A combined prediction method for gas concentration based on dynamic optimal selection of indicators

Abstract

The present invention discloses a combined prediction method for gas concentration based on dynamic optimization of indicators, including: obtaining historical monitoring data of multiple indicators affecting future gas concentration; calculating the Spearman rank correlation coefficient between each indicator and the gas concentration indicator respectively, dynamically optimizing the indicators according to the calculated Spearman rank correlation coefficient to obtain optimized indicators; calculating the feature matrix of each optimized indicator; inputting the feature matrix of each optimized indicator into a preset single-indicator prediction model one by one to respectively obtain the predicted values of the maximum gas concentration within a preset time period after the current moment corresponding to each optimized indicator; inputting the predicted values corresponding to each optimized indicator into a preset combined prediction model to obtain the final predicted value for predicting gas overlimit conditions. The present invention can achieve dynamic and accurate overlimit prediction of the gas concentration and its overlimit conditions in coal mine excavation and working faces.

Description

Technical Field

[0001] The present invention relates to the technical field of gas concentration prediction, and particularly relates to a combined gas concentration prediction method based on dynamic index optimization. Background Art

[0002] Gas disasters are one of the most serious disasters in the process of coal mining and excavation. Once gas disasters such as gas explosions and coal and gas outbursts occur, they will not only cause serious losses of life and property, but also seriously affect the safe production of mines. Accurately predicting the gas concentration in the mining and excavation working face, discovering the situation of gas overrun in advance, and taking effective measures in time to reduce the gas concentration are the key to preventing gas disasters. The safety monitoring system can monitor indicators such as gas concentration, wind speed, and temperature in the mining and excavation working face in real time. If the gas overrun can be predicted, relevant measures can be taken in advance to reduce the gas concentration and reduce the production losses caused by the start and stop of mining equipment due to gas overrun. The multi-source monitoring data in the safety monitoring system provides conditions for gas overrun prediction. The development of deep learning technologies such as neural networks makes it possible to dynamically and accurately predict gas concentration.

[0003] However, in the current methods for predicting gas concentration using deep learning technologies, the prediction indicators are relatively fixed, and the accuracy of gas concentration prediction needs to be further improved. How to effectively extract the correlation features between the factors affecting the change trend of gas concentration, consider the degree of correlation between relevant factors and gas concentration at each moment, and establish the best prediction model to improve the accuracy and robustness of the model is still an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a combined gas concentration prediction method based on dynamic index optimization to solve the technical problem of low prediction accuracy in the existing gas concentration prediction technology.

[0005] To solve the above technical problem, the present invention provides the following technical solutions:

[0006] On the one hand, the present invention provides a combined gas concentration prediction method based on dynamic index optimization, and the combined gas concentration prediction method based on dynamic index optimization includes the following steps:

[0007] Obtain the historical monitoring data of multi-source indicators affecting future gas concentration; wherein, the indicators include gas concentration; the historical monitoring data is the monitoring data within a preset time period before the current moment.

[0008] Based on the obtained historical monitoring data, calculate the Spearman rank correlation coefficient between each index and the gas concentration index respectively, and dynamically optimize the multiple indices according to the calculated Spearman rank correlation coefficients corresponding to each index to obtain the optimized indices; wherein, the optimized indices include the gas concentration and the indices corresponding to the Spearman rank correlation coefficients whose calculation results are greater than the preset threshold;

[0009] Calculate the feature matrix of each optimized index;

[0010] Input the feature matrices of the optimized indices into the preset single-index prediction model one by one to obtain the predicted values of the maximum gas concentration within the preset time period after the current moment corresponding to each optimized index;

[0011] Input the predicted values corresponding to the optimized indices into the preset combined prediction model to obtain the final predicted value.

[0012] Further, when dynamically optimizing the multiple indices according to the calculated Spearman rank correlation coefficients corresponding to each index to obtain the optimized indices, if there is no Spearman rank correlation coefficient greater than the preset threshold, select the top preset number of indices with the largest Spearman rank correlation coefficients as the optimized indices.

