A method for intelligently identifying abnormal CO disturbance in mines and determining early warning values ​​of coal spontaneous combustion

Abstract

The present invention discloses a method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for coal spontaneous combustion, comprising the steps of: obtaining historical time series data of CO concentration in the upper corner when there is no sign of coal spontaneous combustion in the goaf during normal mining of the working face; dividing the obtained data into various subsequences, and marking whether each subsequence is an interference sequence; establishing an upper corner CO concentration time series classification model using an LSTM network; establishing a CO concentration time series data reconstruction model in a non-interference state using an LSTM autoencoder network; determining an abnormal warning threshold for coal spontaneous combustion in the goaf according to the reconstruction error during the training of the CO concentration time series data reconstruction model in a non-interference state; and providing a real-time warning for the coal spontaneous combustion state in the goaf. The method is driven by the upper corner CO concentration data measured on site, fully considering the correlation of its time dimension, and can effectively reduce the missed reports and false reports of coal spontaneous combustion in the goaf, and reduce the blindness of fire prevention and extinguishing measures in the goaf.

Description

Technical Field

[0001] The present invention relates to the field of mine coal spontaneous combustion early warning, and in particular to a method for intelligently identifying abnormal CO disturbance in a mine and determining a coal spontaneous combustion early warning value. Background Art

[0002] Spontaneous combustion of coal in goaf is one of the main disasters faced in the process of coal mining in underground mines. Once it occurs, it will cause heavy casualties and huge economic losses. Timely and accurate warning of spontaneous combustion of coal in goaf is of great significance to ensure safe and efficient production of coal mines. CO is an indicator gas that characterizes the spontaneous combustion of coal in goaf. The "Detailed Rules for Fire Prevention and Extinguishing in Coal Mines" clearly points out that 24ppm of carbon monoxide cannot be simply used as the critical value of the indicator gas for spontaneous combustion of coal. At present, the indicator gas measurement experiment is usually used to determine the critical value of CO concentration warning for spontaneous combustion of coal in goaf. The upper corner of the working face is the main accumulation area of ​​CO gas in the goaf and is the key monitoring point for spontaneous combustion of coal in the goaf. However, in the process of coal mining, the underground situation is complex and changeable. The CO gas in the upper corner is often disturbed by non-coal spontaneous combustion factors such as pressure from the working face and blasting, resulting in an abnormal increase in the CO gas concentration in the upper corner. The critical value of CO concentration warning obtained by small laboratory experiments is difficult to effectively reflect the actual ignition state of the goaf, making it impossible for coal mining operators to make accurate judgments, which is easy to cause missed reports and false reports, seriously affecting the progress of coal mining. In addition, the occurrence and development of spontaneous combustion of coal in goaf is a gradual evolution rather than a sudden change process. The traditional method of early warning of spontaneous combustion of coal in goaf using the critical value of CO concentration early warning fails to take this feature into account and is obviously not scientific enough. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a method for intelligently identifying abnormal disturbances in mine CO and determining warning values ​​for coal spontaneous combustion, so as to improve the accuracy of warnings for coal spontaneous combustion in goafs and reduce the blindness of fire prevention and extinguishing measures in goafs.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solution: comprising the following steps:

[0005] Step 1: Obtain the historical time series data of CO concentration in the upper corner of the goaf when there is no sign of spontaneous combustion of coal during normal mining of the working face;

[0006] Step 2: Use the sliding window technology to divide the upper corner CO concentration historical time series data obtained in step 1 into subsequences, and mark each subsequence type as an interference sequence and a non-interference sequence according to the corresponding period of time when there is interference from non-coal spontaneous combustion factors during the actual production of the coal mine; the non-coal spontaneous combustion factors include working face pressure, blasting, coal cutting, and trackless rubber-wheeled vehicle gathering;

[0007] Step 3: Using all subsequences obtained in step 2 as training samples, an LSTM network is used to establish an upper corner CO concentration time series classification model;

[0008] Step 4: Using the non-interference type subsequence obtained in step 2 as a training sample, an LSTM autoencoder network is used to establish a CO concentration time series data reconstruction model under the non-interference state;

[0009] Step 5: Determine the abnormal warning threshold of spontaneous combustion of coal in the goaf area based on the reconstruction error of the CO concentration time series data reconstruction model during the training period under the interference-free state obtained in step 4;

[0010] Step 6: Use the CO sensor to collect CO concentration data at the upper corner of the working face in real time, use the same sliding window technology as step 2 to generate a CO concentration time series, and use the data obtained in steps 3 to 5 to provide real-time warning of spontaneous combustion of coal in the goaf.

