Coke oven gas generation amount prediction method based on multi-time scale decomposition
By using an LSTM model with multi-timescale decomposition and attention mechanism, the problems of high computational complexity and low accuracy in predicting coke oven gas production are solved, and more efficient prediction results are achieved.
Patent Information
- Application Number
- CN202510627724.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-15
- Publication Date
- 2025-10-24
AI Technical Summary
Existing methods for predicting coke oven gas production have high computational complexity, making it difficult to effectively handle nonlinear fluctuations caused by changes in operating conditions. Furthermore, traditional models struggle to capture long-term time dependencies, resulting in limited prediction accuracy.
A multi-timescale decomposition method is adopted, which decomposes coke oven gas generation data through EEMD, calculates sample entropy and mutual information coefficient, reconstructs periodic components and trend components, and uses an LSTM model with added attention mechanism for prediction. The prediction model is constructed by combining the optimal lag period and key process parameters.
It effectively removes noise, improves signal quality, reduces model computational complexity, and enhances the accuracy and reliability of coke oven gas generation prediction.
Smart Images
Figure CN120832976A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of coke oven gas generation amount prediction, and particularly relates to a coke oven gas generation amount prediction method based on multi-time scale decomposition. BACKGROUND
[0002] As a very important secondary energy medium in the energy system of a steel enterprise, gas is a high-quality gaseous fuel generated in the steel production process. The effective utilization rate of gas is an important indicator of the energy management level of an enterprise. The importance of gas prediction and scheduling is reflected in the following four aspects: first, gas has high heat, and its full utilization can greatly reduce the energy consumption and production cost per ton of steel; second, gas emission will cause environmental pollution and needs to be effectively controlled; third, the generation amount of gas is difficult to predict, and the generation amount of gas is closely related to the production process. The generation of gas involves various physical and chemical processes, and the mechanism is complex; fourth, there are many types of gas, including coke oven gas, blast furnace gas, converter gas and mixed gas.
[0003] Gas generation amount prediction is an important basis for rational gas scheduling, and an energy balance scheduling system is an important carrier for realizing gas scheduling. The prediction result of the gas generation amount is an important basis for formulating a reasonable energy use plan and designing a dynamic gas scheduling method, and directly affects the effectiveness of subsequent optimization of the gas system. Predicting the generation amount of gas can help to grasp the supply and demand trend of the gas system in advance, control the stability of the gas tank, avoid the imbalance between supply and demand of the gas system, maximize the use of gas resources, and reduce energy loss and waste.
[0004] Coke oven gas is one of the most important by-product gases in the steel production process, and has the characteristics of high calorific value, large output and wide application. It is widely used for heating, power generation and as a synthetic raw material in steel enterprises. The fluctuation of its output will directly affect the energy scheduling and stable operation of the system.
[0005] The existing coke oven gas generation amount prediction mainly includes: (1) traditional machine learning methods, including least squares support vector machine, autoregressive difference moving average model, etc. The traditional machine learning method has high requirements for the stationarity of time series, and it is difficult to effectively process the nonlinear fluctuations caused by the change of the working condition of the coal gas production, and cannot fully excavate the complex nonlinear relationship between multiple variables, resulting in limited prediction accuracy. (2) BP neural network method, first detect and remove abnormal data through isolated forest algorithm, and input the normalized data into the BP neural network for training. For example, the input parameters include historical coal gas generation amount and process parameters affecting coke oven gas generation amount, and the output is the predicted value at the future time. The BP neural network is difficult to capture the long sequence time dependence, resulting in a significant increase in prediction error in the fluctuating data. (3) Deep learning model, using long short-term memory network (LSTM), gated recurrent unit (GRU) and other time series processing. Through gray correlation analysis to screen input parameters, and using box plot and Hampel filter to process abnormal values, the model robustness is improved, and the model training requires a large amount of computing resources, and the parameter tuning process is tedious. SUMMARY
[0006] The purpose of the present application is to solve the problem of high computational complexity of the existing coke oven gas generation amount prediction deep learning method, and to propose a coke oven gas generation amount prediction method based on multi-time scale decomposition, comprising the following steps:
[0007] S1, collect the historical data of coke oven gas generation amount and process parameters affecting coke oven gas generation amount, and smooth the coke oven gas generation amount data;
[0008] S2, decompose the processed coke oven gas generation amount data using EEMD to obtain multiple intrinsic mode function (IMF) components, and calculate the average instantaneous frequency of each intrinsic mode function component by Hilbert transform to remove high-frequency noise signals in the intrinsic mode function component;
[0009] S3, evaluate the complexity of each intrinsic mode function component by calculating the sample entropy, and reconstruct the intrinsic mode function component into periodic component and trend component according to the complexity of the intrinsic mode function component;
[0010] S4, based on the periodic component, trend component and process parameters affecting coke oven gas generation amount, calculate the mutual information coefficient to obtain the key process parameters affecting coke oven gas generation amount;
[0011] S5, use LSTM with attention mechanism as the prediction model, obtain the optimal lag period of the model input by Bayesian information criterion, and use the periodic component, trend component, key process parameters and optimal lag period as the input variables of the prediction model to predict the coke oven gas generation amount.