[0013] Further, calculating the Spearman rank correlation coefficient between each index and the gas concentration index includes:

[0014] Take the index for which the Spearman rank correlation coefficient is to be calculated as the current index;

[0015] Based on the historical monitoring data, calculate the time series of the gas concentration index composed of the historical monitoring data of the gas concentration and the time series of the current index composed of the historical monitoring data of the current index;

[0016] Calculate the Spearman rank correlation coefficient between the time series of the current index and the time series of the gas concentration index, and take the calculation result as the Spearman rank correlation coefficient between the current index and the gas concentration index.

[0017] Further, the calculating the feature matrix of each optimized index includes:

[0018] Calculate the time series composed of the historical monitoring data of each optimized index corresponding to itself respectively;

[0019] Take the optimized index for which the feature matrix is to be calculated as the current index;

[0020] Calculate the time series features and spatial topological features of the current index; among them, the time series features of the current index include: the monitoring value and the first-order difference value of the current index at the current moment, and the statistical features of the time series of the current index, where the statistical features include: the maximum value, average value, root mean square value, variance, standard deviation, coefficient of dispersion, peak factor, skewness, kurtosis, and range of the time series; the spatial topological feature of the current index is: the time series features of the other indicators except the current index;

[0021] Based on the time series features and spatial topological features of the current index, obtain its corresponding feature matrix.

[0022] Furthermore, the single-index prediction model is a Bi-LSTM (Bi-directional Long Short-Term Memory) model.

[0023] Furthermore, the step of inputting the feature matrices of each preferred index into a preset single-index prediction model one by one to obtain the predicted values of the maximum gas concentration within a preset time period after the current moment for each preferred index includes:

[0024] Adopt the method of deviation standardization to map the feature matrices of each preferred index to the interval [0, 1] respectively to obtain the standardized feature matrices;

[0025] Input the standardized feature matrices into the single-index prediction model one by one, then input the output results into a 3-layer DenseNet, realize feature reuse through the connection of features on the channel, and use the activation function RELU to de-linearize between DenseNet layers; set a Dropout layer to discard neurons from the network with a probability of 20% to prevent overfitting of the model, adopt dense connection and weight sharing to filter out process noise and interference information, learn the predicted values of different features through a supervised method, and finally obtain the predicted values of the maximum gas concentration within a preset time period after the current moment corresponding to the corresponding preferred index through inverse standardization.

[0026] Furthermore, the combined prediction model is an LSTM (Long short-term memory) model.

[0027] Furthermore, the step of inputting the predicted values corresponding to each preferred index into a preset combined prediction model to obtain the final predicted value includes:

[0028] Standardize the predicted values corresponding to each preferred index to obtain the standardized predicted values;

[0029] The standardized predicted values are successively input into a two-layer LSTM. A Dropout layer is set to discard neurons from the network with a probability of 20% to prevent overfitting of the model. The DenseNet layer is used for dense connection and weight sharing. Finally, the final predicted value of the gas concentration within a preset time period after the current moment is obtained through inverse standardization.

[0030] If the final predicted value is greater than the set critical value of the gas concentration, it indicates that there is a possibility of gas concentration exceeding the limit within the preset duration after the current moment; otherwise, it indicates that there is no possibility of gas concentration exceeding the limit.

[0031] On the other hand, the present invention also provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the above method.

[0032] On another aspect, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the above method.