[0011] Preferably, in step 1, the time intervals of the historical time series data of the upper corner CO concentration are equal and less than or equal to 15 minutes.

[0012] Preferably, in step 2, the sliding window width of the sliding window technology is greater than or equal to 20, that is, the sliding window contains at least 20 data points, and the moving time step of the sliding window technology is 1 time point.

[0013] Preferably, in step 2, as long as there are data points in the subsequence that are in the interference period of non-coal spontaneous combustion factors, the subsequence is marked as an interference sequence; if all data points in the subsequence are not in the interference period of non-coal spontaneous combustion factors, the subsequence is marked as a non-interference sequence.

[0014] Preferably, in step 3, the LSTM network includes an input layer, a hidden layer and an output layer, and the overall structure is constructed by a LSTM recurrent unit;

[0015] The input layer neuron nodes correspond to the CO concentration values ​​at each moment of the time series, the number of hidden layers and neuron nodes are determined by experimental method, and the sigmoid function is selected as the output layer classifier;

[0016] During the training process, the goal was to minimize the cross entropy loss function, and the Adam adaptive algorithm was used to optimize the cross entropy loss function until the cross entropy loss function converged, and the upper corner CO concentration time series classification model was obtained.

[0017] Preferably, the LSTM recurrent unit includes a forget gate, an input gate, an output gate and a unit state;

[0018] The operation process of the LSTM recurrent unit is as follows: First, according to the current input x tand the external state h at the previous moment t-1 Calculate the forget gate f t , input gate i t , output gate o t and candidate state c t ;

[0019] Then the forget gate f t , input gate i t and the internal state c at the previous moment t-1 Combined with the current internal state c t Make updates;

[0020] Finally, the current internal state c t AND output gate o t Combined, we can get the external state h at the current moment t ;

[0021] The calculation formulas for the forget gate, input gate, output gate, and unit state are as follows:

[0022] Forget gate: f t =σ(W f *[h t-1 ,x t ]+b f )

[0023] Input gate: i t =σ(W i *[h t-1 ,x t ]+b i )

[0024] Output gate: o t =σ(W o *[h t-1 ,x t ]+b o )

[0025] Candidate status: c t =tanh(W c *[h t-1 ,x t ]+b c )

[0026] The internal state at the current moment: c t =f t ·c t-1 +i t · c t

[0027] The external state at the current moment: h t =ot ·tanh(c t )

[0028] Where W is the weight matrix, b is the bias, σ(·) is the sigmoid function, and tanh(·) is the hyperbolic tangent function.

[0029] Preferably, in step 4, the LSTM autoencoder network includes an encoder and a decoder, both of which are constructed by LSTM recurrent units;

[0030] The input of the encoder corresponds to the original value of the CO concentration at each moment of the time series, the output of the encoder is the input of the decoder, and the output of the decoder corresponds to the reconstructed value of the CO concentration at each moment of the time series;

[0031] The mean square error between the original value of CO concentration at each moment of the time series input by the encoder and the reconstructed value of CO concentration at each moment of the time series output by the decoder is selected as the loss function. The Adam adaptive algorithm is used to optimize the loss function until convergence, and the reconstruction model of CO concentration time series data under the interference-free state is obtained.

[0032] Preferably, the mean square error calculation formula between the original value of the CO concentration at each moment of the time series input by the encoder and the reconstructed value of the CO concentration at each moment of the time series output by the decoder is:

[0033]

[0034] Where n is the number of data contained in the time series; x i is the original value of the CO concentration data at the i-th moment in the time series; y i is the reconstructed value of the CO concentration data at the i-th moment in the time series.

[0035] Preferably, in step 5, the goaf coal spontaneous combustion abnormal warning threshold adopts the maximum reconstruction error during the training of the CO concentration time series data reconstruction model under the interference-free state obtained in step 4.

