Neural Network-Based Filtering Method for Failure Data of Mine Catalytic Sensors

Abstract

The present invention relates to a method for filtering failure data of a mine-used catalytic sensor based on a neural network, belonging to the field of sensor data filtering, and comprising the following steps: S1: Obtain the response curve of the mine-used catalytic sensor; S2: Perform interval sampling on the response curve as sample material, obtain multiple points therefrom as a set of sample data, and sample multiple sets of sample data; S3: Input the sample data into the neural network, construct and train a failure data recognition model; S4: Simulate the failure data recognition model; S5: Apply the trained failure data recognition model to the sensor system to filter the failure data. The present invention applies the neural network to the sensor system, determines and filters the failure data, and greatly improves the detection accuracy of the sensor.

Description

Technical Field

[0001] The present invention belongs to the field of sensor data filtering, and relates to a method for filtering failure data of mine catalytic sensors based on a neural network. Background Art

[0002] Sensors such as methane, oxygen, and hydrogen sulfide sensors in coal mines constitute the perception layer of the coal mine safety system and play a crucial role in coal mine production safety. Currently, most mine sensors are based on the principle of electrochemical catalysis. When the sensitive element of the sensor comes into contact with the monitored gas, an electrochemical reaction occurs, which causes a change in the output voltage of the circuit. The concentration of the monitored gas can be deduced based on the change in the output voltage. When the gas concentration is relatively high, the coal mine safety monitoring system will immediately execute a power-off operation to cut off the underground power transmission line to avoid gas explosion disasters. Since sensors based on the electrochemical principle may have a certain probability of error and false alarm phenomena during use, and false alarms are often regarded as dangerous signals by the safety monitoring system and executed, this will cause the interruption of the underground power transmission circuit, resulting in unnecessary losses. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method for filtering failure data of mine catalytic sensors based on a neural network.

[0004] To achieve the above purpose, the present invention provides the following technical solutions:

[0005] A method for filtering failure data of mine catalytic sensors based on a neural network, comprising the following steps:

[0006] S1: Obtain the response curve of the mine catalytic sensor;

[0007] S2: Use the response curve as sample material and sample at a certain time interval, and obtain multiple points therefrom as a set of sample data, and sample multiple sets of sample data;

[0008] S3: Construct a neural network, use the sample data as input variables to train and simulate the neural network, and complete the construction of the failure data recognition model;

[0009] S4: Apply the failure data recognition model to the catalytic sensor system to filter failure data.

[0010] Further, in step S1, obtain the response curves of the mine catalytic sensor under different concentrations of combustible gases.

[0011] Further, in step S2, use the response curve described in step S1 as sample material, sample once every 1 second, and take every 10 points as a set of sample data, and collect a total of 9 sets of sample data.

[0012] Further, the neural network described in step S3 is a BP neural network.

[0013] Further, the number of hidden layers of the BP neural network described in step S3 is:

[0014]

[0015] where h is the number of neurons in the hidden layer, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an adjustment constant between 1 and 10;

[0016] Take the response time parameter of the methane sensor as 10 attributes of the input sample, and the output y i value is the probability that the sample is close to the true value;

[0017] Select the Sigmoid function that can reflect the binary probability as the activation function. This function maps the values from negative infinity to positive infinity to the range of (0,1), and can effectively represent the probability of the sample;

[0018] X = [x 1 , x 2 , x 3 , x 4 … x i … x 10 T , Wj = [w j1 , w j2 , w j3 … w ji … w jm . If we consider x 0 = 1, w j1 = b j , then

[0019] X = [x 0 , x 1 , x 2 , x 3 , x 4 … x i … x 10 T , Wj = [w j0 , w j1 , w j2 , w j3 … w ji … w jm ;

[0020] The input Z j of the hidden layer node j is expressed as:

[0021] j = 0,1,2...5

[0022] ​​The output node y is expressed as:

[0023]

[0024] y = [y 1 , y 2 , y 3 ,... y p ,... y 12 T , and the set takes the above 12 samples as inputs, and uses x 1 , x 2 , x 3 ,..., x p , x 12 to represent. After the p-th sample is input into the neural network, the obtained y p uses the mean squared error method to calculate the error E of the p-th sample p :

[0025]

[0026] The global error E is equal to:

[0027]

[0028] where t p is the expected input, t = [0.9, 0.9, 0.9,... 0.9, 0.1, 0.1, 0.1] T , to minimize the global error E, adjust w ij and v j weights, so that the response curve samples of normal mine-used catalytic sensors output approach 0.9, while the response curve samples of abnormal ones approach 0.1.

