Heterogeneous multimode neural network compensation algorithm for multi-gas detector
By using a heterogeneous multi-mode neural network compensation algorithm to collect and process gas and environmental parameter signals in real time, and constructing a convolutional neural network, the problems of environmental factors and gas interference in multi-gas detection are solved, thereby improving detection accuracy and real-time performance.
Patent Information
- Application Number
- CN202511768480.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-02-27
AI Technical Summary
Existing compensation algorithms cannot accurately compensate for the collected data, resulting in reduced data detection accuracy, especially in multi-gas detection environments where environmental factors such as temperature, humidity, and pressure, as well as mutual interference between gases, are severe.
A heterogeneous multimodal neural network compensation algorithm is adopted. By collecting gas and environmental parameter signals in real time, initial compensation and binarization processing are performed to construct a convolutional neural network. Nonlinear transformation is set between the convolutional layer and the pooling layer to train the convolutional kernel and eliminate environmental factors and gas interference.
It improves the accuracy of gas detection, enhances the anti-interference capability of the all-in-one sensor, achieves effective compensation of gas data, and improves detection accuracy and real-time performance.
Smart Images
Figure CN121583379A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of gas detection, in particular to a heterogeneous multi-modal neural network compensation algorithm for a multi-gas detector. BACKGROUND
[0002] In the process of coal production, a variety of toxic and harmful gases are generated, which seriously affect the safety production of underground workers. Therefore, industry technical personnel have designed different gas detection equipment, for example, a single gas detector can detect oxygen, carbon monoxide, hydrogen sulfide, methane and other gases. The disadvantage is that in a mixed gas working environment, only a single gas can be detected, and workers need to carry more gas detection equipment to meet the early warning of toxic and harmful gases in the underground working environment. Therefore, technical personnel have designed multi-in-one gas detectors, for example, a four-in-one gas detector, which has the advantage of being able to detect different gases in the same environment simultaneously.
[0003] Currently, gas detection equipment mainly uses two-point straight line calibration method or three-point broken line calibration method for compensation, which is not suitable for non-linear sensor detection equipment. In addition, when using multi-in-one sensors to detect multiple gases, there are not only environmental factors such as temperature, humidity and pressure that interfere with gas detection, but also mutual interference between multiple gases, which leads to the fact that existing compensation algorithms cannot accurately compensate for collected data, reducing data detection accuracy. SUMMARY
[0004] In order to solve the technical problem that the existing compensation algorithm cannot accurately compensate for the collected data, reducing the data detection accuracy, the purpose of the present application is to provide a heterogeneous multi-modal neural network compensation algorithm for a multi-gas detector, the technical solution adopted is as follows: The present application proposes a heterogeneous multi-modal neural network compensation algorithm for a multi-gas detector, the algorithm comprising: Real-time acquisition of concentration signals of different gases and environmental signals of different environmental parameters using a multi-in-one gas sensor; Adjusting the amplitude of the concentration signal of each gas according to the amplitude of each environmental signal at the current time point to obtain the initial compensation amplitude of the concentration signal of each gas at the current time point; binaryzation processing of the initial compensation amplitude of each concentration signal and the amplitude of each environmental signal to obtain the binary value of each gas and each environmental parameter at the current time point; Training the constructed convolutional neural network to obtain a trained convolutional neural network, the input of the convolutional neural network being the binary value of each gas and each environmental parameter at the time point, and the output of the convolutional neural network being the compensated concentration value of each gas; The binary values of each gas and each environmental parameter are input to the input layer nodes of the trained convolutional neural network in time sequence, and the compensated concentration value of each gas is output.
[0005] Further, the environmental parameters at least include temperature parameter, humidity parameter and air pressure parameter.
