CN114429589B - Hydrogen leakage concentration distribution prediction method and system - Google Patents

Abstract

The invention relates to the technical field of hydrogen safe utilization, in particular to a method and a system for predicting hydrogen leakage concentration distribution, wherein the method comprises the following steps: acquiring a plurality of groups of training data sets; the training data set comprises a training concentration set and a training image set; preprocessing a training image set to obtain a training gray level image set; matching the gray value in the training gray image set with the concentration value in the training concentration set to obtain a training gray concentration mapping set; training a gray level concentration conversion model based on a training gray level concentration mapping set, and training a concentration distribution prediction model based on a training gray level image set; based on the real-time leakage image set, combining the trained concentration distribution prediction model to obtain a prediction gray level image; and obtaining the predicted hydrogen concentration distribution based on the predicted gray level image and the trained gray level concentration conversion model. The invention improves the efficiency and the precision of the concentration distribution prediction.

Description

A kind of hydrogen leakage concentration distribution prediction method and system

technical field

The invention relates to the technical field of hydrogen safety utilization, in particular to a hydrogen leakage concentration distribution prediction method and system.

Background technique

Due to the high efficiency, cleanliness and renewable advantages of hydrogen, the use of hydrogen as energy storage and energy supply has attracted more and more attention from the political, academic and industrial circles. However, hydrogen has the characteristics of being flammable and explosive, having a large diffusion coefficient and easily deteriorating the mechanical properties of materials, and when the volume fraction of hydrogen in the air reaches 4%-75.6%, only 0.017mJ of ignition energy can ignite hydrogen and explode. There are potential leakage and explosion hazards in the process of hydrogen preparation, storage, transportation, filling and use. At the same time, the explosion of hydrogen is a combined chain reaction of deflagration and detonation in the diffusion range, and the resulting flame propagation speed is close to the speed of sound. Therefore, it is of great significance to study the hydrogen leakage and diffusion law and predict the concentration distribution of hydrogen in the air when it leaks.

The existing hydrogen leakage concentration distribution prediction methods are based on the basic governing equations and numerical models of fluid mechanics, and three models are used for numerical simulation research: the low-pressure integral leakage model, the high-pressure under-expanded jet model and the simplified two-region model. However, the calculation and simulation time required by this method is long, and the accuracy is not high.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a method and system for predicting the distribution of hydrogen leakage concentration, the whole process has small calculation amount, high efficiency, more realistic and high accuracy.

For achieving the above object, the present invention provides following scheme:

A hydrogen leakage concentration distribution prediction method, comprising:

Obtain several groups of training data sets; the training data sets include training concentration sets and training image sets;

Preprocessing the training image set to obtain a training grayscale image set;

Matching the grayscale values in the training grayscale image set with the concentration values in the training concentration set to obtain a training grayscale concentration map set;

Perform training on the grayscale concentration conversion model based on the training grayscale concentration mapping set, to obtain the trained grayscale concentration conversion model;

The concentration distribution prediction model is trained based on the training grayscale image set to obtain the trained concentration distribution prediction model;

Obtain a real-time leaked image set, and obtain a predicted grayscale sequence based on the real-time leaked image set and the trained concentration distribution prediction model;

The predicted hydrogen concentration distribution is obtained based on the predicted grayscale sequence and the trained grayscale concentration conversion model.

Preferably, the acquiring several groups of training data sets, including:

Build an experimental bench and arrange M ₁ ×M ₂ concentration sensors, where M ₁ is the number of concentration sensors in the height direction, and M ₂ is the number of concentration sensors in the width direction;

Carry out a hydrogen leakage experiment, use a concentration sensor to record the concentration value of a fixed point of hydrogen leakage to obtain the training concentration set; use a high-speed camera to record a complete image of the hydrogen leakage process to obtain the training image set.

Preferably, the training image set is preprocessed to obtain a training grayscale image set, including:

Perform grayscale processing on the training image set to obtain an initial training grayscale image set;

Performing convolution filtering processing on the initial training grayscale image set to obtain a training filtering image set;

dividing each training filtered image in the training filtered image set to obtain a divided image set;

Sampling the divided image set to obtain the training grayscale image set.