[0036] Preferably, in step 6, the real-time early warning process of spontaneous combustion of coal in goaf is as follows:

[0037] Input the generated CO concentration time series into the upper corner CO concentration time series classification model established in step 3:

[0038] If the upper corner CO concentration time series classification model identifies an interference sequence, continue monitoring and repeat step 6;

[0039] If the upper corner CO concentration time series classification model identifies it as a non-interference series, the generated CO concentration time series is input into the CO concentration time series data reconstruction model under the non-interference state established in step 4, and the reconstruction error is calculated:

[0040] If the reconstruction error is less than or equal to the warning threshold set in step 5, it means that there is no risk of spontaneous combustion of coal in the goaf, and monitoring continues and step 6 is repeated;

[0041] If the reconstruction error is greater than the warning threshold set in step 5, it means that there is a risk of spontaneous combustion of coal in the goaf and it is in an abnormal state, and a warning is issued.

[0042] The beneficial effects of the present invention are:

[0043] (1) The present invention is driven by the field measured CO concentration data and adopts a method combining supervised learning and unsupervised learning to establish an intelligent early warning model for spontaneous combustion of coal in goaf areas, which provides double protection for the prevention of spontaneous combustion of coal in goaf areas. It can effectively reduce the false alarm rate of missed reports of spontaneous combustion of coal in goaf areas, reduce the blindness of fire prevention and extinguishing measures in goaf areas and the cost of fire prevention and extinguishing treatment, and help realize unmanned and intelligent coal mining.

[0044] (2) Spontaneous combustion of coal in goaf is a gradual process rather than a sudden process. The present invention processes the CO concentration data of the upper corner into a time series, fully considers the correlation of CO concentration in the time dimension, and utilizes the advantages of LSTM in processing time series. Compared with the prior art method of using a single-point threshold of CO concentration to warn of spontaneous combustion of coal in goaf, the present invention is more in line with the actual occurrence and development process of spontaneous combustion of coal in goaf and has higher accuracy.

[0045] (3) In the coal mine production process, the goaf is mostly in a non-combustible state, and effective ignition samples are difficult to obtain. The present invention only uses the historical data of CO concentration in the upper corner of the goaf when there is no sign of spontaneous combustion of coal during normal mining of the working face, making sample acquisition easier. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A flow chart of a method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal provided by an embodiment of the present invention;

[0048] Figure 2LSTM network structure diagram provided for an embodiment of the present invention;

[0049] Figure 3 A diagram showing the structure of a LSTM loop unit according to an embodiment of the present invention;

[0050] Figure 4 This is a diagram of the LSTM autoencoder network structure provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, a method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal includes the following steps:

[0053] Step 1: Obtain historical time series data of CO concentration in the upper corner of a coal mine working face when there is no sign of spontaneous combustion of coal in the goaf during normal mining. In this embodiment, the time interval of the time series is 1 minute.

[0054] Step 2: Use the sliding window technology to divide the upper corner CO concentration historical time series data obtained in step 1 into subsequences, and mark each subsequence type as an interference sequence and a non-interference sequence according to the corresponding time period with or without interference from non-coal spontaneous combustion factors during the actual production of the coal mine. Among them, non-coal spontaneous combustion factors include working face pressure, blasting, coal cutting, and trackless rubber-wheeled vehicle gathering.

[0055] As long as there are data points in the subsequence that are in the interference period of non-coal spontaneous combustion factors, the subsequence is marked as an interference sequence; if all data points in the subsequence are not in the interference period of non-coal spontaneous combustion factors, the subsequence is marked as a non-interference sequence;

[0056] In this embodiment, the sliding window width is 20, that is, the sliding window contains 20 data points, and the moving step of the sliding window is 1 time point.

[0057] Step 3: Using all subsequences as training samples, the LSTM network is used to establish the upper corner CO concentration time series classification model.

[0058] like Figure 2As shown in the figure, the LSTM network includes an input layer, a hidden layer, and an output layer. The overall structure is built by the LSTM recurrent unit. The input layer neuron nodes correspond to the CO concentration values ​​at each moment of the time series. The number of hidden layers and neuron nodes is determined by experimental methods.