[0006] Further, the initial compensation amplitude value of the concentration signal of each gas at the current time point comprises: Taking any gas as a target gas, when the amplitude value of the environmental signal of the temperature parameter at the current time point belongs to a preset first temperature range, the amplitude value of the concentration signal of the target gas at the current time point is adjusted in a negative direction to obtain a temperature compensation amplitude value of the concentration signal of the target gas at the current time point; when the amplitude value of the environmental signal of the temperature parameter at the current time point belongs to a preset second temperature range, the amplitude value of the concentration signal of the target gas at the current time point is not adjusted, and the amplitude value of the concentration signal of the target gas is taken as the temperature compensation amplitude value of the concentration signal of the target gas at the current time point; when the amplitude value of the environmental signal of the temperature parameter at the current time point belongs to a preset third temperature range, the amplitude value of the concentration signal of the target gas at the current time point is adjusted in a positive direction to obtain a temperature compensation amplitude value of the concentration signal of the target gas at the current time point. According to the amplitude value of the environmental signal of the humidity parameter at the current time point, the temperature compensation amplitude value of the concentration signal of the target gas at the current time point is adjusted in a negative direction to obtain a humidity compensation amplitude value of the concentration signal of the target gas at the current time point. According to the amplitude value of the environmental signal of the air pressure parameter at the current time point, the humidity compensation amplitude value of the concentration signal of the target gas at the current time point is adjusted in a positive direction to obtain an initial compensation amplitude value of the concentration signal of the target gas at the current time point.
[0007] Further, the binary values of each gas and each environmental parameter at the current time point comprise: For any gas, if the initial compensation amplitude value of the concentration signal of the gas at the current time point exceeds the standard safety threshold value corresponding to the gas, the binary value of the gas at the current time point is set to the value 1, otherwise the binary value of the gas at the current time point is set to the value 0. For any environmental parameter, if the amplitude value of the environmental signal of the environmental parameter at the current time point belongs to the preset effective range corresponding to the environmental parameter, the binary value of the environmental parameter at the current time point is set to the value 1, otherwise the binary value of the environmental parameter at the current time point is set to the value 0.
[0008] Further, the convolutional neural network comprises an input layer, a convolutional layer, a pooling layer and an output layer, and does not comprise a full connection layer, the pooling layer adopts a direct connection fusion method, a nonlinear transformation is arranged between the convolutional layer and the pooling layer, and the number of nodes of each layer is equal to the sum of the number of all gases and the number of all environmental parameters.
[0009] Further, the nonlinear transformation arranged between the convolutional layer and the pooling layer of the convolutional neural network is realized by a ReLU activation function.
[0010] Further, the trained convolutional neural network comprises: an arbitrary node of an input layer of the convolutional neural network is taken as a target input node, and an arbitrary node of a convolutional layer of the convolutional neural network is taken as a target convolutional node; a zero voltage value of a sensor corresponding to the target input node is used to simplify a convolutional calculation formula from the target input node to the target convolutional node, to obtain a simplified formula of the convolutional calculation formula from the target input node to the target convolutional node; the simplified formula of the convolutional calculation formula is:
[0011] wherein, represents a response function from the target input node to the target convolutional node; represents a zero voltage value of a sensor corresponding to the target input node; represents a weight value in a convolutional kernel from the target input node to the target convolutional node; represents a time point.
[0012] the simplified formula of the convolutional calculation formula from all nodes of the input layer to the target convolutional node is added, and a sum value of the zero voltage values of the sensors corresponding to all input nodes is equal to a numerical value 0, to obtain a convolutional kernel of the target convolutional node; sample data collected by the multi-in-one gas sensor is used to train the convolutional kernel of the convolutional neural network, to obtain a trained convolutional neural network.
[0013] Further, the convolutional calculation formula from the target input node to the target convolutional node is:
[0014] wherein, represents a response function from the target input node to the target convolutional node; represents an excitation function in the target input node; represents an impact function from the target input node to the target convolutional node; represents a time point.
[0015] Furthermore, the step of training the convolutional kernel of the convolutional neural network using sample data collected by the multi-in-one gas sensor to obtain the trained convolutional neural network includes: The concentration sample signals of different gases and environmental sample signals of different environmental parameters collected by the multi-in-one gas sensor are binarized and then input into the corresponding nodes of the input layer of the convolutional neural network. The loss between the output value and the true value of the convolutional neural network is calculated, and the convolution kernel of each node of the convolutional layer of the convolutional neural network is optimized using the gradient descent method to obtain the trained convolutional neural network.
[0016] Furthermore, after outputting the compensation concentration value for each gas, the compensation concentration value also needs to be transmitted to the monitoring center via wired or wireless means.