[0029] Furthermore, in step S3, the input sample X is normalized and then simulated.

[0030] Furthermore, in step S4, using the trained failure data recognition model, each time a certain concentration of combustible gas is detected, the response curve of the mine-used catalytic sensor is sampled, and then the sampled data is used as a sample to be input into the network for recognition. When the output result is close to 0.1 and the error is less than 0.1%, it is determined as failure data for filtering.

[0031] The beneficial effects of the present invention are as follows: The present invention applies a neural network to the sensor system to determine and filter failure data, greatly improving the detection accuracy of the sensor.

[0032] ​Other advantages, objects and features of the present invention will be set forth in part in the following description, and in part will be obvious to those skilled in the art upon examination of the following, or may be learned by practice of the present invention. The objects and other advantages of the present invention may be realized and attained by the means of the instrumentalities and combinations particularly pointed out hereinafter. Description of the Drawings

[0033] In order to make the objects, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, where:

[0034] Figure 1 (a), (b), and (c) are respectively the response curve diagrams of the methane sensor under methane gas concentrations of 1000 ppm, 8000 ppm, and 6000 ppm;

[0035] Figure 2 is the structure diagram of the BP neural network;

[0036] Figure 3 is the curve diagram of the Sigmoid function;

[0037] Figure 4 is the schematic diagram of the Matlab simulation vector and error analysis. (a) is the schematic diagram of vectors P and T; (b) is the graph of the change of the error value with the number of training times. Detailed Embodiments

[0038] The following specific examples illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following examples only illustrate the basic concept of the present invention in a schematic manner. Without conflict, the following examples and the features in the examples can be combined with each other.

[0039] Among them, the drawings are only for illustrative purposes, showing only schematic diagrams, not physical diagrams, and should not be construed as a limitation to the present invention; in order to better illustrate the embodiments of the present invention, some components in the drawings will be omitted, enlarged or reduced, and do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0040] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the accompanying drawings. This is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the accompanying drawings are only for illustrative purposes and should not be construed as a limitation of the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0041] The present invention proposes to use the principle of neural network to optimize the reliability of sensors for monitoring harmful gases, which has important practical significance for avoiding the interruption of production operations (cost waste) caused by false alarm signals issued by failure detection.

[0042] Using the principle of neural network, simulate the response curve of catalytic sensors, establish an input-output model, and finally obtain data that effectively filters out failure detections, achieving the purpose of improving the detection accuracy of sensors.

[0043] Taking the methane sensor as an example, under methane gas with different concentrations, the response time curve is as Figure 1 shown.

[0044] Step 1: Select effective and invalid data samples. Using the above response time curve as sample material, sample once every 1 s, and a total of 10 points in each test are used as a sample data. In this way, there are 9 groups of sample data.

[0045] The input and output of normal samples are shown in Table 1:

[0046] Table 1

[0047] Serial number Xi Yi Description 1 120 160 90 60 55 40 37 36 38 35 0.9 Normal 2 80 130 80 60 55 40 37 36 38 35 0.9 Normal 3 100 140 90 60 55 40 37 36 38 35 0.9 Normal 4 98 108 76 54 40 36 33 34 32 33 0.9 Normal 5 97 110 67 40 39 37 36 35 33 30 0.9 Normal 6 105 117 109 43 40 36 35 29 28 27 0.9 Normal 7 70 104 70 53 35 33 29 28 27 27 0.9 Normal 8 70 95 70 56 35 33 32 34 32 31 0.9 Normal 9 52 99 70 56 34 33 30 32 29 28 0.9 Normal 10 119 0 119 0 119 0 119 0 119 0 0.1 Step 11 98 98 98 98 98 98 98 98 98 98 0.1 Mutation 12 102 100 99 101 103 101 98 103 99 101 0.1 Wire break

[0048] Step 2: Construct a BP neural network according to the samples. The number of hidden layers is determined according to the empirical formula, where h is the number of neurons in the hidden layer, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an adjustment constant between 1 and 10. Here, the input vector has 10 attribute values, so m is equal to 10. The output has only one probability value, so n is equal to 1. a is taken as 2, so h is equal to 5, and 5 intermediate nodes are taken.