[0012] Further, the coke oven gas generation amount data is smoothed using a Savitzky-Golay filter, and the filtering target is to minimize the following objective function at each point:
[0013]
[0014] where a0,...,a k denote the coefficient vector, a l denotes the l-th coefficient vector, j denotes the offset position of each point in the sliding window relative to the current center point, y i+j denotes the j-th coke oven gas generation amount data in the window with the center point y i , and [-m,m] denotes a sliding window with a length of 2m+1.
[0015] Further, the coke oven gas generation amount data after smoothing is decomposed using the EEMD method to obtain a group of intrinsic mode function (IMF) components:
[0016]
[0017] wherein denotes the coke oven gas generation amount data after smoothing, IMF i (t) denotes the i-th intrinsic mode function component, r(t) is a residual term, and M is the total number of modal components.
[0018] The average instantaneous frequency of each intrinsic mode function component is obtained by Hilbert transform, denoted as:
[0019]
[0020] wherein z i (t) denotes the analytic signal of the i-th intrinsic mode function, H() denotes Hilbert transform, i denotes the i-th, A i (t) denotes the amplitude envelope of the i-th IMF component, j denotes the imaginary unit, and θ i (t) denotes the instantaneous phase of the i-th IMF component.
[0021] Let the analysis interval be [t1,t2], then the average instantaneous frequency of the i-th component is:
[0022]
[0023] wherein denotes the average instantaneous frequency of the i-th component.
[0024] Further, S3 is specifically:
[0025] Calculate the sample entropy corresponding to each component:
[0026]
[0027] Where SampEn(m, r, N) represents the sample entropy with parameters (m, r, N), m represents the dimension of the template vector, r represents the similarity tolerance, N represents the number of intrinsic mode function components, A m (r) represents the probability that the two sequences of the template vector match m+1 points, B m (r) represents the probability that the two sequences of the template vector match m points under the similarity tolerance r, A i represents the number of pairs of template vectors with length m+1 whose distance is less than r, B i represents the number of pairs of template vectors with length m whose distance is less than r.
[0028] According to the size of the sample entropy, the intrinsic mode function components are reconstructed into periodic components and trend components:
[0029]
[0030] Further, based on the periodic components, trend components, and process parameters affecting the coke oven gas generation amount, the key process parameters affecting the coke oven gas generation amount are obtained by calculating the mutual information coefficient, represented as:
[0031]
[0032] Where I(y i ; x k ) represents the mutual information coefficient of y i and x k , y i represents the i-th periodic component or trend component signal, x k represents the k-th process parameter, p(y i , x k ) represents the joint probability of y i and x k , p(y i ) represents the probability of y i , and p(x k ) represents the probability of x k .
[0033] Based on the size of the mutual information coefficient, the key process parameters affecting the coke oven gas generation amount are selected.