[0017] The present invention has the following beneficial effects: This invention addresses the issue of environmental interference and inter-gas interference, which hinders existing compensation algorithms from accurately processing collected data and reducing detection precision. Therefore, this invention first performs preliminary compensation on the concentration signal amplitude of each gas based on the amplitude of the environmental signal at the current time point. The initial compensated amplitude of the concentration signal and the amplitude of the environmental signal are then binarized to provide a relatively clean and simplified data input for the subsequent neural network, effectively improving its computational efficiency and enhancing the anti-interference capability of the multi-sensor. Finally, a constructed convolutional neural network further eliminates environmental interference and inter-gas interference. This invention proposes… A direct connection fusion method for pooling layers in convolutional neural networks suitable for this system is proposed. This method reduces the impact of independent influencing factors on the graphical complexity of the neural network and avoids information loss during dimensionality reduction. A nonlinear transformation is set between the convolutional and pooling layers to effectively remove interference from environmental factors and mutual interference between gases. At the same time, the convolutional kernel of the convolutional neural network is constructed and trained to obtain a convolutional neural network that can effectively compensate for environmental interference and mutual interference between gases. Then, the binary values of each gas and each environmental parameter are input to the input layer nodes of the trained convolutional neural network in chronological order, and the compensated concentration value of each gas is output, so that the detection data is effectively compensated and the detection accuracy of gas data is improved. Attached Figure Description
[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, 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.
[0019] Figure 1 The flowchart illustrates a heterogeneous multimodal neural network compensation algorithm for a multi-gas detector, as provided in one embodiment of the present invention. Detailed Implementation
[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a heterogeneous multimodal neural network compensation algorithm for a multi-gas detector proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0022] The following description, in conjunction with the accompanying drawings, details a specific scheme for a heterogeneous multimodal neural network compensation algorithm for a multi-gas detector provided by the present invention.
[0023] Please see Figure 1 This illustrates a heterogeneous multimodal neural network compensation algorithm for a multi-gas detector provided by an embodiment of the present invention. The algorithm includes: Step S1: Use an all-in-one gas sensor to collect concentration signals of different gases and environmental signals of different environmental parameters in real time.
[0024] This invention first utilizes a multi-in-one sensor to collect concentration signals of different gases in real time. In the process of safe production in coal mines, the detection of oxygen, carbon monoxide, hydrogen sulfide, and methane is crucial. Therefore, in one embodiment of this invention, a four-in-one sensor is specifically used, and the gases detected are mainly oxygen, carbon monoxide, hydrogen sulfide, and methane. In other embodiments of this invention, the types of gases can be expanded according to the specific implementation scenario, and are not limited here.
[0025] Multi-functional sensors typically have built-in environmental detection sensing modules. Therefore, this embodiment of the invention also requires the use of a multi-functional gas sensor to collect environmental signals of different environmental parameters in real time. Generally speaking, the environmental factors that have a significant impact on the gas detection process are temperature, humidity, and air pressure. Therefore, the environmental parameters detected in this embodiment of the invention include at least temperature, humidity, and air pressure parameters. In other embodiments of the invention, the environmental parameters can be expanded according to the specific implementation scenario, which is not limited here.
[0026] Step S2: Adjust the amplitude of the concentration signal of each gas according to the amplitude of each environmental signal at the current time point to obtain the initial compensation amplitude of the concentration signal of each gas at the current time point; perform binarization processing on the initial compensation amplitude of each concentration signal and the amplitude of each environmental signal to obtain the binary value of each gas and each environmental parameter at the current time point.
[0027] During gas concentration detection using an all-in-one gas sensor, interference from environmental factors such as temperature, humidity, and air pressure, as well as mutual interference between gases, can lead to a decrease in the sensor's detection accuracy. Compensation is needed to correct this. Therefore, this embodiment of the invention first adjusts the amplitude of the concentration signal for each gas based on the amplitude of the environmental signal for each environmental parameter at the current time point. This obtains the initial compensation amplitude of the concentration signal for each gas at the current time point, thereby achieving preliminary compensation for environmental interference. This provides relatively clean data input for the subsequent neural network and enhances the anti-interference capability of the all-in-one sensor.