[0033] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0034] The present invention first obtains multivariate monitoring data such as gas concentration, wind speed, temperature, humidity, air pressure, etc. in the coal mine safety monitoring system to obtain monitoring indicators. Then, the Spearman rank correlation coefficient between each monitoring indicator at the current moment and the gas concentration indicator to be predicted is calculated, and index dynamic optimization is carried out according to the Spearman rank correlation coefficient. After that, the time series and spatial topological features of each optimized indicator are calculated to establish a feature matrix. The feature matrix is successively input into the Bi-LSTM single-index model to obtain the predicted values of the maximum gas concentration within a specified time after the current moment for each indicator. Finally, the predicted values obtained for each indicator are input into the LSTM combined model again to obtain the final predicted value Y predict , and then the gas overrun situation is predicted based on the predicted value and the critical value. The present invention can realize the dynamic and accurate prediction of the gas concentration and its abnormal conditions in the coal mine excavation and working face. The present invention is based on the real-time monitoring data in the safety monitoring system, considers the change of the correlation degree between other indicators and the gas concentration to be predicted over time, dynamically optimizes the prediction indicators, different indicators correspond to different moments, truly realizes the dynamic optimization of indicators, ensures the optimal sample indicators at the prediction moment, and improves the prediction accuracy of the gas concentration. In addition, the gas overrun situation can be predicted according to the gas concentration predicted value and its critical value, providing a decision-making basis for safety monitoring personnel, which has certain significance for improving the coal mine safety production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0036] Figure 1 It is a schematic execution flowchart of the gas concentration combined prediction method based on dynamic index optimization provided by the embodiments of the present invention;

[0037] Figure 2 It is a schematic implementation diagram of dynamically optimizing indicators according to the Spearman rank correlation coefficient (SRC) provided by the embodiments of the present invention;

[0038] Figure 3 It is a network structure diagram of the Bi-LSTM single-index prediction model provided by the embodiments of the present invention;

[0039] Figure 4 It is a network structure diagram of the LSTM optimized index combination prediction model provided by the embodiments of the present invention;

[0040] Figure 5 It is a gas concentration prediction result diagram provided by the embodiments of the present invention; among them, (a) is the gas concentration prediction result diagram corresponding to the sensor MM256, (b) is the gas concentration prediction result diagram corresponding to the sensor MM263, and (c) is the gas concentration prediction result diagram corresponding to the sensor MM264;

[0041] Figure 6 It is a gas over-limit situation prediction result diagram provided by the embodiments of the present invention; among them, (a) is the gas over-limit situation prediction result diagram corresponding to the sensor MM256, (b) is the gas over-limit situation prediction result diagram corresponding to the sensor MM263, and (c) is the gas over-limit situation prediction result diagram corresponding to the sensor MM264. Specific embodiments

[0042] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the drawings.

[0043] The first embodiment

[0044] This embodiment provides a combined prediction method for gas concentration based on dynamic optimization of indicators. Based on multivariate monitoring data, the Spearman rank correlation between gas concentration and other indicators is analyzed, and considering the degree of correlation between other indicators and gas concentration changes over time, indicators are dynamically optimized; the time series and spatial topological features of the optimized indicators are extracted and input into the Bi-LSTM single-indicator model, and the output results are input into the LSTM combined model again to establish a combined prediction model for gas concentration based on indicator dynamic optimization and Bi-LSTMs. Realize dynamic prediction of gas concentration to achieve the purpose of predicting gas overrun in advance.

[0045] Next, starting from the multivariate monitoring data of a fully mechanized coal mining face in a certain mine, the implementation process of the gas concentration combined prediction method of this embodiment will be described. Its execution flow is as Figure 1 shown, including the following steps:

[0046] S1. Obtain the historical monitoring data of multivariate indicators that affect future gas concentration;

[0047] Among them, the historical monitoring data of multivariate indicators is: the monitoring data within the preset time window length l from the current moment t c forward, obtained from the coal mine safety monitoring system; the multivariate indicators obtained in this embodiment include: gas concentration, wind speed, temperature, humidity, air pressure, etc., a total of 28 monitoring indicators.

[0048] Specifically, in this embodiment, three key sensors, MM263, MM264, and MM256, are arranged near the upper corner of the fully mechanized coal mining face. Gas accumulation is likely to occur at this location. The three sensors can monitor the gas concentration changes at this location in real time. When the gas concentration monitored by any one of the sensors reaches the alarm level (the early warning critical value is 1.0% for all), the shearer will automatically shut down. The gas concentrations monitored by the three key sensors, MM263, MM264, and MM256, are used as the gas concentration indicators to be predicted respectively.