Preferably, the matching of the grayscale values in the training grayscale image set and the concentration values in the training concentration set to obtain a training grayscale concentration map set includes:

Determine the position of each said concentration value in the training filter image set of the training filter image to obtain the initial position data set;

Based on the initial position data set, searching is performed in the training grayscale image set to obtain a mapping relationship; the mapping relationship is that each of the concentration values corresponds to a pixel block;

The training grayscale density map set is obtained based on the grayscale value of each of the pixel blocks, each of the density values and the mapping relationship.

Preferably, the grayscale concentration conversion model is trained based on the training grayscale concentration map set to obtain the trained grayscale concentration conversion model, specifically:

Iteratively trains the grayscale density conversion model based on the training grayscale density map set, and evaluates the grayscale density conversion model based on the error loss value until the error loss value is less than the error loss setting value, and obtains The trained grayscale concentration conversion model; the loss error value includes mean square error and regularization loss.

The present invention also provides a hydrogen leakage concentration distribution prediction system, comprising:

The data acquisition module acquires several groups of training data sets; the training data sets include training concentration sets and training image sets;

a preprocessing module, which preprocesses the training image set to obtain a training grayscale image set;

a matching module, which matches the grayscale values in the training grayscale image set with the concentration values in the training concentration set to obtain a training grayscale concentration map set;

a first training module, which trains a grayscale concentration conversion model based on the training grayscale concentration mapping set to obtain the trained grayscale concentration conversion model;

The second training module, based on the training grayscale image set, trains the concentration distribution prediction model to obtain the trained concentration distribution prediction model;

The grayscale prediction module obtains a real-time leaked image set, and obtains a predicted grayscale sequence based on the real-time leaked image set and the trained concentration distribution prediction model;

The concentration distribution prediction module obtains the predicted hydrogen concentration distribution based on the predicted grayscale sequence and the trained grayscale concentration conversion model.

Preferably, the data acquisition module includes:

Experiment building unit, build an experimental bench, and arrange M ₁ × M ₂ concentration sensors, where M ₁ is the number of concentration sensors in the height direction, and M ₂ is the number of concentration sensors in the width direction;

The data acquisition unit performs the hydrogen leakage experiment, uses the concentration sensor to record the concentration value of the fixed position of the hydrogen leakage, and obtains the training concentration set; uses the high-speed camera to record the complete image of the hydrogen leakage process to obtain the training image set.

Preferably, the preprocessing module includes:

a grayscale processing unit, which performs grayscale processing on the training image set to obtain an initial training grayscale image set;

A filtering unit, which performs convolution filtering on the initial training grayscale image set to obtain a training filtering image set;

a dividing unit, which divides each training filtered image in the training filtered image set to obtain a divided image set;

A sampling unit, for sampling the divided image set to obtain the training grayscale image set.

Preferably, the matching module includes:

a position determining unit, for determining the position of each of the concentration values in the training filter image set in the training filter image set, to obtain an initial position data set;

a mapping unit, searching in the training grayscale image set based on the initial position data set to obtain a mapping relationship; the mapping relationship is that each of the concentration values corresponds to a pixel block;

The matching unit obtains the training grayscale density map set based on the grayscale value of each of the pixel blocks, each of the density values and the mapping relationship.

Preferably, the first training module is specifically:

Iteratively trains the grayscale density conversion model based on the training grayscale density map set, and evaluates the grayscale density conversion model based on the error loss value until the error loss value is less than the error loss setting value, and obtains The trained grayscale concentration conversion model; the loss error value includes mean square error and regularization loss.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

The invention relates to a hydrogen leakage concentration distribution prediction method and system. The method includes: acquiring several sets of training data sets; the training data set includes a training concentration set and a training image set; preprocessing the training image set to obtain a training grayscale image set ; Match the gray values in the training gray image set with the concentration values in the training concentration set to obtain the training gray concentration map set; train the gray concentration conversion model based on the training gray concentration map set, and train the gray concentration conversion model based on the training gray image set. Based on the real-time leaked image set, combined with the trained concentration distribution prediction model, the predicted grayscale image is obtained; based on the predicted grayscale image and the trained grayscale concentration conversion model, the predicted hydrogen concentration distribution is obtained. The present invention improves the efficiency and accuracy of concentration distribution prediction.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings required in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the present invention. In the embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative labor.