[0059] In this embodiment, the number of hidden layers and the number of neuron nodes are 2 and 10 respectively, and the sigmoid function is selected as the output layer classifier. During the training process, the cross entropy loss function is minimized as the goal, and the Adam adaptive algorithm is used to optimize the cross entropy loss function until the cross entropy loss function converges to obtain the upper corner CO concentration time series classification model.

[0060] like Figure 3 As shown in Figure 1, the LSTM recurrent unit includes a forget gate, an input gate, an output gate, and a unit state. The operation process of the LSTM recurrent unit is as follows: First, according to the current input x t and the external state h at the previous moment t-1 Calculate the forget gate f t , input gate i t , output gate o t and candidate state c t , and then the forget gate f t , input gate i t and the internal state c at the previous moment t-1 Combined with the current internal state c t Update, and finally update the current internal state c t AND output gate o t Combined, we can get the external state h at the current moment t The calculation formulas for the forget gate, input gate, output gate and unit state are as follows:

[0061] Forget gate: f t =σ(W f *[h t-1 ,x t ]+b f )

[0062] Input gate: i t =σ(W i *[h t-1 ,x t ]+b i )

[0063] Output gate: o t =σ(W o *[h t-1 ,x t ]+b o )

[0064] Candidate status:c t =tanh(W c *[h t-1 ,x t ]+b c )

[0065] The internal state at the current moment: c t =f t ·c t-1 +i t · c t

[0066] The external state at the current moment: h t =o t ·tanh(c t )

[0067] Where W is the weight matrix, b is the bias, σ(·) is the sigmoid function, and tanh(·) is the hyperbolic tangent function.

[0068] Step 4: Using the non-interference type subsequence as training samples, the LSTM autoencoder network is used to establish a CO concentration time series data reconstruction model under the non-interference state.

[0069] like Figure 4 As shown in the figure, the LSTM autoencoder network includes an encoder and a decoder, both of which are built by LSTM recurrent units.

[0070] Among them, the input of the encoder corresponds to the original value of the CO concentration at each moment of the time series, the output of the encoder is the input of the decoder, and the output of the decoder corresponds to the reconstructed value of the CO concentration at each moment of the time series.

[0071] The mean square error (i.e., reconstruction error) between the original value of the CO concentration at each moment of the time series input by the encoder and the reconstructed value of the CO concentration at each moment of the time series output by the decoder is selected as the loss function. Taking the minimum mean square error as the goal, the Adam adaptive algorithm is used to optimize the loss function until convergence, and the reconstruction model of CO concentration time series data under undisturbed state is obtained.

[0072] The formula for calculating the mean square error between the original value of the CO concentration at each moment of the time series input by the encoder and the reconstructed value of the CO concentration at each moment of the time series output by the decoder is:

[0073]

[0074] Where n is the number of data contained in the time series; x i is the original value of the CO concentration data at the i-th moment in the time series; y iis the reconstructed value of the CO concentration data at the i-th moment in the time series.

[0075] Step 5: The maximum reconstruction error during the reconstruction model training of the CO concentration time series data under the interference-free state is used as the abnormal warning threshold of spontaneous combustion of coal in the goaf. In this embodiment, the abnormal warning threshold of spontaneous combustion of coal in the goaf is 0.3226;

[0076] Step 6: Use the CO sensor to collect the CO concentration data of the upper corner of the working surface in real time (the sampling frequency is the same as the time interval of the historical time series data of the upper corner CO concentration obtained in step 1), and use the same sliding window technology as step 2 to generate the CO concentration time series.

[0077] The real-time early warning process of spontaneous combustion of coal in goaf is as follows:

[0078] The generated CO concentration time series is input into the upper corner CO concentration time series classification model:

[0079] If the upper corner CO concentration time series classification model identifies an interference sequence, continue monitoring and repeat step 6;

[0080] If the upper corner CO concentration time series classification model identifies it as a non-interference series, the generated CO concentration time series is input into the CO concentration time series data reconstruction model under the non-interference state, and the reconstruction error is calculated:

[0081] If the reconstruction error is less than or equal to 0.3226, it means that there is no risk of spontaneous combustion of coal in the goaf, and monitoring continues and step 6 is repeated;

[0082] If the reconstruction error is greater than 0.3226, it means that there is a risk of spontaneous combustion of coal in the goaf and it is in an abnormal state, and an early warning is issued.