[0034] Further, the key process parameters affecting the coke oven gas generation amount are: coking state and volatile matter of blended coal;
[0035] The coking state is divided as follows:
[0036]
[0037] Wherein, t is the interval between actual coal loading time and current time, r is the coking cycle of the coke oven.
[0038] Further, the optimal lag period of the model input is represented by the Bayesian information criterion as follows:
[0039] BIC=k·lnn-2lnL,
[0040] Wherein, k is the number of free parameters in the model, L is the maximum likelihood estimation of the model on the coke oven gas generation data, n is the sample size, BIC represents the Bayesian criterion, and the optimal lag period is obtained when BIC reaches the minimum value.
[0041] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the coke oven gas generation prediction method based on multi-time scale decomposition.
[0042] The application further provides an electronic device, which comprises a processor and a memory, and the processor and the memory are connected with each other, wherein the memory is used for storing a computer program, the computer program comprises computer readable instructions, the processor is configured to call the computer readable instructions, and the coke oven gas generation prediction method based on multi-time scale decomposition is executed.
[0043] The application further provides a computer program product, which comprises computer program / instructions, and the computer program / instructions are executed by a processor to implement the steps of the coke oven gas generation prediction method based on multi-time scale decomposition.
[0044] The technical scheme provided by the application has the beneficial effects that:
[0045] The present application considers the multi-time scale characteristics of coke oven gas generation, decomposes the coke oven gas generation using the EEMD method, and calculates the average instantaneous frequency of each mode. Then, the periodic component and the trend component are reconstructed by calculating the sample entropy, the main factors affecting the coke oven gas generation and the optimal lag period of each component signal are obtained by calculating the mutual information coefficient and the Bayesian information criterion, and the coke oven gas generation is predicted using the LSTM with attention mechanism as the prediction model, with the periodic component, the trend component, the key process parameters and the optimal lag period as the input variables of the prediction model. The present application uses the EEMD method to decompose the complex signal into several intrinsic mode functions with different characteristic scales, the intrinsic mode function represents the fluctuation characteristics of the original signal at different time scales, so that the local characteristics and change rules of the signal can be more clearly analyzed, and the noise in the signal can be effectively removed, the quality and signal-to-noise ratio of the signal are improved, the LSTM with attention mechanism is used as the prediction model, and the periodic component, the trend component, the key process parameters and the optimal lag period are used as the input variables of the prediction model, which effectively reduces the model calculation complexity and improves the prediction accuracy and reliability. BRIEF DESCRIPTION OF DRAWINGS
[0046] Figure 1 is a flowchart of the coke oven gas generation prediction method based on multi-time scale decomposition of the present application embodiment;
[0047] Figure 2 is 11 intrinsic mode functions obtained by the EEMD method of the present application embodiment;
[0048] Figure 3 is the average instantaneous frequency of each intrinsic mode function obtained by the Hilbert transform of the present application embodiment;
[0049] Figure 4 is the reconstruction result of the intrinsic mode function component of the present application embodiment;
[0050] Figure 5 is the calculation result of selecting the periodic component and the trend component by calculating the mutual information coefficient of the present application embodiment, Figure 5 (a) in (a) represents the calculation result of selecting the periodic component by calculating the mutual information coefficient, and (b) represents the calculation result of selecting the trend component by calculating the mutual information coefficient;
[0051] Figure 6 is the input variable lag period result of the periodic component and the trend component of the present application embodiment, Figure 6 (a) in (a) is the input variable lag period result of the periodic component, Figure 6 (b) in (b) is the input variable lag period result of the trend component;
[0052] Figure 7is a comparison chart of the coke oven gas generation amount prediction result based on the LSTM-Attention model and the actual coke oven gas generation amount of an embodiment of the present application;
[0053] Figure 8 is a block diagram of an electronic device in an exemplary embodiment of the present application. DETAILED DESCRIPTION
[0054] To make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described below with reference to the drawings.
