Early warning method for difficult coke pushing

Abstract

The invention provides an early warning method for difficult coke pushing, which comprises the following steps: S100: preprocessing historical data of coke pushing current of a coke oven to obtain preprocessed historical data of the coke pushing current; s200, analyzing the preprocessed coke pushing current historical data based on requirements of an SARIMA model, and then determining model parameters of the SARIMA model to obtain a prediction model; s300, performing reverse normalization processing on an output result based on the prediction model to obtain a coke pushing current prediction value; and S400, based on the predicted value of the coke pushing current, combining historical data and a preset rule to realize early warning of difficult coke pushing. According to the early warning method for difficult coke pushing, the coking coke pushing production rule is considered, the irregular periodic characteristics of the coke pushing current data are fully excavated, accurate prediction is carried out by adopting the SARIMA time sequence prediction method, and compared with the prior art, the model prediction precision is higher, the training time is short, and the prediction efficiency is high.

Description

Technical Field

[0001] The invention relates to the field of coking, and in particular to an early warning method for difficult coking. Background Art

[0002] Difficult coke pushing means that the current of the coke pushing motor exceeds the maximum current specified by the motor, or the coke pushing is difficult and cannot be done smoothly, which disrupts the normal heating of the coke oven, easily damages the hot oven masonry, shortens the service life of the coke oven, easily damages the coke oven machinery, and affects the output and quality of coke.

[0003] The coking process involves many physical and chemical reactions and has a complex mechanism. Although the main influencing factors of the coke pushing current can be determined through mechanism analysis, it is difficult to determine the degree of influence of these main influencing factors on the coke pushing current, and the factors will affect each other, and the specific degree of interaction is also difficult to determine. Therefore, although the existing prediction methods for difficult coke pushing consider the various factors that affect the coke pushing current and can give certain prediction results, they do not consider the weight of the influence of each factor on the coke pushing current and the interaction between the factors. In addition, the prediction model established by deep learning methods such as artificial neural networks is difficult to obtain the optimal parameters, the training time is long, and the prediction results are accidental, which makes it difficult for the calculation efficiency and prediction accuracy of the prediction model to meet the needs of coking production. Summary of the invention

[0004] The main purpose of the present invention is to provide a difficult-to-focus early warning method to solve the technical problems of insufficient computational efficiency and prediction accuracy of the prediction model.

[0005] To achieve the above object, the present invention provides an early warning method for difficult focus pushing, comprising the following steps:

[0006] S100: Preprocessing the historical data of the coke pushing current of the coke oven to obtain preprocessed historical data of the coke pushing current.

[0007] S200: Analyze the pre-processed push-focus current historical data based on the requirements of the SARIMA model, and then determine the model parameters of the SARIMA model to obtain a prediction model. The model parameters include the order p of the autoregressive term, the order d of the difference, the order q of the moving average term, the order P of the seasonal autoregressive term, the order D of the seasonal difference, the order Q of the seasonal moving average term, and the seasonal cycle length s.

[0008] S300: Based on the prediction model, the output result is denormalized to obtain the predicted value of the push-focus current.

[0009] S400: Based on the predicted value of the pushing current, combined with historical data and preset rules, a difficult pushing warning is implemented.

[0010] According to an embodiment of the present application, step 100 includes:

[0011] Perform at least one of missing value interpolation, outlier elimination, and normalization on the historical data of the push-focus current.

[0012] According to an embodiment of the present application, the step of interpolating missing values ​​of the historical data of the push-focus current includes:

[0013] The missing values ​​in the historical data of the push-focus current are processed by the mean filling method.

[0014] The step of normalizing the historical data of the push-focus current comprises:

[0015] The Min-Max normalization method is used to normalize the historical data of the push-coke current after missing value interpolation and outlier removal. The normalization formula is as follows:

[0016]

[0017] Among them, x N is the expected value of the normalized historical data of the push current; min is the minimum value in the historical data of the push-focus current; x max

[0018] is the maximum value in the historical data of the push-focus current; x is any data in the historical data of the push-focus current.

[0019] According to an embodiment of the present application, step 200 includes:

[0020] S210: The pre-processed focus current historical data is converted into corresponding stable first time series and second time series through seasonal differential transformation and non-seasonal differential transformation respectively. The first time series and the second time series then obtain the differential order d and the seasonal differential order D based on the number of seasonal differential transformations and the number of non-seasonal differential transformations respectively.

[0021] S220: Determine the optimal values ​​of the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term and the order Q of the seasonal moving average term through the Bayesian Information Criterion BIC parameter optimization algorithm.