Fig. 1 is the flow chart of the hydrogen leakage concentration distribution prediction method of the present invention;

2 is a schematic diagram of the arrangement of the concentration sensor of the present invention;

3 is a schematic diagram of an eight-bit grayscale map of the present invention;

4 is a schematic diagram of a training filter image of the present invention;

5 is a structural diagram of a grayscale concentration conversion model of the present invention;

FIG. 6 is a structural diagram of the hydrogen leakage concentration distribution prediction system of the present invention.

Symbol description: 1-data acquisition module, 2-preprocessing module, 3-matching module, 4-first training module, 5-second training module, 6-grayscale prediction module, 7-concentration distribution prediction module.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The purpose of the present invention is to provide a method and system for predicting the distribution of hydrogen leakage concentration, the whole process has small calculation amount, high efficiency, more realistic and high accuracy.

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

Fig. 1 is a flow chart of the method for predicting the hydrogen leakage concentration distribution of the present invention. As shown in Figure 1, the present invention provides a method for predicting the hydrogen leakage concentration distribution, including:

In step S1, several groups of training data sets are obtained; the training data sets include a training concentration set and a training image set.

Specifically, the step S1 includes:

Step S11, build an experimental bench, and arrange M ₁ ×M ₂ density sensors, where M ₁ is the number of density sensors in the height direction, and M ₂ is the number of density sensors in the width direction. Taking 10×2 as an example, the arrangement of the concentration sensors is shown in FIG. 2 , and each four-pointed star in FIG. 2 represents one of the concentration sensors.

In step S12, a hydrogen leakage experiment is performed, and a concentration sensor is used to record the concentration value of a fixed point of hydrogen leakage to obtain the training concentration set; a high-speed camera is used to record a complete image of the hydrogen leakage process to obtain the training image set. For the sake of safety, the experiment replaced hydrogen with helium, which has a density similar to that of hydrogen but is not flammable and explosive, and used a high-speed camera to record the entire hydrogen process. Sampling images and concentration distributions of hydrogen leakage at time step, 1000 training images of hydrogen leakage of size X × Y and hydrogen concentration distribution data of size (M ₁ × M ₂ ) × 1000 can be obtained, where X is the size of the training image height, Y is the width of the training image.

In step S2, the training image set is preprocessed to obtain a training grayscale image set.

Preferably, the step S2 includes:

Step S21, performing grayscale processing on the training image set to obtain an initial training grayscale image set. The initial training grayscale images in the initial training grayscale image set are eight-bit grayscale images, as shown in FIG. 3 .

Step S22, performing convolution filtering on the initial training grayscale image set to obtain a training filtered image set. The training filtered images in the training filtered image set are shown in FIG. 4 . Preferably, a 3×3 convolution kernel is used for convolution filtering processing. The center of the convolution kernel is the origin (0, 0). The convolution kernel is superimposed by the Gaussian convolution kernel and the mean filter, which can eliminate random errors in the image. The calculation formula of the convolution kernel is as follows:

，

；

,

;

为卷积核，

为高斯卷积核。In the formula: x is the abscissa of the convolution kernel, y is the ordinate of the convolution kernel,

is the convolution kernel,

is the Gaussian convolution kernel.

Step S23: Divide each training filtered image in the training filtered image set to obtain a divided image set.

Since the density value measured by the density sensor is the density of an area, the training filter image needs to be divided so that the area can be represented by a pixel block. The division calculation formula is as follows:

；

；

;

;

为训练滤波图像的宽度，

为浓度传感器检测区域的宽度，

为划分后像素块的宽度，

为训练滤波图像的高度，

为浓度传感器检测区域的高度，

为划分后像素块的高度。where:

is the width of the training filtered image,

is the width of the detection area of the concentration sensor,

is the width of the divided pixel block,

is the height of the training filtered image,

is the height of the detection area of the concentration sensor,

is the height of the divided pixel block.

为高度步长，以

为宽度步长对所述划分图像集中的划分图像进行采样，得到训练灰度图像，所述训练灰度图像的大小为

。Step S24, sampling the divided image set to obtain the training grayscale image set. and with

is the height step, with

Sampling the divided images in the divided image set for the width step to obtain a training grayscale image, and the size of the training grayscale image is

.