[0049] S2. Based on the obtained historical monitoring data, calculate the Spearman rank correlation coefficients between each indicator and the gas concentration indicator respectively, and dynamically optimize the multivariate indicators according to the calculated Spearman rank correlation coefficients corresponding to each indicator to obtain optimized indicators; among them, the optimized indicators include gas concentration, and the indicators corresponding to the Spearman rank correlation coefficients whose calculation results are greater than the preset threshold;

[0050] Specifically, in this embodiment, as Figure 2 shown, the above S2 includes:

[0051] S21. At the current moment t c , based on the historical monitoring data, calculate from tc Previously, the time series C(t) of the gas concentration index composed of the monitoring data of the gas concentration within the time window length l = 1d is C(t) = {C 1 (t 1 ), C 2 (t 2 ), C 3 (t 3 ), …, C l (t l )}; where C i (t i ) represents the monitored value of the gas concentration at time t i , i = 1, 2, 3 … l; and the time series X(t) of the current index to be calculated for the Spearman rank correlation coefficient composed of the monitoring data of the current index within the time window length l = 1d before t c is X(t) = {x 1 (t 1 ), x 2 (t 2 ), x 3 (t 3 ), …, x l (t l )}, where x i (t i ) represents the monitored value of the current index at time t i .

[0052] S22. Calculate the Spearman rank correlation coefficient between the time series of each index and the time series of the gas concentration respectively, and take the calculation result as the Spearman rank correlation coefficient (SRC) between the corresponding index and the gas concentration index, to obtain the Spearman rank correlation coefficient matrix ρ = (ρ 1 , ρ 2 , …, ρ n ); where p j represents the Spearman rank correlation coefficient between the j-th index and the gas concentration index, j = 1, 2, 3 … n, and n represents the total number of indexes.

[0053] S23. According to the magnitude of the correlation coefficient values in ρ = (ρ 1 , ρ 2 , …, ρ n ), preferentially select the corresponding indexes with medium correlation or above, that is, select the indexes with p j ≥ 0.4 to obtain m indexes corresponding to ρ' = (ρ' 1 , ρ' 2 , …, ρ' m ) as the preferred index set; where ρ' mRepresents the m-th preferred index. If the number of indices m in the preferred index set is less than 2, that is, except for the gas concentration index, no other index has a Spearman rank correlation coefficient higher than 0.4 with it. At this time, the largest 5 p 1 , ρ 2 , …, ρ n ) will be selected from ρ = (ρ j ), and the corresponding indices, gas concentration, temperature, and humidity indices will be selected. A total of m preferred indices will jointly form the final preferred index set.

[0054] S3. Calculate the characteristic matrix of each preferred index;

[0055] Specifically, in this embodiment, the above S3 includes:

[0056] S31. For the m indices dynamically and preferably selected at the current moment t c , calculate the time series composed of the historical monitoring data of each preferred index respectively;

[0057] S32. Use the index for which the characteristic matrix is to be calculated as the current index (taking gas concentration as an example in this embodiment);

[0058] S33. Calculate the time series characteristics and spatial topological characteristics of the gas concentration;

[0059] Among them, the time series characteristics of the gas concentration include: the gas concentration monitoring value c c at the moment t c (t c ); the first-order difference value D c of the gas concentration at the moment t c (t c ), D c (t c ) = c c (t c ) - C c-1 (t c-1 ); before the current moment t c , the statistical characteristics of the gas concentration time series C l (t) = {C c-l (t c-l ), …, C c-2 (t c-2 ), C c-1 (t c-1 ), C c (t c )} with a time window length l = 1d. Specifically, in this embodiment, the obtained statistical characteristics include: C lThe maximum value, average value, root mean square value, variance, standard deviation, coefficient of variation, peak factor, skewness, kurtosis, and range of (t). In this embodiment, for the gas concentration index at the current moment t c A total of 12 features are calculated and used as the time series features of the gas concentration index.

[0060] The spatial topological features of the gas concentration are: at the current moment t c For the other m - 1 optimized indicators in the m indicators dynamically optimized except for the gas concentration indicator at the time series moment t c Time series features In this embodiment, these are used as the spatial topological features of the gas concentration index.