[0055] The flowchart of the coke oven gas generation amount prediction method based on multi-time scale decomposition of the embodiment of the present application is as shown in Figure 1 , and specifically includes the following steps:
[0056] S1, collect the coke oven gas generation amount and the historical data of the process parameters affecting the coke oven gas generation amount, and perform smoothing processing on the coke oven gas generation amount data.
[0057] The historical data of the coke oven gas generation amount is collected, the sampling period is set to 1 min, and the actual coal charging time, actual coking time, planned coal charging time, planned coking time, volatile matter of blended coal and the historical data of continuous months of the coking process are collected. The coke oven gas generation amount data collected may contain interference of multiple coupled noise signals, and the Savitzky-Golay filter is used for smoothing processing, which preliminarily removes multiple coupled noises, and the signal-to-noise ratio of the smoothed data is 36.056. The basic characteristics of the original data are retained while denoising.
[0058] The original sequence of the coke oven gas generation amount is y={y1,y2,K,y n}, and the target of the SG filtering is to minimize the following objective function at each point, and the filtered signal is
[0059]
[0060] wherein a0,...,a k represents a coefficient vector, a l represents the lth coefficient vector, j represents the offset position of each point in the sliding window relative to the current center point, y i+k represents the jth coke oven gas generation amount data in the window with the center point y i , and [-m,m] represents a sliding window with a length of 2m+1.
[0061] S2, using Ensemble Empirical Mode Decomposition (EEMD) to decompose the processed coke oven gas generation data, obtain a plurality of Intrinsic Mode Function (IMF) components, and obtain the average instantaneous frequency of each intrinsic mode function component by Hilbert transform method, and remove the high frequency noise signal in the intrinsic mode function component.
[0062] The coke oven gas generation has a multi-time scale characteristic, and the coke oven gas generation is decomposed by using the EEMD method to obtain a group of intrinsic mode function IMF components:
[0063]
[0064] Among them, The coke oven gas generation data after smoothing is represented by z i (t) represents the i-th intrinsic mode function component, r(t) is a residual term, and M is the total number of mode components.
[0065] Reference Figure 2 , Figure 2 The 11 intrinsic mode functions obtained by the EEMD method in the embodiment of the application can be observed to have the characteristics of high frequency noise of IMF1 and IMF2. The average instantaneous frequency of each intrinsic mode function component is obtained by Hilbert transform method.
[0066] The coke oven gas generation IMF component obtained by decomposition is used to verify the multi-time scale characteristic, and the analytic signal thereof is obtained by using Hilbert transform:
[0067]
[0068] Among them, z i (t) represents the i-th intrinsic mode function, H() represents Hilbert transform, i represents the i-th, A i (t) represents the amplitude envelope of the i-th IMF component, j represents an imaginary unit, and θ i (t) represents the instantaneous phase of the i-th IMF component.
[0069] After the coke oven gas generation IMF component is converted into the corresponding analytic signal, the effectiveness of the decomposition can be verified by calculating the average instantaneous frequency of each component. Assuming that the analysis interval is [t1, t2], the average instantaneous frequency of the i-th IMF component is:
[0070]
[0071] Among them, The average instantaneous frequency of the i-th IMF component is represented.
[0072] Reference Figure 3 , Figure 3 The average instantaneous frequency of the 11 intrinsic mode functions is calculated by the Hilbert transform, and the numbers 1-11 on the abscissa represent the serial numbers of the intrinsic mode functions. It can be observed that the average instantaneous frequencies of IMF1 and IMF2 are 2.21 Hz and 1.54 Hz, respectively, which are much higher than those of other intrinsic mode functions, so they are classified as high-frequency noise signals. IMF1 and IMF2 are removed from the 11 intrinsic mode functions, and only the remaining 9 intrinsic mode functions are left for the following steps.
[0073] S3, complexity evaluation of each intrinsic mode function component is performed by calculating sample entropy, and each intrinsic mode function component is reconstructed into a periodic component and a trend component according to the complexity of the intrinsic mode function component.