Step S3, matching the grayscale values in the training grayscale image set with the concentration values in the training concentration set to obtain a training grayscale concentration map set.

Specifically, the step S3 includes:

Step S31: Determine the position of each of the concentration values in the training filter image set in the training filter image set to obtain an initial position data set.

Specifically, the position of each density value on the training filtered image is determined based on the position of each density sensor on the training filtered image, and the process of determining the location of the density sensor on the training filtered image is as follows:

Taking a square area as the background of hydrogen diffusion, the coordinate system is established with the lower right corner of the square as the origin; with the lower right corner of the training filtered image as the origin of the coordinate system, the position of the concentration sensor on the training filtered image can be obtained according to the following formula:

；

;

为浓度传感器在现实中的高度方向上的坐标值，

为浓度传感器在现实中的宽度方向上的坐标值，

为浓度传感器在训练滤波图像上高度方向的坐标值，

为浓度传感器在训练滤波图像上宽度方向的坐标值，

为正方形的高度，

为正方形的宽度。where:

is the coordinate value of the concentration sensor in the height direction in reality,

is the coordinate value of the density sensor in the width direction in reality,

is the coordinate value of the height direction of the density sensor on the training filtered image,

is the coordinate value of the density sensor in the width direction on the training filtered image,

is the height of the square,

is the width of the square.

Step S32, searching in the training grayscale image set based on the initial position data set to obtain a mapping relationship; the mapping relationship is that each density value corresponds to a pixel block.

Step S33, obtaining the training grayscale density map set based on the grayscale value of each of the pixel blocks, each of the density values, and the mapping relationship.

，N为训练灰度图像集中训练灰度图像的数量。Convert the training grayscale image into a grayscale value sequence, and traverse the training grayscale image set to obtain a grayscale value sequence set; the size of the grayscale value sequence set is

, N is the number of training grayscale images in the training grayscale image set.

进行逐步采样，得到灰度值时间序列集。所述灰度值时间序列集包括

个灰度值时间序列。即将训练灰度图像每个像素块的灰度值按照时间序列进行采样，得到所述灰度值时间序列集。The gray value sequence set is based on the time series according to

Step-by-step sampling is performed to obtain a gray value time series set. The gray value time series set includes

gray value time series. That is to say, the gray value of each pixel block of the training gray image is sampled according to a time series to obtain the gray value time series set.

Sampling the training concentration set according to a time series to obtain a concentration value time series set, the concentration value time series set includes M ₁ ×M ₂ concentration value time series, that is, the concentration values obtained by each concentration sensor are carried out according to the time series. Sampling to obtain the concentration value time series set.

Matching the gray value time series set and the concentration value time series set based on the initial position data set to obtain the training gray concentration map set.

In step S4, the gray-scale density conversion model is trained based on the training gray-scale density map set to obtain the trained gray-scale density conversion model.

Preferably, 80% are selected from the training grayscale concentration map set for training, and the remaining 20% are used for testing, and the grayscale concentration conversion model is built, using the grayscale value as the feature input and the concentration value as the final output , for iterative training. The loss function of the grayscale concentration conversion model is the mean square error, the activation function is ReLU, and a regularization layer is added, and the regularization loss and the mean square error are combined together as the overall error loss value. In the process of back propagation , the size of the entire grayscale density conversion model will be reduced, and all the weight coefficients will be as small as possible, reducing the impact of random errors on the final output. The grayscale density conversion model is shown in FIG. 5 . When the error loss value is less than the error loss setting value and no overfitting occurs, the training ends, and the trained grayscale density conversion model is obtained. In this embodiment, the set value of the error loss is 10%.

The calculation formula of the mean square error is as follows:

；

;

为灰度值对应的真实的浓度值，

为灰度值对应的预测的浓度值。In the formula: MSE is the mean square error, M is the number of gray values input in the current iteration,

is the real density value corresponding to the gray value,

is the predicted density value corresponding to the gray value.

The regularization loss is as follows:

；

;

In the formula: L2 is the regularization loss, Z is the number of weight coefficients, λ is the regularization coefficient, and θ _z is the zth weight coefficient.

The calculation formula of the error loss value is as follows:

；

;

为误差损失值。where:

is the error loss value.