[0028] Preferably, in one embodiment of the present invention, the method for obtaining the initial compensation amplitude of the concentration signal of each gas at the current time point specifically includes: In this embodiment of the invention, compensation is performed sequentially based on temperature, humidity, and air pressure. Initial temperature compensation is performed first, using any gas as the target gas. The method of adjusting the amplitude of the target gas concentration signal at the current time point varies depending on the current temperature. When the amplitude of the ambient temperature signal at the current time point falls within a preset first temperature range, the amplitude of the concentration signal needs to be appropriately reduced. Therefore, the amplitude of the target gas concentration signal at the current time point is negatively adjusted to obtain the temperature-compensated amplitude of the target gas concentration signal at the current time point. When the amplitude of the ambient temperature signal at the current time point falls within a preset second temperature range, no adjustment is made to the amplitude of the target gas concentration signal at the current time point. The amplitude of the target gas concentration signal is adjusted, and the amplitude of the target gas concentration signal is used as the temperature compensation amplitude of the target gas concentration signal at the current time point. When the amplitude of the ambient temperature signal at the current time point falls within the preset third temperature range, the amplitude of the concentration signal needs to be appropriately increased. Therefore, the amplitude of the target gas concentration signal at the current time point is positively adjusted to obtain the temperature compensation amplitude of the target gas concentration signal at the current time point. The temperature values included in the preset first temperature range, preset second temperature range, and preset third temperature range increase sequentially. Generally speaking, for coal mine production processes, the operating ambient temperature of the detection equipment is usually between -10 and 55 degrees Celsius. In one embodiment of the present invention, the preset first temperature range is set to... The preset second temperature range is set to The preset third temperature range is set to The preset first temperature range, preset second temperature range, and preset third temperature range can also be set by the implementer according to the specific implementation scenario, and are not limited here.
[0029] As an example, in one embodiment of the present invention, the expression for the temperature compensation amplitude of the target gas concentration signal at the current time point can be specifically as follows:
[0030] in, This represents the temperature compensation amplitude of the target gas concentration signal at the current time point; This indicates the amplitude of the target gas concentration signal at the current time point; The amplitude of the environmental signal representing the temperature parameter at the current time point; This indicates a standard temperature value, which is 25 degrees Celsius. This indicates the preset first adjustment parameter, whose value range is: In one embodiment of the present invention, the following is used: Set to 0.02, It can also be set by the implementer according to the specific implementation scenario, and there are no restrictions here.
[0031] Then, preliminary humidity compensation is performed. Higher humidity will cause the amplitude of the concentration signal to be higher. Therefore, based on the amplitude of the environmental signal of the humidity parameter at the current time point, the temperature compensation amplitude of the target gas concentration signal at the current time point can be negatively adjusted to obtain the humidity compensation amplitude of the target gas concentration signal at the current time point.
[0032] As an example, in one embodiment of the present invention, the expression for the humidity compensation amplitude of the target gas concentration signal at the current time point can be specifically as follows:
[0033] in, This indicates the humidity compensation amplitude of the target gas concentration signal at the current time point; This represents the temperature compensation amplitude of the target gas concentration signal at the current time point; The amplitude of the environmental signal representing the humidity parameter at the current point in time; This indicates the preset second adjustment parameter, whose value range is: In one embodiment of the present invention, the following is used: Set to 0.01, It can also be set by the implementer according to the specific implementation scenario, and there are no restrictions here.
[0034] Then, the air pressure is compensated. Based on the amplitude of the environmental signal of the air pressure parameter at the current time point, the humidity compensation amplitude of the target gas concentration signal at the current time point is positively adjusted to obtain the initial compensation amplitude of the target gas concentration signal at the current time point. Here, air pressure compensation is performed in a nonlinear parabolic manner.
[0035] As an example, in one embodiment of the present invention, the expression for the initial compensation amplitude of the target gas concentration signal at the current time point can be specifically as follows:
[0036] in, This represents the initial compensation amplitude of the target gas concentration signal at the current time point; This indicates the humidity compensation amplitude of the target gas concentration signal at the current time point; This represents the amplitude of the environmental signal indicating the air pressure parameter at the current time point; This represents standard atmospheric pressure, with a value of 101 kPa. This indicates the preset third adjustment parameter, whose value range is: In one embodiment of the present invention, the following is used: Set to 0.001, It can also be set by the implementer according to the specific implementation scenario, and there are no restrictions here.
[0037] Using the same method described above, the amplitude of the concentration signal of each gas can be adjusted in real time, thereby obtaining the initial compensation amplitude of the concentration signal of each gas at the current time point. Then, the initial compensation amplitude of the concentration signal of each gas at the current time point and the amplitude of the environmental signal of each environmental parameter at the current time point are binarized to obtain the binary value of each gas and each environmental parameter at the current time point. Subsequently, the obtained binary value can be input into the constructed neural network, thereby simplifying data input, facilitating neural network processing, improving the computational efficiency of the neural network, and ensuring the real-time performance of gas detection.