[0074] Due to the complexity and variability of the coking process, the noise in the original coke oven gas generation data cannot be completely removed by the Savitzky-Golay filter and EEMD decomposition, and the proportion of noise in each component is inconsistent, resulting in differences in the order of each component. Therefore, the complexity of each intrinsic mode function component is evaluated by calculating the sample entropy according to the following formula:
[0075]
[0076] Wherein, SampEn(m, r, N) represents the sample entropy with parameters (m, r, N), m represents the dimension of the template vector, r represents the similarity tolerance, N represents the number of intrinsic mode function components, A m (r) represents the probability that two sequences of the template vector match m+1 points, B m (r) represents the probability that two sequences of the template vector match m points under the similarity tolerance r, A i represents the number of pairs of template vectors with a length of m+1 whose distance is less than r, B i represents the number of pairs of template vectors with a length of m whose distance is less than r.
[0077] The components with sample entropy values between 0.6 and 1 are added together as periodic components. The frequency of the periodic component is relatively low compared to the high-frequency noise component, and it has regular and periodic fluctuations, but the fluctuation aggregation characteristics are more obvious compared to the trend component. The components with sample entropy values between 0 and 0.6 are added together as trend components, and the trend component reflects the long-term trend change of the coke oven gas generation. Figure 4 The reconstruction results of the intrinsic mode function components of the embodiment of the present application are shown in
[0078]
[0079] S4, based on the periodic component, the trend component and the process parameters affecting the coke oven gas generation, the key process parameters affecting the coke oven gas generation are obtained by calculating the mutual information coefficient.
[0080] According to the mechanism analysis, the coke oven gas generation is mainly affected by the coking state and the volatile matter of the blended coal. The coking state is divided as follows:
[0081]
[0082] Wherein, J=t / r, t is the interval of the actual coal charging time and the current time, and r is the coking period of the coke oven.
[0083] Based on the size of the mutual information coefficient, the key process parameters affecting the coke oven gas generation are selected. In the embodiment of the present application, the mutual information coefficient is used to analyze the relationship between each component and the gas generation, and the strategy used is as follows:
[0084]
[0085] Wherein, I(y i ; x k ) represents the mutual information coefficient of y i and x k , y i represents the i th periodic component or trend component signal, x k represents the k th process parameter, p(y i , x k ) represents the joint probability of y i and x k , p(y i ) represents the probability of y i , and p(x k ) represents the probability of x k .
[0086] In the embodiment of the present application, the calculation results of the periodic component and the trend component selected by calculating the mutual information coefficient are as shown in Figure 5 Figure 5 (a) represents the calculation result of the periodic component selected by calculating the mutual information coefficient, and (b) represents the calculation result of the trend component selected by calculating the mutual information coefficient. adf1#2# V adf3# represents the volatile matter of the blended coal of the No. 1 and No. 2 combined coke oven, V o1#2# represents the volatile matter of the blended coal of the No. 3 coke oven, N o3# represents the number of ovens in the initial coking stage of the No. 1 and No. 2 combined coke oven, N m1#2# N represents the number of the combined coke ovens of No. 1 and No. 2 in the middle stage of coking m3# N represents the number of the combined coke ovens of No. 1 and No. 2 in the middle stage of coking e1#2# N represents the number of the combined coke ovens of No. 1 and No. 2 in the middle stage of coking e3# N represents the number of the combined coke ovens of No. 1 and No. 2 in the middle stage of coking adf3# N represents the number of the combined coke ovens of No. 1 and No. 2 in the middle stage of coking e3# The mutual information coefficient values of N adf1#2# , the number of the combined coke ovens of No. 1 and No. 2 in the middle stage of coking, and N m3# , the number of the combined coke ovens of No. 3 in the middle stage of coking, and N e3# , the number of the combined coke ovens of No. 3 in the end stage of coking, are 0.2608, 0.4238 and 0.2293, respectively.
[0087] Since the coking process has a long time delay characteristic, the current state is greatly affected by the historical state. In order to improve the accuracy of prediction, the method obtains the lag period of the input variable of the model by calculating the Bayesian information criterion. The optimal lag period corresponds to the minimum value of the Bayesian criterion, which is represented as:
[0088] BIC = k·lnn-2lnL,
[0089] wherein k is the number of free parameters in the model, L is the maximum likelihood estimation of the model on the coke oven gas generation data, n is the sample size, BIC represents the Bayesian criterion, and the optimal lag period is obtained when the BIC reaches the minimum value.