In step S5, the concentration distribution prediction model is trained based on the training grayscale image set to obtain the trained concentration distribution prediction model.

个时间步为长度进行选取，作为一个灰度预测集，用每个时间步的前

步灰度值

来预测第

个时间步的图像灰度值

，所述浓度分布预测模型为长短期记忆网络（Long Short-Term Memory，简称LSTM），所述浓度分布预测模型的激活函数选ReLU,损失函数选均方误差，全连接层添加正则化层，训练过程与所述灰度浓度转化模型的训练过程相同，得到训练好的所述浓度分布预测模型。Preferably, the gray value time series set is

The length of time steps is selected as a grayscale prediction set, using the previous value of each time step

step gray value

to predict the

image grayscale values at time steps

, the concentration distribution prediction model is a long short-term memory network (Long Short-Term Memory, referred to as LSTM), the activation function of the concentration distribution prediction model selects ReLU, the loss function selects the mean square error, and the full connection layer adds a regularization layer, The training process is the same as the training process of the grayscale concentration conversion model, and the trained concentration distribution prediction model is obtained.

In step S6, a real-time leaked image set is acquired, and a predicted grayscale sequence is obtained based on the real-time leaked image set and the trained concentration distribution prediction model.

Using the process of step S2 to preprocess the real-time leaked image set to obtain a real-time gray value time series set;

Based on the real-time gray value time series set and the trained concentration distribution prediction model, an initial predicted gray value sequence is obtained, and the initial predicted gray value sequence is interpolated to obtain the predicted gray value sequence.

In step S7, a predicted hydrogen concentration distribution is obtained based on the predicted grayscale sequence and the trained grayscale concentration conversion model.

FIG. 6 is a structural diagram of the hydrogen leakage concentration distribution prediction system of the present invention. As shown in FIG. 6 , the present invention provides a hydrogen leakage concentration distribution prediction system, including: data acquisition module 1, preprocessing module 2, matching module 3, first training module 4, second training module 5, grayscale prediction Module 6 and Concentration Distribution Prediction Module 7.

The data acquisition module 1 acquires several sets of training data sets; the training data sets include a training concentration set and a training image set.

The preprocessing module 2 preprocesses the training image set to obtain a training grayscale image set;

The matching module 3 matches the grayscale values in the training grayscale image set with the concentration values in the training concentration set to obtain a training grayscale concentration map set.

The first training module 4 trains the grayscale concentration conversion model based on the training grayscale concentration map set to obtain the trained grayscale concentration conversion model.

The second training module 5 trains the concentration distribution prediction model based on the training grayscale image set to obtain the trained concentration distribution prediction model.

The grayscale prediction module 6 obtains a real-time leaked image set, and obtains a predicted grayscale sequence based on the real-time leaked image set and the trained concentration distribution prediction model.

The concentration distribution prediction module 7 obtains the predicted hydrogen concentration distribution based on the predicted grayscale sequence and the trained grayscale concentration conversion model.

As an optional implementation manner, the data acquisition module 1 of the present invention includes: an experiment construction unit and a data acquisition unit.

The experimental building unit is used to build an experimental bench, and M1×M2 concentration sensors are arranged, wherein M1 is the number of concentration sensors in the height direction, and M2 is the number of concentration sensors in the width direction.

The data acquisition unit is used for hydrogen leakage experiments, using a concentration sensor to record the concentration value of a fixed point of hydrogen leakage to obtain the training concentration set; using a high-speed camera to record a complete image of the hydrogen leakage process to obtain the training image set.

As an optional implementation manner, the preprocessing module 2 of the present invention includes: a grayscale processing unit, a filtering unit, a dividing unit and a sampling unit.

The grayscale processing unit performs grayscale processing on the training image set to obtain an initial training grayscale image set.

The filtering unit performs convolution filtering on the initial training grayscale image set to obtain a training filtered image set.

The dividing unit divides each training filtered image in the training filtered image set to obtain a divided image set.

The sampling unit samples the divided image set to obtain the training grayscale image set.

As an optional implementation manner, the matching module 3 of the present invention includes: a position determination unit, a mapping unit and a matching unit.

The position determination unit determines the position of each of the concentration values in the training filter image set in the training filter image set to obtain an initial position data set.