[0022] S230: Perform STL decomposition on the second time series, and determine the value of the focusing current period S in combination with the irregular periodicity of the focusing operation in actual production.

[0023] According to an embodiment of the present application, the step of determining the optimal values ​​of the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term, and the order Q of the seasonal moving average term by using the Bayesian Information Criterion BIC parameter optimization algorithm includes:

[0024] The calculation formula of BIC is as follows:

[0025] BIC=-2*ln(L)+k*ln(n)

[0026] Among them, L is the maximum likelihood value of the model, k is the number of parameters of the model, and n is the sample size of the data.

[0027] According to an embodiment of the present application, the step of performing STL decomposition on the second time series and determining the value of the coke pushing current period S in combination with the irregular periodicity of the actual production coke pushing operation includes:

[0028] S231: Detrending the second time series:

[0029] in, is the second time series after detrending, Y t is the second time series, is the result of the ith iteration of the trend component. i is the number of iterations.

[0030] S232: Smoothing each subsequence using the local weighted regression LOWESS algorithm, and then extending it by one period each before and after, to obtain a new subsequence

[0031] S233: Calculate the trend component of the atomic sequence using low-pass filtering Remove the periodic differences and then calculate the seasonal component of the new time series by the additivity of the time series The calculation formula is:

[0032]

[0033] S234: Remove the seasonal component in the subsequence to obtain the seasonal component of the second time series and trend component The calculation formula is:

[0034]

[0035]

[0036] S235: Determine whether it converges. If not, repeat the process of steps S231 to S234. If it converges, t, the seasonal component of the second time series Trend Component Get the residual component of the time series

[0037] According to an embodiment of the present application, according to the second time series Y t , the seasonal component of the second time series Trend Component Get the residual component of the time series The calculation formula is:

[0038]

[0039] According to an embodiment of the present application, between steps S200 and S300, the following steps are further included:

[0040] The standard residuals are used to test the residual normality of the SARIMA model to determine whether the residuals obey the normal distribution.

[0041] The Durbin-Watson method is used to test the residual correlation of the SARIMA model to determine whether there is a first-order correlation in the residuals.

[0042] After the SARIMA model passes the residual normality test and the residual correlation test, step S300 is performed.

[0043] According to an embodiment of the present application, step S400 includes:

[0044] The preset rules are constructed by using expert experience analysis method.

[0045] In the above-mentioned early warning method for difficult coke pushing, the production rules of coke pushing in coking are taken into consideration, the irregular periodic characteristics of the coke pushing current data itself are fully explored, and the SARIMA time series prediction method is used for accurate prediction. Compared with the existing technical solutions, the model has higher prediction accuracy, short training time and high prediction efficiency. 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 the structures shown in these drawings without paying creative work.

[0047] Figure 1 It is a flow chart of a difficult-to-push-focus early warning method according to an embodiment of the present application;

[0048] Figure 2is a flowchart of step S200 of an implementation mode of the present application;

[0049] Figure 3 It is a comparison chart between the predicted value and the actual value of the push-focus current in one embodiment of the present application.

[0050] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with the implementation methods and with reference to the accompanying drawings. 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] It should be noted that all directional indications (such as up, down, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0053] In addition, in the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features.

[0054] Furthermore, the technical solutions between the various embodiments of the present invention may be combined with each other, but this must be based on the fact that they can be implemented by ordinary technicians in the field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0055] The applicant has conducted extensive research on the production rules of coke pushing in coke making. Considering that the coke pushing current is relatively stable during the data collection time of each coke oven coking process, and affected by the degree of repair of the coke oven carbonization chamber, the carbonization chamber will have difficulty in pushing coke every certain period of time, that is, the peak value of the coke pushing current of different furnaces will exceed the normal threshold every certain period, so there are obvious non-stationary and periodic time series characteristics. In view of this feature, the SARIMA time series prediction method is used to mine the periodic characteristics of the coke pushing current data itself to predict the coke pushing current of future furnaces, so as to give an early warning of difficult coke pushing.

[0056] Based on this, the present invention provides an early warning method for difficult focus pushing, see Figure 1, including the following steps:

[0057] S100: Preprocessing the historical data of the coke pushing current of the coke oven to obtain preprocessed historical data of the coke pushing current.

[0058] Coke pushing current is an important parameter in the coke production process, generally referring to the maximum current in the coke pushing process. The internal cause of the change of coke pushing current is that there is a certain linear relationship between coke pushing resistance and coke pushing current, so the change of coke pushing current can be studied by studying the change of resistance.

[0059] The historical data of the push-focus current may be a collection of the push-focus current in the past time, such as a collection of the push-focus current in the past year or several months.