[0083] Spontaneous combustion of coal in goaf is a process that evolves gradually over time, and goaf is mostly non-ignitable during coal mine production, making it difficult to obtain sample data on spontaneous combustion of coal in goaf. In addition, during coal mining, CO gas at the upper corners of the working face is often disturbed by non-coal spontaneous combustion factors such as working face pressure and blasting, making it impossible for coal mining operators to make accurate judgments, which can easily lead to missed or false reports of spontaneous combustion of coal in goaf.

[0084] The method of the present invention only uses the historical time series data of CO concentration in the upper corner when there is no sign of spontaneous combustion of coal in the goaf during the normal advancement of the working face, divides it into subsequences by using the sliding window technology, and marks each subsequence according to the actual mining situation. By using the advantages of LSTM in processing time series and fully considering the correlation of CO concentration in the time dimension, the upper corner CO concentration time series classification model and the CO concentration time series data reconstruction model under the interference-free state are established, which are used for real-time early warning of spontaneous combustion of coal in the goaf.

[0085] The principle is that the upper corner CO concentration time series classification model can effectively identify whether the upper corner CO concentration time series collected in real time is an interference series.

[0086] If yes, continue monitoring;

[0087] Since the samples used to establish the early warning model are all data under the safe state of the goaf, if not, there are two possibilities: the goaf is in a safe state and the goaf is in an abnormal state.

[0088] The CO concentration time series data reconstruction model under the undisturbed state can effectively reconstruct the CO concentration time series data of the safe state of the goaf, but cannot reconstruct the CO concentration time series data of the abnormal state of the goaf. Then, the collected CO concentration time series is input into the CO concentration time series data reconstruction model under the undisturbed state. If the calculated reconstruction error is greater than the warning threshold, it means that there is a risk of spontaneous combustion of coal in the goaf and it is in an abnormal state, and an early warning is issued. Otherwise, it means that there is no risk of spontaneous combustion of coal in the goaf and it is in a safe state, and monitoring continues.

[0089] The present invention is driven by the time series data of CO concentration measured on site, which is more in line with the characteristic that spontaneous combustion of coal in goaf is a gradual rather than sudden process. It can effectively reduce the false alarm rate of missed alarm of spontaneous combustion of coal in goaf, and help to realize unmanned and intelligent coal mining.

[0090] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal, characterized in that: The following steps are involved: Step 1: Obtain the historical time series data of CO concentration in the upper corner of the goaf when there is no sign of spontaneous combustion of coal during normal mining of the working face; Step 2: Use the sliding window technology to divide the upper corner CO concentration historical time series data obtained in step 1 into subsequences, and mark each subsequence type as an interference sequence and a non-interference sequence according to the corresponding period of time when there is interference from non-coal spontaneous combustion factors during the actual production of the coal mine; Step 3: Using all subsequences obtained in step 2 as training samples, an LSTM network is used to establish an upper corner CO concentration time series classification model; Step 4: Using the non-interference type subsequence obtained in step 2 as a training sample, an LSTM autoencoder network is used to establish a CO concentration time series data reconstruction model under the non-interference state; Step 5: Determine the abnormal warning threshold of spontaneous combustion of coal in the goaf area based on the reconstruction error of the CO concentration time series data reconstruction model during the training period under the interference-free state obtained in step 4; Step 6: Use the CO sensor to collect CO concentration data at the upper corner of the working face in real time, use the same sliding window technology as step 2 to generate a CO concentration time series, and use the data obtained in steps 3 to 5 to provide real-time warning of spontaneous combustion of coal in the goaf.

2. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 1, characterized in that: In step 1, the time intervals of the historical time series data of upper corner CO concentration are equal and less than or equal to 15 minutes.

3. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 1, characterized in that: In step 2, the sliding window width of the sliding window technology is greater than or equal to 20, that is, the sliding window contains at least 20 data points, and the moving time step of the sliding window technology is 1 time point.

4. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 1, characterized in that: In step 2, as long as there are data points in the subsequence that are in the interference period of non-coal spontaneous combustion factors, the subsequence is marked as an interference sequence; if all data points in the subsequence are not in the interference period of non-coal spontaneous combustion factors, the subsequence is marked as a non-interference sequence.