[0090] The calculation result is shown in Table 1, Figure 6 Figure 6 which is the input variable lag period result of the cycle component and the trend component in the embodiment of the present application. The cycle component selects 14 lag periods as the input of the prediction model, and the trend component selects 7 lag periods as the input of the prediction model.
[0091] S5, using a long short-term memory network (LSTM) with an added attention mechanism (Attention) as a prediction model (LSTM-Attention), obtaining the optimal lag period of the model input through the Bayesian information criterion, taking the cycle component, the trend component, the key process parameters and the optimal lag period as the input variables of the prediction model, and predicting the coke oven gas generation. Reference Figure 7 , Figure 7 is a comparison chart of the coke oven gas generation amount prediction result based on the LSTM-Attention model and the actual coke oven gas generation amount of an embodiment of the present application.
[0092] The attention mechanism is introduced to self-learn the weight alpha of each hidden state h t of the LSTM t , and the context vector c is formed by weighting and combining all historical states according to the weight, so that the model focuses on the key moment, improves the interpretability and accuracy of the prediction, and enhances the understanding ability of the model to the operation law of the coke oven.
[0093] e t = v Τ tanh(Wh t +b)
[0094]
[0095] Wherein, v and W are weight parameters, b is a bias parameter, e t is the attention score of the hidden state h t , T represents the number of hidden states, and e j is the attention score of the hidden state h j .
[0096] In order to show the prediction effect of the present method (LSTM-Attention), root meanapuare error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) are used to measure the prediction effect, and the experimental results are shown in Table 1.
[0097] Table 1: Prediction error indicators of coke oven gas generation amount
[0098] Model RMSE MAE MAPE LSTM-Attention 233.82 319.46 0.27％
[0099] In an exemplary embodiment, a computer readable storage medium is included, and the computer readable storage medium stores a computer program, which is executed by a processor to realize the coke oven gas generation amount prediction method based on multi-time scale decomposition described above.
[0100] Please refer to Figure 8 , in an exemplary embodiment, an electronic device is also included, which comprises at least one processor, at least one memory, and at least one communication bus.
[0101] The computer program is stored in the memory, and includes computer readable instructions. The processor invokes the computer readable instructions stored in the memory to execute the coke oven gas generation amount prediction method based on multi-time scale decomposition.
[0102] In an exemplary embodiment, a computer program product is also included, comprising computer program / instructions, wherein the computer program / instructions, when executed by the processor, implement the steps of the coke oven gas generation amount prediction method based on multi-time scale decomposition.
[0103] The above description of disclosed embodiments enables one of ordinary skill in the art to make and use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A coke oven gas generation amount prediction method based on multi-time scale decomposition, characterized by, The method comprises the following steps: S1, collecting historical data of coke oven gas generation amount and process parameters affecting the coke oven gas generation amount, and performing smoothing processing on the coke oven gas generation amount data; S2, decomposing the processed coke oven gas generation amount data using EEMD to obtain multiple intrinsic mode function components, and removing high-frequency noise signals in the intrinsic mode function components by calculating the average instantaneous frequency of each intrinsic mode function component through Hilbert transform; S3, evaluating the complexity of each intrinsic mode function component by calculating sample entropy, and reconstructing the intrinsic mode function components into periodic components and trend components according to the complexity of the intrinsic mode function components; S4, obtaining key process parameters affecting the coke oven gas generation amount based on the periodic components, trend components and process parameters affecting the coke oven gas generation amount, and calculating the mutual information coefficient; S5, using an LSTM with an attention mechanism as a prediction model, obtaining an optimal lag period of the input of the model through the Bayesian information criterion, taking the periodic components, trend components, key process parameters and optimal lag period as input variables of the prediction model, and predicting the coke oven gas generation amount.