[0060] The historical data of the focus current is preprocessed to remove invalid, repeated or redundant data, thereby improving the quality and accuracy of the data or reducing the risks of overfitting and underfitting.

[0061] Therefore, in some embodiments, step 100 includes: performing at least one of missing value interpolation, outlier elimination, and normalization on the historical data of the focus current.

[0062] The historical data of the current of the push-coke is processed by one or more of the following processes: missing value interpolation, outlier elimination, and normalization, as required. For example, the historical data of the current of the push-coke is processed by three processes: missing value interpolation, outlier elimination, and normalization.

[0063] Outliers and missing values ​​will affect the final fitting results and therefore need to be processed accordingly.

[0064] Outliers refer to data points in a data set that do not conform to the general rules of the data or exceed the standard limits. Due to measurement errors, data input errors, abnormal data changes, etc., outliers may appear in the historical data of the current. Missing values ​​refer to the situation where the values ​​of certain attributes or features in the historical data of the current are missing or empty. Methods for processing missing values ​​of the current data include nearest neighbor filling, random forest interpolation, and mean filling.

[0065] In some specific embodiments, the step of interpolating missing values ​​of the historical data of the push-focus current includes:

[0066] The missing values ​​in the historical data of the push-focus current are processed by the mean filling method.

[0067] The main steps of the mean imputation method include: 1. Calculate the mean of all non-missing data. 2. Use the calculated mean to fill in the missing values.

[0068] The mean imputation method replaces missing values ​​with the average value of the feature. Its operation is simple and convenient. The mean imputation method is used to process abnormal data and missing values, which improves the availability and accuracy of the data set compared with the existing technical solutions.

[0069] In some specific embodiments, the step of normalizing the historical data of the push-focus current includes:

[0070] The Min-Max normalization method is used to normalize the historical data of the push-coke current after missing value interpolation and outlier removal. The normalization formula is as follows:

[0071]

[0072] Among them, x N is the expected value of the normalized historical data of the push current; min is the minimum value in the historical data of the push-focus current; x max

[0073] is the maximum value in the historical data of the push-focus current; x is any data in the historical data of the push-focus current.

[0074] Min-Max normalization can convert all data in the historical data of the current of the focus after missing value interpolation and outlier removal to between 0 and 1. This method converts each data point into a relative value relative to the maximum and minimum values ​​based on the maximum and minimum values ​​of the original data. For example, if the maximum value in the original data is 100 and the minimum value is 0, then for the data point with an original value of 50, the normalized value is 0.5.

[0075] S200: Analyze the pre-processed push-focus current historical data based on the requirements of the SARIMA model, and then determine the model parameters of the SARIMA model to obtain a prediction model. The model parameters include the order p of the autoregressive term, the order d of the difference, the order q of the moving average term, the order P of the seasonal autoregressive term, the order D of the seasonal difference, the order Q of the seasonal moving average term, and the seasonal cycle length s.

[0076] SARIMA model is the abbreviation of Seasonal Autoregressive Integrated Moving Average model, which is a time series forecasting model that can handle data with seasonality and trend.

[0077] The meanings of the parameters in SARIMA(p,d,q)(P,D,Q)s are as follows:

[0078] p: The order of the autoregressive term. An autoregressive model is a linear model that predicts future values ​​by using past values ​​of a time series as independent variables to predict future values. The order p represents the number of past values ​​used in the autoregressive model.

[0079] d: The order of differencing. Difference is a method used to eliminate seasonality and trend in time series data by comparing the observations of the time series with their previous observations. The order d indicates how many times the difference operation needs to be performed.

[0080] q: The order of the moving average term. A moving average model is a smoothing model that predicts future values ​​by using the average of observations from past time periods. The order q represents the number of past observations used in the moving average model.

[0081] P: The order of the seasonal autoregressive term. This is a parameter added to take into account the seasonal effects of the time series. The number of past seasonal observations used in the seasonal autoregressive model is the order P.

[0082] D: Order of seasonal difference. This is a parameter added to take into account the seasonal effect of the time series. Seasonal difference transformation is used to eliminate the seasonal effect of time series data. The order D indicates how many times the seasonal difference transformation operation needs to be performed.

[0083] Q: The order of the seasonal moving average term. This is a parameter added to take into account the seasonal effects of the time series. The seasonal moving average model is used to further smooth the time series data and reduce the seasonal fluctuations of the data. The order Q represents the number of past seasonal observations used in the seasonal moving average model.