[0061] It should be noted that after calculating the time series features and spatial topological features of the indicators, a feature matrix can be obtained through direct combination or weighted combination of the two. Of course, other methods can also be used to obtain the feature matrix based on the time series features and spatial topological features. Specifically, in this embodiment, the method for obtaining the feature matrix based on the time series features and spatial topological features is as follows:

[0062] S34. For the m prediction indicators dynamically optimized at the current moment t c Calculate the feature matrix for each indicator with a lag step of h = 5 and a prediction step of p = 30 min That is The prediction step p = 30 min, that is, the prediction time interval is 30 min each time.

[0063] S4. Input the feature matrices of each optimized indicator into a preset single - indicator prediction model one by one, and obtain the predicted values of the maximum gas concentration within a preset time period after the current moment corresponding to each optimized indicator;

[0064] Among them, in this embodiment, the single - indicator prediction model used is the Bi - LSTM (Bi - directional Long Short - Term Memory) model, and its network structure is as Figure 3 shown.

[0065] Based on the above, the implementation process of the above S4 includes:

[0066] S41. Obtain the feature matrix set X = [X 1 , X 2 , …, X m of the m optimized indicators;

[0067] S42. Adopt the method of deviation standardization to normalize X = [X 1 , X 2 , …, Xm each feature matrix X in i is mapped to between [0, 1] respectively to obtain the standardized feature matrix X N ;

[0068] In S43, input X N into the Bi-LSTM, and then input its output result into the 3-layer DenseNet. Feature reuse is achieved through the connection of features on the channel, and the activation function RELU is used between the DenseNet layers to de-linearize; a Dropout layer is set to discard neurons from the network with a probability of 20% to prevent model overfitting. Dense connection and weight sharing are adopted to filter out process noise and interference information, and the predicted values of different features are learned in a supervised manner. Finally, the corresponding time t of the preferred index is obtained through inverse standardization c Then, the maximum value Y of the gas concentration within T time after that. T is defined as the prediction length, and T = 30 min. Furthermore, the prediction value set Y = [Y 1 , Y 2 , …, Y m of the maximum value of the gas concentration corresponding to each preferred index is obtained.

[0069] S5. Input the predicted values corresponding to each preferred index into the preset combined prediction model to obtain the final predicted value;

[0070] Among them, in this embodiment, the combined prediction model used is the LSTM (Long short-term memory) model, and its network structure is as Figure 4 shown.

[0071] Based on the above, the specific implementation process of the above S5 includes:

[0072] S51. Standardize Y = [Y 1 , Y 2 , …, Y m to obtain Y N ;

[0073] S52. Input Y N into the 2-layer LSTM in sequence. Similarly, set the Dropout layer to discard neurons from the network with a probability of 20% to prevent model overfitting. Adopt dense connection and weight sharing of the DenseNet layer, and finally obtain the final predicted value Y of the gas concentration within T time after time t c through inverse standardization; predict ;

[0074] S53. Compare Y predict with the set critical value of the gas concentration. If Y predictIf it is greater than the set critical value of gas concentration, it means that at time t c After that, within T = 30 min, there is a possibility that the gas concentration exceeds the limit, and corresponding danger elimination measures need to be taken in a timely manner; otherwise, it means safety and normal production can be carried out.

[0075] In summary, based on the multi-source monitoring data in the safety monitoring system in this embodiment, the Spearman rank correlation coefficient (SRC) between the gas concentration index to be predicted and other indexes is calculated. Considering the change of the correlation degree over time, the indexes are dynamically optimized. The time series features and spatial topological features of each index are extracted to establish a feature matrix. The feature matrices of m prediction indexes are input into the Bi-LSTM single-index model one by one to obtain the predicted value of the maximum gas concentration within T time after the current time t c for each index. The predicted values obtained for each index are input into the LSTM combined model again to obtain the final predicted value Y predict , and then the gas concentration exceeding the limit situation can be predicted according to the finally obtained predicted value and the preset critical value. The prediction results are as Figure 5 shown.