[0038] Preferably, in one embodiment of the present invention, the method for obtaining the binary values of each gas and each environmental parameter at the current time point specifically includes: For any gas, if the initial compensation amplitude of the gas concentration signal at the current time point exceeds the standard safety threshold corresponding to the gas, then the binary value of the gas at the current time point is set to the value 1; otherwise, the binary value of the gas at the current time point is set to the value 0. Each gas corresponds to a standard safety threshold, and exceeding the threshold indicates a dangerous phenomenon. In coal mine safety production, the standard safety threshold of each gas is a known value.
[0039] For any environmental parameter, if the amplitude of the environmental signal at the current time point belongs to the preset effective range corresponding to the environmental parameter, then the binary value of the environmental parameter at the current time point is set to the value 1; otherwise, the binary value of the environmental parameter at the current time point is set to the value 0. In one embodiment of the present invention, the preset effective range corresponding to the temperature parameter is set to -10 to 55 degrees Celsius, the preset effective range corresponding to the humidity parameter is set to 10% to 95%, and the preset effective range corresponding to the air pressure parameter is set to 85 kPa to 115 kPa. The preset effective range corresponding to each environmental parameter can also be set by the implementer according to the specific implementation scenario, and is not limited here.
[0040] Step S3: Train the constructed convolutional neural network to obtain the trained convolutional neural network. The input of the convolutional neural network is the binary value of each gas and each environmental parameter at a given time point, and the output of the convolutional neural network is the compensation concentration value of each gas.
[0041] To further eliminate interference from environmental factors and mutual interference between gases, this invention constructs a convolutional neural network (CNN), trains it, and outputs a compensation concentration value for each gas, thereby achieving effective compensation for gas detection. In this embodiment, the input of the constructed CNN is a binary value of each gas and each environmental parameter at a given time point. That is, each node in the input layer of the CNN corresponds to one gas or one environmental parameter, and the nodes in the input layer take the binary value of each gas or environmental parameter at a given time point in chronological order. The output of the CNN is the compensation concentration value for each gas. The CNN includes an input layer, a convolutional layer, a pooling layer, and an output layer, but does not include a fully connected layer, thus simplifying the CNN model structure, reducing the computational complexity of the CNN, and facilitating its implementation in intelligent coal mine sensing systems. Furthermore, the number of nodes in each layer is equal to the sum of the number of all gases and the number of all environmental parameters. In one embodiment of this invention, a four-in-one sensor is used, meaning there are four gases and three detected environmental parameters, resulting in seven nodes per layer.
[0042] Meanwhile, the pooling layer of the constructed convolutional neural network adopts a direct connection fusion method, which reduces the impact of independent influencing factors on the complexity of the neural network graph and avoids information loss during dimensionality reduction of traditional pooling layers. Furthermore, a nonlinear transformation is set between the convolutional layer and the pooling layer of the convolutional neural network, which can effectively remove the mutual interference between gases and the interference of temperature, humidity and air pressure on gas detection. In one embodiment of the present invention, the nonlinear transformation set between the convolutional layer and the pooling layer is implemented through the ReLU activation function.
[0043] Then, in this embodiment of the invention, it is also necessary to construct the convolution kernel of each node of the convolutional layer of the convolutional neural network, and train the convolution kernel of the convolutional neural network to obtain the trained convolutional neural network. Subsequently, the binary values of each gas and each environmental parameter obtained above at the current time point can be input into the trained convolutional neural network to achieve effective compensation for the detection of each gas concentration.
[0044] Preferably, in one embodiment of the present invention, the method for obtaining the trained convolutional neural network specifically includes: Use any node in the input layer of the convolutional neural network as the target input node, and any node in the convolutional layer of the convolutional neural network as the target convolutional node. By using the zero-point voltage value of the sensor corresponding to the target input node, the convolution calculation formula from the target input node to the target convolution node is simplified to obtain the simplified formula for the convolution calculation formula from the target input node to the target convolution node. The simplified formula for convolution is:
[0045] in, This represents the response function from the target input node to the target convolution node; This represents the zero-point voltage value of the sensor corresponding to the target input node. For example, when the target input node corresponds to a temperature parameter, then... This represents the zero-point voltage value of the temperature sensor corresponding to the target input node; This represents the weight values in the convolution kernel from the target input node to the target convolution node; Indicates a point in time.