[0084] s: Seasonal cycle length. This is a parameter added to take into account the seasonal effects of the time series. It represents the length of the cycle for each season in the time series of the data, for example, if the data is collected hourly, the seasonal cycle length may be 24 (a day).

[0085] The selection of these parameters needs to be determined based on the characteristics of the actual data and the prediction requirements. That is, the pre-processed push-focus current historical data is analyzed based on the requirements of the SARIMA model to determine the above parameters and thus determine the SARIMA model.

[0086] In some embodiments, see Figure 2 , step 200 comprises:

[0087] S210: The pre-processed focus current historical data are converted into corresponding stable time series through seasonal difference transformation and non-seasonal difference transformation respectively, and then the difference order d and seasonal difference order D are obtained based on the number of seasonal difference transformation and the number of non-seasonal difference transformation respectively.

[0088] The pre-processed push-coke current historical data is converted into a first time series through seasonal difference transformation. The first time series obtains a difference order d based on the number of seasonal difference transformations.

[0089] The pre-processed push-focus current historical data is converted into a second time series through non-seasonal difference transformation. The second time series obtains the difference order D based on the number of non-seasonal difference transformations.

[0090] The purpose of this step is to perform stabilization processing (seasonal difference transformation and non-seasonal difference transformation) on the pre-processed historical data of the push-focus current to convert it into a stable time series, and obtain the values ​​of the difference d and D through the number of stabilization processes.

[0091] The SARIMA model stabilizes the data to make it meet the requirements of time series analysis. Specifically, the SARIMA model requires the data to be stationary, that is, the statistical characteristics of the data remain constant or change slowly over time. If the data is not stationary, the SARIMA model may be biased, resulting in a decrease in prediction accuracy.

[0092] In order to make the data meet the requirements of stationarity, the pre-processed historical data of the current can be subjected to non-seasonal difference transformation or seasonal difference transformation to eliminate trends and seasonal changes.

[0093] Seasonal difference transformation divides the time series into segments according to seasons and differentiates each segment of data to eliminate seasonal changes.

[0094] Through stabilization processing, the pre-processed push-coke current historical data can meet the requirements of the SARIMA model and improve the prediction accuracy and stability of the model. At the same time, the outliers and noise in the pre-processed push-coke current historical data can also be eliminated to make the data smoother and more stable.

[0095] S220: Determine the optimal values ​​of the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term and the order Q of the seasonal moving average term through the Bayesian Information Criterion BIC parameter optimization algorithm.

[0096] Among them, Bayesian Information Criterion (BIC) is a statistical method based on the Bayesian school of thought, which is used to evaluate the quality of the model. It evaluates the model by calculating the Bayesian risk of the model, which can balance the complexity and fit of the model and avoid overfitting.

[0097] In some embodiments, the step of determining the optimal values ​​of the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term, and the order Q of the seasonal moving average term by using the Bayesian Information Criterion BIC parameter optimization algorithm comprises:

[0098] The calculation formula of BIC is as follows:

[0099] BI C=-2*ln(L)+k*ln(n)

[0100] Among them, L is the maximum likelihood value of the model, k is the number of parameters of the model, and n is the sample size of the data. Generally speaking, the smaller the BIC value, the better the model.

[0101] In other embodiments, the Akaike information criterion (AIC) is used to determine the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term, and the order Q of the seasonal moving average term.

[0102] S230: Perform STL decomposition on the second time series, and determine the value of the focusing current period S in combination with the irregular periodicity of the focusing operation in actual production.

[0103] The main principle of the STL algorithm decomposition process is to decompose the data at a certain moment into trend component, periodic component and remainder based on LOESS (locally weighted scatter plot smoothing).

[0104] In some embodiments, the step of performing STL decomposition on the second time series and determining the value of the coke pushing current period S in combination with the irregular periodicity of the actual production coke pushing operation includes:

[0105] S231: Detrending the second time series:

[0106] in, is the second time series after detrending, Y t is the second time series. The second time series is not detrended and can be called the original series.

[0107] is the result of the ith iteration of the trend component. i is the number of iterations.

[0108] Detrending is a technique used to remove trends or cyclical changes from time series data in order to better analyze fluctuations and outliers in the data.

[0109] In the detrending process, a subsequence refers to a sequence of continuous data points selected from the original time series data. The length of the subsequence can be selected according to actual needs and data characteristics, and usually requires experiments and adjustments based on actual conditions.

[0110] In the detrending process, the trend and periodic changes in the data can be eliminated by segmenting the original time series data and fitting a trend line or smooth curve to each segment. In this process, each segment of data can be regarded as a subsequence. By fitting or smoothing the subsequence, the time series data after trend removal can be obtained, which is convenient for further data analysis and processing.