2. The coke oven gas generation amount prediction method based on multi-time scale decomposition according to claim 1, characterized in that, The coke oven gas generation amount data is smoothed using a Savitzky-Golay filter, and the target of the filtering is to minimize the following objective function at each point: where a0,...,a k denote the coefficient vector, a l denotes the l-th coefficient vector, j denotes the offset position of each point in the sliding window relative to the current center point, y i+j denotes the j-th coke oven gas generation data in the window with the center point y i , and [-m, m] denotes a sliding window with a length of 2m+1.
3. The coke oven gas generation amount prediction method based on multi-time scale decomposition according to claim 1, characterized in that, The coke oven gas generation amount data after smoothing is decomposed using the EEMD method to obtain a group of intrinsic mode function IMF components: wherein denotes the smoothed coke oven gas generation amount data, IMF i (t) denotes the i-th intrinsic mode function component, r(t) is the residual term, and M is the total number of mode components. The average instantaneous frequency of each intrinsic mode function component is calculated through Hilbert transform and is expressed as: wherein z i (t) represents the analytic signal of the i-th intrinsic mode function, H() represents the Hilbert transform, i represents the i-th, A i (t) represents the amplitude envelope of the i-th IMF component, j represents the imaginary unit, θ i (t) represents the instantaneous phase of the i-th IMF component; Let the analysis interval be [t1, t2], and the average instantaneous frequency of the i-th component is: wherein, denotes the average instantaneous frequency of the i-th component.
4. The coke oven gas generation amount prediction method based on multi-time scale decomposition according to claim 1, characterized in that, S3 is specifically: The sample entropy corresponding to each component is calculated: where SampEn(m, r, N) represents the sample entropy with parameters (m, r, N), m represents the dimension of the template vector, r represents the similarity tolerance, and N represents the number of intrinsic mode function components, A m (r) represents the probability that the two sequences of the template vector match m+1 points, B m (r) represents the probability that the two sequences of the template vector match m points under the similarity tolerance r, A i represents the number of pairs of template vectors of length m+1 whose distance is less than r, B i represents the number of pairs of template vectors of length m whose distance is less than r; The intrinsic mode function components are reconstructed into periodic components and trend components according to the size of the sample entropy:
5. The coke oven gas generation amount prediction method based on multi-time scale decomposition according to claim 1, characterized in that, Based on the periodic components, trend components and process parameters affecting the coke oven gas generation amount, the key process parameters affecting the coke oven gas generation amount are obtained by calculating the mutual information coefficient, and are expressed as: where I(y i ; x k ) denotes the mutual information coefficient of y i and x k , y i denotes the i-th periodic component or trend component signal, x k denotes the k-th process parameter, p(y i , x k ) denotes the joint probability of y i and x k , p(y i ) denotes the probability of y i , and p(x k ) denotes the probability of x k ; Based on the size of the mutual information coefficient, the key process parameters affecting the coke oven gas generation amount are selected.
6. The coke oven gas generation amount prediction method based on multi-time scale decomposition according to claim 5, characterized in that, The key process parameters affecting the coke oven gas generation amount are: coking state and volatile matter of blended coal; The coking state is divided as follows: Where t is the interval between the actual coal charging time and the current time, and r is the coking period of the coke oven.
7. The coke oven gas generation amount prediction method based on multi-time scale decomposition according to claim 1, characterized in that, The optimal lag period of the input of the model is obtained through the Bayesian information criterion and is expressed as: BIC=k·lnn-2lnL, Where k is the number of free parameters in the model, L is the maximum likelihood estimate of the model on the coke oven gas generation amount data, n is the sample size, BIC represents the Bayesian criterion, and the optimal lag period is obtained when the BIC reaches the minimum value.
8. A computer readable storage medium storing a computer program, characterized in that: The computer program is executed by the processor to implement the method of any one of claims 1-7.
9. An electronic device, comprising: The computer program comprises computer readable instructions, and the processor is configured to invoke the computer readable instructions to execute the method of any one of claims 1-7.
10. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the steps of the method of any one of claims 1-7.