[0049]

[0050] Take the response time parameter of the methane sensor as the 10 attributes of the input sample, and the output y i value as the probability that the sample is close to the true value. The constructed neural network is as Figure 2as shown

[0051] The activation function selects the Sigmoid function that can reflect binary probability, such as Figure 3 as shown, this function maps the values from negative infinity to positive infinity to the range of (0, 1), and can effectively represent the probability of the sample.

[0052] X = [x 1 , x 2 , x 3 , x 4 … x i … x 10 T , Wj = [w j1 , w j2 , w j3 … w ji … w jm , if we consider x0 = 1, w j1 = b j , then

[0053] X = [x 0 , x 1 , x 2 , x 3 , x 4 … x i … x 10 T , Wj = [w j0 , w j1 , w j2 , w j3 … w ji … w jm . The input Z j of the hidden layer node j can be expressed as:

[0054] j = 0, 1, 2... 5

[0055] The output node y can be expressed as:

[0056]

[0057] y = [y 1 , y 2 , y 3 ,... y p ,... y 12 T , The set takes the above 12 samples as input, and uses x 1 , x 2 , x 3 ,..., x p , x 12 to represent, and the y obtained after the p-th sample is input into the neural network​​​p The error E of the p-th sample is calculated in the way of mean squared error p .

[0058]

[0059] The global error E is equal to:

[0060]

[0061] where t p is the expected input, t = [0.9, 0.9, 0.9,...0.9, 0.1, 0.1, 0.1] T . To minimize the global error E, adjust w ij and v j weights so that the output of the normal methane sensor response curve sample approximates 0.9, while the output of the abnormal response curve sample approximates 0.1.

[0062] The S-shaped activation function is very flat in the region outside the interval (0, 1), and the discrimination is too small. For example, when the parameter a = 1 in the S-shaped function f(X), the difference between f(100) and f(5) is only 0.0067. Therefore, the input sample X needs to be normalized before using matlab simulation.

[0063] Step 3, simulation and network effectiveness analysis.

[0064] Taking X as the input P and Y as the output T, the matlab implementation is as follows:

[0065] netl = newff(minmax(P), [5 1], {'logsig', 'purelin'}, 'traingd');

[0066] netl.trainParam.goal = 1e-5;

[0067] net1.trainParam.epochs = 1500;

[0068] netl.trainParam.lr = 0.05;

[0069] net1.trainParam.showWindow = 1;

[0070] netl = train(netl, P, T);

[0071] Figure 4 (a) is a schematic diagram of the vectors P and T Figure 4(b) is the graph of the error value varying with the number of training times. It can be seen that there is some fluctuation at the beginning of the network training, and then the error decreases linearly with the number of training times. The expected value of 1e-5 is reached at the 292nd training time.

[0072] The weight set of the intermediate layer Wj is as follows:

[0073]

[0074] The weight set of the output layer Zj is as follows:

[0075] 0.7262 -0.6749 -0.8370 0.7925 0.4417

[0076] The threshold B1 set of the intermediate layer is as follows:

[0077] 1.1861 -1.3526 -3.1100 1.5138 0.3402

[0078] The threshold of the output layer is 0.2165. With the above parameters, the entire neural network is constructed.

[0079] To verify the reliability of the network, randomly select abnormal samples of the above X and add noise spikes, increase the peak value, etc. to verify the reliability of the network.