[0076] Furthermore, to verify the effectiveness of the gas concentration combined prediction method provided in this embodiment, the prediction results are analyzed and evaluated below in combination with the predicted value of gas concentration and the set critical value (1%). The coefficient of determination (R 2 ) is used to evaluate the prediction effect of gas concentration; the prediction efficiency (R) is used to evaluate the prediction effect of the gas concentration exceeding the limit situation. Among them, the coefficient of determination (R 2 ) and the prediction efficiency (R) are defined as follows:

[0077]

[0078]

[0079] In the formula, n is the total number of samples; y i is the true value of gas concentration; is the predicted value of gas concentration. TP is the number of samples where the predicted value exceeds the limit and the true value also exceeds the limit, that is, the number of samples where the over-limit is accurately reported; FN is the number of samples where the predicted value does not exceed the limit but the true value exceeds the limit, that is, the number of missed reports; FP is the number of samples where the predicted value exceeds the limit but the true value does not exceed the limit, that is, the number of false reports; TN is the number of samples where the predicted value does not exceed the limit and the true value also does not exceed the limit, that is, the number of samples where the non-over-limit is accurately reported.

[0080] Specifically, the parameter settings in this embodiment are as follows: the time window length \(l = 1d\), the lag step \(h = 5\), the prediction length \(T = 30min\), and the prediction step \(p = 30min\). That is, based on the multi - variable monitoring data of the previous 1d before the current moment, a feature matrix with a preferred index length of 5 is calculated, and the preferred index is used to predict the gas concentration over - limit situation within 30min after the current moment, with a prediction every 30min. For the three key sensors MM256, MM263, and MM264, in 325 sets of test - set data samples, the determination coefficients (\(R\ 2 ) are 0.980, 0.974, and 0.940 respectively; the prediction results of the gas over - limit situation are as shown in Figure 6 . It can be seen from this that the prediction efficacies (\(R\)) of the gas over - limit situations of the three key sensors are 77.3%, 75.0%, and 87.5% respectively. This proves the effectiveness of the technical solution provided by the present invention in predicting the gas concentration and its over - limit situation.

[0081] Second Embodiment

[0082] This embodiment provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the method of the first embodiment.

[0083] This electronic device may have relatively large differences due to configuration or performance, and may include one or more processors (central processing units, CPU) and one or more memories. Among them, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to perform the above - mentioned method.

[0084] Third Embodiment

[0085] This embodiment provides a computer - readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the method of the first embodiment above. Among them, this computer - readable storage medium can be ROM, random access memory, CD - ROM, magnetic tape, floppy disk, and optical data storage device, etc. The instruction stored therein can be loaded and executed by the processor in the terminal to perform the above - mentioned method.

[0086] In addition, it should be noted that the present invention can be provided as a method, a device, or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer - usable storage media containing computer - usable program code.

[0087] Embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing terminal device generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0088] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, such that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0089] It should also be noted that in this article, the terms "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or terminal device including the said element.