5. The method for intelligently identifying abnormal CO disturbance in a mine and determining the early warning value of coal spontaneous combustion according to claim 1, characterized in that: In step 3, the LSTM network includes an input layer, a hidden layer, and an output layer, and the overall structure is built by the LSTM recurrent unit; The input layer neuron nodes correspond to the CO concentration values ​​at each moment of the time series, the number of hidden layers and neuron nodes are determined by experimental method, and the sigmoid function is selected as the output layer classifier; During the training process, the goal was to minimize the cross entropy loss function, and the Adam adaptive algorithm was used to optimize the cross entropy loss function until the cross entropy loss function converged, and the upper corner CO concentration time series classification model was obtained.

6. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 5, characterized in that: The LSTM recurrent unit includes a forget gate, an input gate, an output gate and a unit state; The operation process of the LSTM recurrent unit is as follows: First, according to the current input x t and the external state h at the previous moment t-1 Calculate the forget gate f t , input gate i t , output gate o t and candidate state c t ; Then the forget gate f t , input gate i t and the internal state c at the previous moment t-1 Combined with the current internal state c t Make updates; Finally, the current internal state c t AND output gate o t Combined, we can get the external state h at the current moment t ; The calculation formulas for the forget gate, input gate, output gate, and unit state are as follows: Forget gate: f t =σ(W f *[h t-1 ,x t ]+b f ) Input gate: i t =σ(W i *[h t-1 ,x t ]+b i ) Output gate: o t =σ(W o *[h t-1 ,x t ]+b o ) Candidate status: c t =tanh(W c *[h t-1 ,x t ]+b c ) The internal state at the current moment: c t =f t ·c t-1 +i t ·c t The external state at the current moment: h t =o t ·tanh(c t ) Where W is the weight matrix, b is the bias, σ(·) is the sigmoid function, and tanh(·) is the hyperbolic tangent function.

7. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 1, characterized in that: In step 4, the LSTM autoencoder network includes an encoder and a decoder, both of which are built by LSTM recurrent units; The input of the encoder corresponds to the original value of the CO concentration at each moment of the time series, the output of the encoder is the input of the decoder, and the output of the decoder corresponds to the reconstructed value of the CO concentration at each moment of the time series; The mean square error between the original value of CO concentration at each moment of the time series input by the encoder and the reconstructed value of CO concentration at each moment of the time series output by the decoder is selected as the loss function. The Adam adaptive algorithm is used to optimize the loss function until convergence, and the reconstruction model of CO concentration time series data under the interference-free state is obtained.

8. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 7, characterized in that: The mean square error calculation formula between the original value of the CO concentration at each moment of the time series input by the encoder and the reconstructed value of the CO concentration at each moment of the time series output by the decoder is: Where n is the number of data contained in the time series; x i is the original value of the CO concentration data at the i-th moment in the time series; y i is the reconstructed value of the CO concentration data at the i-th moment in the time series.

9. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 1, characterized in that: In step 5, the abnormal warning threshold of spontaneous combustion of coal in goaf area adopts the maximum reconstruction error during the training of the CO concentration time series data reconstruction model under the interference-free state obtained in step 4.

10. A method for intelligently identifying abnormal CO disturbance in a mine and determining a warning value for spontaneous combustion of coal as claimed in claim 1, characterized in that: In step 6, the real-time early warning process of spontaneous combustion of coal in goaf is as follows: Input the generated CO concentration time series into the upper corner CO concentration time series classification model established in step 3: If the upper corner CO concentration time series classification model identifies an interference sequence, continue monitoring and repeat step 6; If the upper corner CO concentration time series classification model identifies it as a non-interference series, the generated CO concentration time series is input into the CO concentration time series data reconstruction model under the non-interference state established in step 4, and the reconstruction error is calculated: If the reconstruction error is less than or equal to the warning threshold set in step 5, it means that there is no risk of spontaneous combustion of coal in the goaf, and monitoring continues and step 6 is repeated; If the reconstruction error is greater than the warning threshold set in step 5, it means that there is a risk of spontaneous combustion of coal in the goaf and it is in an abnormal state, and a warning is issued.

Images