention feature map; The multi-attention feature map is displayed on a human-computer interaction platform using relevant MATLAB programs.

2. The method for monitoring oil tank leakage according to claim 1, characterized in that, The step of acquiring capacitance and liquid level signals related to the monitored oil tank through the set measuring device specifically includes: Based on the characteristics of the substances around the oil tank being monitored in the deep-buried area, a map showing the location and distribution of the electrode plate sensor and the liquid level sensor was developed. The capacitance signal between the electrode plate sensors is obtained by combining the sensitive field formed between the monitored outer wall of the oil tank and the electrode plate sensor, and the electrostatic field formed by the electrode plate sensor in the sensitive field. The liquid level signal is obtained by the distributed liquid level sensors in the monitoring tank interlayer.

3. The method for monitoring oil tank leakage according to claim 1, characterized in that, The step of converting the capacitance and liquid level signals into corresponding electrical information specifically includes: The capacitance signal is processed by an AC excitation measurement circuit to obtain voltage information. The AC excitation measurement circuit includes an operational amplifier circuit, an amplifier circuit, a demodulation circuit, and a filter circuit. A charge amplification circuit is used to convert liquid level information into electrical quantity information; The voltage and electrical information are sampled in a timing manner using a hardware description language.

4. The method for monitoring oil tank leakage according to claim 1, characterized in that, The steps of training a neural network and fusing the electrical information to obtain the fused feature map specifically include: The electrical information is preprocessed to obtain preprocessed data; The preprocessed data is input and a BP neural network is trained to output the desired feature vector. The desired feature vectors are fused using a weighted mechanism to obtain a fused feature vector, and then the feature map is reconstructed through deconvolution.

5. The method for monitoring oil tank leakage according to claim 1, characterized in that, The step of enhancing the feature map based on the residual network to expand the feature map set to be monitored specifically includes: The feature map is sharpened by calculating the gradient difference; The direction of data variance variation is captured by calculating the covariance matrix to determine the sampling direction; The feature map is explicitly expanded along the sampling direction to form an expanded feature map set.

6. The method for monitoring oil tank leakage according to claim 1, characterized in that, The steps of extracting low-order and high-order feature maps from the enhanced feature map set based on the feature extraction network specifically include: The input feature map set is segmented using a segmentation module; The adjustment module adjusts the number of image channels for the feature map set. The extraction module extracts low-order and high-order feature maps from the adjusted feature map set.

7. The method for monitoring oil tank leakage according to claim 1, characterized in that, The step of using a feature fusion algorithm to fuse the low-order and high-order feature maps to generate a multi-attention feature map specifically includes: Upsampling is used to adjust the size of the high-order feature map to be consistent with that of the low-order feature map, and a portion of the low-order feature map is concatenated with the high-order feature map to obtain an initial fused feature map; The initial fused feature map size is adjusted to be consistent with the size of the remaining low-order feature maps by upsampling, and the initial fused feature map and the remaining low-order feature maps are then concatenated to obtain the final fused feature map. The final fused feature map is processed by a multi-attention mechanism algorithm to generate a multi-attention feature map.

8. An oil tank leakage monitoring system, characterized in that, include: Acquisition module: Used to acquire capacitance and liquid level signals related to the monitored oil tank through the set measuring device; The conversion module is used to convert the capacitance and liquid level signals into corresponding electrical information; The training module is used to train a neural network and fuse the electrical information to obtain a fused feature map; An enhancement module is used to enhance the feature maps based on a residual network to expand the feature map set to be monitored; The extraction module is used to extract low-order and high-order feature maps from the enhanced feature map set based on the feature extraction network; The fusion module is used to fuse the low-order and high-order feature maps using a feature fusion algorithm to generate a multi-attention feature map; The demonstration module is used to display the multi-attention feature map on a human-computer interaction platform using relevant MATLAB programs.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the oil tank leakage monitoring method as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the oil tank leakage monitoring method as described in any one of claims 1 to 7.

Images