[0090] Finally, it should be noted that the above description is the preferred embodiment of the present invention. It should be pointed out that although the preferred embodiments of the present invention have been described, for those skilled in the art of this technology, once they know the basic creative concept of the present invention, without departing from the principle described in the present invention, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A combined prediction method for gas concentration based on dynamic optimization of indicators, characterized in that, it includes: Obtain historical monitoring data of multiple indicators affecting future gas concentration; wherein, the indicators include gas concentration; the historical monitoring data is the monitoring data within a preset time period before the current moment; Based on the obtained historical monitoring data, calculate the Spearman rank correlation coefficient between each indicator and the gas concentration indicator respectively, and dynamically optimize the multiple indicators according to the calculated Spearman rank correlation coefficient corresponding to each indicator to obtain optimized indicators; wherein, the optimized indicators include gas concentration, and the indicators corresponding to the Spearman rank correlation coefficient with a calculation result greater than a preset threshold; Calculate the feature matrix of each optimized indicator; Input the feature matrices of each optimized indicator into a preset single-indicator prediction model one by one, and respectively obtain the predicted values of the maximum gas concentration within a preset time period after the current moment corresponding to each optimized indicator; Input the predicted values corresponding to each optimized indicator into a preset combined prediction model to obtain the final predicted value; The single-indicator prediction model is a Bi-LSTM model; The step of inputting the feature matrices of each optimized indicator into a preset single-indicator prediction model one by one to respectively obtain the predicted values of the maximum gas concentration within a preset time period after the current moment corresponding to each optimized indicator includes: Adopt the method of deviation standardization to map the feature matrices of each optimized indicator to between [0,1] respectively to obtain the standardized feature matrices; Input the standardized feature matrices into the single-indicator prediction model one by one, and then input its output result into a 3-layer DenseNet. Realize feature reuse through the connection of features on the channel, and adopt the activation function RELU to de-linearize between DenseNet layers; set a Dropout layer to discard neurons from the network with a probability of 20% to prevent model overfitting, adopt dense connection and weight sharing to filter out process noise and interference information, learn the output of different features through a supervised method to obtain the predicted value, and finally obtain the predicted value of the maximum gas concentration within a preset time period after the current moment corresponding to the corresponding optimized indicator through inverse standardization; The combined prediction model is an LSTM model; The step of inputting the predicted values corresponding to each optimized indicator into a preset combined prediction model to obtain the final predicted value includes: Standardize the predicted values corresponding to each optimized indicator to obtain the standardized predicted values; Input the standardized predicted values into a 2-layer LSTM in sequence, set a Dropout layer to discard neurons from the network with a probability of 20% to prevent model overfitting, adopt dense connection and weight sharing of the DenseNet layer, and finally obtain the final predicted value of the gas concentration within a preset time period after the current moment through inverse standardization; If the final predicted value is greater than the set critical value of gas concentration, it means that there is a possibility of gas concentration exceeding the limit within the preset time period after the current moment; otherwise, it means that there is no possibility of gas concentration exceeding the limit.

2. The combined prediction method for gas concentration based on dynamic optimization of indicators according to claim 1, It is characterized in that when dynamically optimizing the multiple indicators according to the Spearman rank correlation coefficients corresponding to the calculated indicators to obtain the optimized indicators, if there is no Spearman rank correlation coefficient greater than the preset threshold, the first preset number of indicators with the largest Spearman rank correlation coefficients are selected as the optimized indicators.

3. The combined prediction method for gas concentration based on dynamic optimization of indicators according to claim 1, It is characterized in that the calculation of the Spearman rank correlation coefficients between each indicator and the gas concentration indicator includes: taking the indicator for which the Spearman rank correlation coefficient is to be calculated as the current indicator; based on the historical monitoring data, calculating the time series of the gas concentration indicator composed of the historical monitoring data of the gas concentration, and the time series of the current indicator composed of the historical monitoring data of the current indicator; calculating the Spearman rank correlation coefficient between the time series of the current indicator and the time series of the gas concentration indicator, and taking the calculation result as the Spearman rank correlation coefficient between the current indicator and the gas concentration indicator.

4. The combined prediction method for gas concentration based on dynamic optimization of indicators according to claim 1, It is characterized in that the calculation of the feature matrix of each optimized indicator includes: respectively calculating the time series composed of the historical monitoring data of each optimized indicator; taking the optimized indicator for which the feature matrix is to be calculated as the current indicator; calculating the time series features and spatial topological features of the current indicator; wherein, the time series features of the current indicator include: the monitoring value and the first-order difference value of the current indicator at the current moment, and the statistical features of the time series of the current indicator, wherein the statistical features include: the maximum value, average value, root mean square value, variance, standard deviation, coefficient of variation, peak factor, skewness, kurtosis and range of the time series; the spatial topological feature of the current indicator is: the time series features of the other indicators except the current indicator; obtaining the corresponding feature matrix based on the time series features and spatial topological features of the current indicator.

Images