The mapping unit searches the training grayscale image set based on the initial position data set to obtain a mapping relationship; the mapping relationship is that each density value corresponds to a pixel block.

The matching unit obtains the training grayscale density map set based on the grayscale value of each of the pixel blocks, each of the density values and the mapping relationship.

As an optional implementation manner, the first training module 4 of the present invention is specifically:

Iteratively trains the grayscale density conversion model based on the training grayscale density map set, and evaluates the grayscale density conversion model based on the error loss value until the error loss value is smaller than the error loss setting value, and obtains The trained grayscale concentration conversion model; the loss error value includes mean square error and regularization loss.

The invention does not rely on simulation software, and uses the real situation of hydrogen diffusion in the air as the input of the whole system, which is more realistic and has high accuracy.

The present invention can be applied to the prediction of the concentration distribution of various gases diffusing in the air, and is not limited to hydrogen.

The present invention is based on the trained model, and the whole process has small calculation amount and high efficiency.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method.

The principles and implementations of the present invention are described herein using specific examples, and the descriptions of the above embodiments are only used to help understand the method and the core idea of the present invention; There will be changes in the specific implementation and application scope. In conclusion, the contents of this specification should not be construed as limiting the present invention.

Claims (8)

1. A hydrogen leakage concentration distribution prediction method characterized by comprising:

acquiring a plurality of groups of training data sets; the training data set comprises a training concentration set and a training image set;

preprocessing the training image set to obtain a training gray level image set;

matching the gray values in the training gray image set with the concentration values in the training concentration set to obtain a training gray concentration mapping set;

training a gray level concentration conversion model based on the training gray level concentration mapping set to obtain the trained gray level concentration conversion model;

training a concentration distribution prediction model based on the training gray level image set to obtain the trained concentration distribution prediction model;

acquiring a real-time leakage image set, and acquiring a prediction gray sequence based on the real-time leakage image set and the trained concentration distribution prediction model; the method comprises the following specific steps: preprocessing the real-time leakage image set by adopting the process of preprocessing the training image set to obtain a training gray level image set to obtain a real-time gray level time sequence set; obtaining an initial predicted gray value sequence based on the real-time gray value time sequence set and the trained concentration distribution prediction model, and interpolating the initial predicted gray value sequence to obtain the predicted gray value sequence;

obtaining the predicted hydrogen concentration distribution based on the predicted gray level sequence and the trained gray level concentration conversion model;

the preprocessing the training image set to obtain a training gray level image set includes:

carrying out gray processing on the training image set to obtain an initial training gray image set;

performing convolution filtering processing on the initial training gray level image set to obtain a training filtering image set; performing convolution filtering processing by adopting a 3 multiplied by 3 convolution kernel; the center of the convolution kernel is taken as an origin (0, 0), the convolution kernel is overlapped by a Gaussian convolution kernel and the mean filtering, and the convolution kernel is calculated by the following formula:

，

；

in the formula: x is the abscissa of the convolution kernel, y is the ordinate of the convolution kernel,

is a convolution kernel, and is a function of the convolution kernel,

is a gaussian convolution kernel;

dividing each training filtering image in the training filtering image set to obtain a divided image set;

and sampling the divided image set to obtain the training gray level image set.

2. The hydrogen leak concentration distribution prediction method according to claim 1, wherein the acquiring of sets of training data includes:

building an experiment bench and arranging M ₁ ×M ₂ A concentration sensor, wherein M ₁ Number of concentration sensors in height direction, M ₂ The number of concentration sensors in the width direction;

performing a hydrogen leakage experiment, and recording the concentration value of the fixed position point of hydrogen leakage by using a concentration sensor to obtain the training concentration set; and recording a complete image of the hydrogen leakage process by using a high-speed camera to obtain the training image set.

3. The method for predicting hydrogen leakage concentration distribution according to claim 1, wherein the matching gray-scale values in the training gray-scale image set with concentration values in the training concentration set to obtain a training gray-scale concentration mapping set comprises:

determining the position of each density value in a training filter image set to obtain an initial position data set;

searching in the training gray level image set based on the initial position data set to obtain a mapping relation; the mapping relation is that each concentration value corresponds to a pixel block;

and obtaining the training gray scale concentration mapping set based on the gray scale value of each pixel block, each density value and the mapping relation.