[0080] Serial number <![CDATA[Input sample x n > <![CDATA[y n > Explanation 1 129 0 129 0 129 0 139 0 129 0 0.1051 Step 2 239 0 239 0 235 0 237 0 232 0 0.1051 Step 3 60 0 69 0 63 0 62 0 61 0 0.1074 Step 4 198 192 196 197 198 195 198 193 192 191 0.1051 Mutation 5 78 71 76 77 78 55 98 93 92 91 0.1052 Mutation 6 55 56 57 59 78 67 68 70 72 71 0.1053 Mutation 7 120 122 119 121 113 111 98 123 99 101 0.1069 Wire break 8 110 111 102 99 97 98 98 103 99 101 0.1068 Wire break 9 83 82 81 80 79 78 76 75 79 89 0.1076 Wire break

[0081] Step 4, effect analysis. As shown in the above table, according to the three important failure factors when the methane sensor detects the gas concentration, randomly construct 9 samples and put them into the trained network for test verification. It can be seen that the network can effectively classify the failure data, and the network output value is near 0.1, and the error is less than 0.1%. It can be seen that the BP neural network constructed this time can effectively perform effective function fitting on the discrete sampling sample data of the input sensor response curve and approximate the sensor response curve.

[0082] Apply the constructed network to the sensor system. Each time a certain concentration of combustible gas is detected, sample the response time curve of the sensor, and then use the sampled data as a sample to input into the network for recognition. When the output result is close to 0.1 and the error is less than 0.1%, it can be basically determined as invalid data for filtering. In this way, the detection accuracy of the sensor is greatly improved.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limitations. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for filtering failure data of mine-used catalytic sensors based on neural networks, characterized in that: It includes the following steps: S1: Obtain the response curve of the mine-used catalytic sensor; S2: Take the response curve as sample material and sample it at a certain time interval, and obtain multiple points from it as a set of sample data, and sample multiple sets of sample data; S3: Build a neural network, use the sample data as input variables to train and simulate the neural network, and complete the construction of the failure data recognition model; the neural network is a BP neural network, and the number of hidden layers of the BP neural network is: where h is the number of neurons in the hidden layer, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an adjustment constant between 1 and 10; Take the response time parameter of the methane sensor as 10 attributes of the input sample, and the output y i value as the probability that the sample is close to the true value; The activation function selects the Sigmoid function that can reflect the binary probability. This function maps the values from negative infinity to positive infinity to the range of (0, 1), and can effectively represent the probability of the sample; X = [x 1 , x 2 , x 3 , x 4 … x i … x 10 T , Wj = [w j1 , w j2 , w j3 … w ji … w jm , if we consider x 0 = 1, w j1 = b j , then​ X = [x 0 , x 1 , x 2 , x 3 , x 4 … x i … x 10 T , Wj = [w j0 , w j1 , w j2 , w j3 … w ji … w jm ;​ The input Z of hidden layer node j j is expressed as: The output node y is expressed as: y = [y 1 , y 2 , y 3 ,... y p ,... y 12 T , The set takes the above 12 samples as input, and uses x 1 , x 2 , x 3 ,..., x p , x 12 to represent. After the p-th sample is input into the neural network, the obtained y p calculates the error E of the p-th sample in the way of mean squared error p :​ The global error E is equal to: where t p is the expected input, t = [0.9, 0.9, 0.9,... 0.9, 0.1, 0.1, 0.1] T , to minimize the global error E, adjust w ij and v j weights, so that the output of the normal mine use catalytic sensor response curve sample approximates 0.9, while the output of the abnormal response curve sample approximates 0.1; S4: Apply the failure data recognition model to the catalytic sensor system to filter the failure data.

2. The method for filtering failure data of mine-used catalytic sensors based on neural networks according to claim 1, characterized in that: In step S1, obtain the response curves of the mine-used catalytic sensor under different concentrations of combustible gas.

3. The method for filtering failure data of mine-used catalytic sensors based on neural networks according to claim 1, characterized in that: In step S2, take the response curve described in step S1 as sample material, sample once every 1 second, take every 10 points as a set of sample data, and collect a total of 9 sets of sample data.

4. The method for filtering failure data of mine-used catalytic sensors based on neural networks according to claim 1, characterized in that: In step S3, normalize the input sample X and then perform simulation.

5. The method for filtering failure data of mine-used catalytic sensors based on neural networks according to claim 1, characterized in that: In step S4, use the trained failure data recognition model. Each time a certain concentration of combustible gas is detected, sample the response curve of the mine-used catalytic sensor, and then use the sampled data as a sample to input the network for recognition. When the output result is close to 0.1 and the error is less than 0.1%, it is determined as failure data for filtering.

Images