[0046] The formula for calculating the convolution from the target input node to the target convolution node is as follows:
[0047] in, This represents the response function from the target input node to the target convolution node; This represents the activation function in the target input node; This represents the impulse function from the target input node to the target convolution node; Indicates a point in time.
[0048] By using the same method described above, a simplified formula for the convolution calculation from each node of the input layer to the target convolution node can be obtained. Then, the simplified formulas for the convolution calculation from all nodes of the input layer to the target convolution node are added together, and the sum of the zero-point voltage values of the corresponding sensors of all input nodes is set to 0. The zero-point offset is then removed to obtain the convolution kernel of the target convolution node.
[0049] The specific implementation process is as follows: First, add the simplified formulas for the convolution calculation from all nodes in the input layer to the target convolution node to obtain:
[0050] in, This indicates the total number of nodes in the input layer. Represents the first input layer The response function from each node to the target convolutional node; Represents the first input layer Each node corresponds to the zero-point voltage value of the sensor; Represents the first input layer The weights in the convolution kernel from each node to the target convolution node are then set as follows: get:
[0051] visible, This refers to the convolution kernel needed for the target convolutional node. The convolution kernel for each node in the convolutional layer can be obtained using the same method described above. Then, using the sample data collected by the all-in-one gas sensor, the convolution kernel of the convolutional neural network is trained to obtain the trained convolutional neural network.
[0052] Preferably, in one embodiment of the present invention, the method for obtaining the trained convolutional neural network further includes: The concentration sample signals of different gases and environmental sample signals of different environmental parameters collected by the multi-in-one gas sensor are binarized and input into the corresponding nodes of the input layer of the convolutional neural network. The loss is calculated between the output value and the true value of the convolutional neural network, and the convolution kernel of each node of the convolutional layer of the convolutional neural network is optimized using the gradient descent method to obtain the trained convolutional neural network. The mean squared error loss function or other loss functions can be used to calculate the loss, and no limitation is made here.
[0053] It should be noted that the training process of a convolutional neural network includes not only training the convolutional kernel, but also training other parameters.
[0054] Step S4: Input the binary values of each gas and each environmental parameter into the input layer nodes of the trained convolutional neural network in chronological order, and output the compensation concentration value of each gas.
[0055] After training the constructed convolutional neural network, the binary values of each gas and each environmental parameter can be input into the input layer nodes of the trained convolutional neural network in chronological order. The convolutional neural network then outputs the compensated concentration value of each gas, thereby compensating for the detection process of the all-in-one gas sensor, eliminating interference from various environmental factors and mutual interference between gases, and improving the gas detection accuracy.
[0056] Meanwhile, in one embodiment of the present invention, after the convolutional neural network outputs the compensation concentration value of each gas, it is also necessary to transmit the compensation concentration value to the monitoring center via wired or wireless means, so that the monitoring center can issue early warnings and human intervention, thereby completing the functional realization and application of the coal mine heterogeneous multi-mode sensing system, realizing the accurate detection of factors affecting coal mine safety production, promoting the intelligent development of coal mines, and accelerating the integration and application of artificial intelligence (AI) with coal mine safety production.
[0057] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0058] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A heterogeneous multimodal neural network compensation algorithm for multi-gas detectors, characterized in that, The method includes: The multi-functional gas sensor is used to collect concentration signals of different gases and environmental signals of different environmental parameters in real time. Based on the amplitude of each environmental signal at the current time point, the amplitude of the concentration signal of each gas is adjusted to obtain the initial compensation amplitude of the concentration signal of each gas at the current time point; the initial compensation amplitude of each concentration signal and the amplitude of each environmental signal are binarized to obtain the binary value of each gas and each environmental parameter at the current time point. The constructed convolutional neural network is trained to obtain a trained convolutional neural network. The input of the convolutional neural network is the binary value of each gas and each environmental parameter at a time point, and the output of the convolutional neural network is the compensation concentration value of each gas. The binary values of each gas and each environmental parameter are input sequentially into each input layer node of the trained convolutional neural network, and the compensated concentration value of each gas is output.
2. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 1, characterized in that, The environmental parameters include at least temperature, humidity, and air pressure.
3. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 2, characterized in that, The initial compensation amplitude of the concentration signal of each gas at the current time point includes: Using any gas as the target gas, when the amplitude of the ambient temperature signal at the current time point falls within a preset first temperature range, the amplitude of the target gas concentration signal at the current time point is negatively adjusted to obtain the temperature compensation amplitude of the target gas concentration signal at the current time point. When the amplitude of the ambient temperature signal at the current time point falls within a preset second temperature range, the amplitude of the target gas concentration signal at the current time point is not adjusted, and the amplitude of the target gas concentration signal is used as the temperature compensation amplitude of the target gas concentration signal at the current time point. When the amplitude of the ambient temperature signal at the current time point falls within a preset third temperature range, the amplitude of the target gas concentration signal at the current time point is positively adjusted to obtain the temperature compensation amplitude of the target gas concentration signal at the current time point. Based on the amplitude of the environmental signal of the humidity parameter at the current time point, the temperature compensation amplitude of the target gas concentration signal at the current time point is negatively adjusted to obtain the humidity compensation amplitude of the target gas concentration signal at the current time point. Based on the amplitude of the environmental signal of the air pressure parameter at the current time point, the humidity compensation amplitude of the target gas concentration signal at the current time point is positively adjusted to obtain the initial compensation amplitude of the target gas concentration signal at the current time point.
4. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 1, characterized in that, The process of obtaining the binary values of each gas and each environmental parameter at the current time point includes: For any gas, if the initial compensation amplitude of the gas concentration signal at the current time point exceeds the standard safety threshold corresponding to the gas, then the binary value of the gas at the current time point is set to the value 1; otherwise, the binary value of the gas at the current time point is set to the value 0. For any environmental parameter, if the amplitude of the environmental signal of the environmental parameter at the current time point is within the preset valid range corresponding to the environmental parameter, then the binary value of the environmental parameter at the current time point is set to the value 1; otherwise, the binary value of the environmental parameter at the current time point is set to the value 0.
5. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 1, characterized in that, The convolutional neural network includes an input layer, a convolutional layer, a pooling layer, and an output layer, but does not include a fully connected layer. The pooling layer uses a direct connection fusion method. A nonlinear transformation is set between the convolutional layer and the pooling layer. The number of nodes in each layer is equal to the sum of the number of all gases and the number of all environmental parameters.
6. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 1, characterized in that, The nonlinear transformation between the convolutional and pooling layers of the convolutional neural network is implemented using the ReLU activation function.
7. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 1, characterized in that, The completed training of the convolutional neural network includes: Use any node of the input layer of the convolutional neural network as the target input node, and any node of the convolutional layer of the convolutional neural network as the target convolutional node. By using the zero-point voltage value of the sensor corresponding to the target input node, the convolution calculation formula from the target input node to the target convolution node is simplified to obtain the simplified formula for the convolution calculation formula from the target input node to the target convolution node. The simplified formula for the convolution calculation is as follows: ;in, This represents the response function from the target input node to the target convolution node; This represents the zero-point voltage value of the sensor corresponding to the target input node; This represents the weight values in the convolution kernel from the target input node to the target convolution node; This represents a time point. The simplified formulas for the convolution calculations from all nodes in the input layer to the target convolution node are added together, and the sum of the zero-point voltage values of the sensors corresponding to all input nodes is set to 0 to obtain the convolution kernel of the target convolution node. The convolutional kernel of the convolutional neural network is trained using sample data collected by the all-in-one gas sensor to obtain the trained convolutional neural network.
8. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 7, characterized in that, The convolution calculation formula from the target input node to the target convolution node is as follows: ;in, This represents the response function from the target input node to the target convolution node; This represents the activation function in the target input node; This represents the impulse function from the target input node to the target convolution node; Indicates a point in time.
9. The heterogeneous multimodal neural network compensation algorithm for a multi-gas detector according to claim 7, characterized in that, The step of training the convolutional kernel of the convolutional neural network using sample data collected by the multi-in-one gas sensor to obtain the trained convolutional neural network includes: The concentration sample signals of different gases and environmental sample signals of different environmental parameters collected by the multi-in-one gas sensor are binarized and then input into the corresponding nodes of the input layer of the convolutional neural network. The loss between the output value and the true value of the convolutional neural network is calculated, and the convolution kernel of each node of the convolutional layer of the convolutional neural network is optimized using the gradient descent method to obtain the trained convolutional neural network.
10. The heterogeneous multimodal neural network compensation algorithm for multi-gas detectors according to claim 1, characterized in that, After outputting the compensation concentration value for each gas, the compensation concentration value also needs to be transmitted to the monitoring center via wired or wireless means.