4. The method for predicting the distribution of hydrogen leakage concentration according to claim 1, wherein the training of a gray concentration conversion model based on the training gray concentration mapping set is performed to obtain the trained gray concentration conversion model, and specifically includes:

iteratively training the gray scale concentration conversion model based on the training gray scale concentration mapping set, and evaluating the gray scale concentration conversion model based on an error loss value until the error loss value is smaller than an error loss set value to obtain the trained gray scale concentration conversion model; the error penalty values include mean square error and regularization penalty.

5. A hydrogen leakage concentration distribution prediction system characterized by comprising:

the data acquisition module acquires a plurality of groups of training data sets; the training data set comprises a training concentration set and a training image set;

the preprocessing module is used for preprocessing the training image set to obtain a training gray image set;

the matching module is used for matching the gray values in the training gray image set with the concentration values in the training concentration set to obtain a training gray concentration mapping set;

the first training module is used for training a gray concentration conversion model based on the training gray concentration mapping set to obtain the trained gray concentration conversion model;

the second training module is used for training a concentration distribution prediction model based on the training gray level image set to obtain the trained concentration distribution prediction model;

the gray level prediction module is used for acquiring a real-time leakage image set and obtaining a prediction gray level sequence based on the real-time leakage image set and the trained concentration distribution prediction model; the method specifically comprises the following steps: preprocessing the real-time leakage image set based on the matching module to obtain a real-time gray value time sequence set; the gray scale prediction module obtains an initial prediction gray scale value sequence based on the real-time gray scale value time sequence set and the trained concentration distribution prediction model, and interpolates the initial prediction gray scale value sequence to obtain the prediction gray scale value sequence;

the concentration distribution prediction module is used for obtaining the predicted hydrogen concentration distribution based on the predicted gray level sequence and the trained gray level concentration conversion model;

the preprocessing module comprises:

the gray processing unit is used for carrying out gray processing on the training image set to obtain an initial training gray image set;

the filtering unit is used for carrying out convolution filtering processing on the initial training gray level image set to obtain a training filtering image set; performing convolution filtering processing by adopting a 3 multiplied by 3 convolution kernel; the center of the convolution kernel is taken as an origin (0, 0), the convolution kernel is overlapped by a Gaussian convolution kernel and the mean filtering, and the convolution kernel is calculated by the following formula:

，

；

in the formula: x is the abscissa of the convolution kernel, y is the ordinate of the convolution kernel,

is a convolution kernel, and is a function of the convolution kernel,

is a gaussian convolution kernel;

the dividing unit is used for dividing each training filtering image in the training filtering image set to obtain a divided image set;

and the sampling unit is used for sampling the divided image set to obtain the training gray level image set.

6. The hydrogen leak concentration distribution prediction system according to claim 5, characterized in that the data acquisition module includes:

experiment building unit, building experiment bench and arranging M ₁ ×M ₂ A concentration sensor, wherein M ₁ Number of concentration sensors in height, M ₂ The number of density sensors in the width direction;

the data acquisition unit is used for carrying out a hydrogen leakage experiment, and recording the concentration value of the hydrogen leakage at the fixed position point by using a concentration sensor to obtain the training concentration set; and recording a complete image of the hydrogen leakage process by using a high-speed camera to obtain the training image set.

7. The hydrogen leak concentration distribution prediction system according to claim 5, characterized in that the matching module includes:

the position determining unit is used for determining the position of each density value in a training filter image set to train a filter image to obtain an initial position data set;

the mapping unit is used for searching in the training gray level image set based on the initial position data set to obtain a mapping relation; the mapping relation is that each concentration value corresponds to a pixel block;

and the matching unit is used for obtaining the training gray scale concentration mapping set based on the gray scale value of each pixel block, each density value and the mapping relation.

8. The hydrogen leak concentration distribution prediction system according to claim 5, characterized in that the first training module is specifically:

iteratively training the gray scale concentration conversion model based on the training gray scale concentration mapping set, and evaluating the gray scale concentration conversion model based on an error loss value until the error loss value is smaller than an error loss set value to obtain the trained gray scale concentration conversion model; the error loss values include mean square error and regularization loss.

Images