[0111] S232: Smoothing each subsequence using the local weighted regression LOWESS algorithm, and then extending it by one period each before and after, to obtain a new subsequence

[0112] S233: Calculate the trend component of the atomic sequence using low-pass filtering Remove the periodic differences and then calculate the seasonal component of the new time series by the additivity of the time series The calculation formula is:

[0113]

[0114] S234: Remove the seasonal component in the subsequence to obtain the seasonal component of the original sequence and trend component The calculation formula is:

[0115]

[0116]

[0117] S235: Determine whether it converges. If not, repeat the process of steps S231 to S234. If it converges, t , the seasonal component of the original series Trend Component Get the residual component of the time series The original sequence is the second time series Y t .

[0118] In some embodiments, according to the second time series Y t , the seasonal component of the original series Trend Component Get the residual component of the time series The calculation formula is:

[0119]

[0120] The original sequence is the second time series Y t .

[0121] In some other embodiments, the schemes for decomposing the current time series data include the classical decomposition method, the X11 decomposition method, etc.

[0122] In step S200, the order p of the autoregressive term, the order d of the difference, the order q of the moving average term, the order P of the seasonal autoregressive term, the order D of the seasonal difference, the order Q of the seasonal moving average term and the length s of the seasonal cycle are determined, thereby determining the push current prediction model SARIMA (p, d, q) (P, D, Q) S .

[0123] S300: Based on the prediction model, the output result is denormalized to obtain the predicted value of the push-focus current.

[0124] Since the preprocessed push current historical data is normalized, the output results of the prediction model need to be denormalized to obtain the push current prediction value. That is, the normalized results are converted back to the original data range, which can return the data to the original measurement scale for subsequent analysis and interpretation.

[0125] Denormalization methods include but are not limited to the following two:

[0126] Inverse mapping method: This method is to fit the original data with the corresponding standard curve or polynomial curve to obtain a mapping curve, and then restore the original data through this curve.

[0127] Comparative analysis method: This method is to compare and analyze the normalized data with the original data to find out the differences and connections between the two, so as to restore the normalized data to the original data.

[0128] S400: Based on the predicted value of the pushing current, combined with historical data and preset rules, a difficult pushing warning is implemented.

[0129] The preset rules are constructed by expert experience analysis. Based on the predicted value of the coke pushing current, combined with historical data and preset rules, a coke pushing current warning rule is established, and the production site determines whether the carbonization chamber of the future furnace will have a difficult coke pushing situation based on the predicted value.

[0130] In some embodiments, between steps S200 and S300, the following steps are further included:

[0131] The standard residuals are used to test the residual normality of the SARIMA model to determine whether the residuals obey the normal distribution.

[0132] The Durbin-Watson method is used to test the residual correlation of the SARIMA model to determine whether there is a first-order correlation in the residuals.

[0133] After the SARIMA model passes the residual normality test and the residual correlation test, step S300 is performed.

[0134] In the above-mentioned early warning method for difficult coke pushing, the production rules of coke pushing in coke making are taken into consideration, the irregular periodic characteristics of the coke pushing current data itself are fully explored, and the traditional SARIMA time series prediction method is used for accurate prediction. Compared with the existing technical solutions, the model has higher prediction accuracy, short training time and high prediction efficiency.

[0135] The above-mentioned early warning method for difficult-to-push focus includes at least the following beneficial effects:

[0136] (1) The SARIMA model-based early warning method for difficult focus pushing of the present invention uses a mean filling method to process abnormal data and missing values, thereby improving the availability and accuracy of the data set compared with the prior art solutions.

[0137] (2) The SARIMA model-based early warning method for difficult coke pushing of the present invention takes into account the production rules of coke pushing in coke making, fully exploits the irregular periodic characteristics of the coke pushing current data itself, and uses the traditional mathematical model prediction method for accurate prediction. Compared with the existing technical solutions, the model has higher prediction accuracy, short training time and high prediction efficiency.

[0138] (3) The SARIMA model-based difficult coke-pushing early warning method of the present invention determines the difficult coke-pushing condition in the carbonization chamber based on historical data and preset rules.

[0139] The SARIMA model-based early warning method for difficult-to-push focus is used to predict the predicted focus current values ​​of 0 to 50 times, and then the predicted values ​​are compared with the actual focus current peak values. The results are as follows: Figure 3 The interval between each shot is 18.5 hours, where 18.5 is the actual focus time of each shot.

[0140] pass Figure 3 It can be seen from the data that the predicted value of the pushing current obtained by the difficult-pushing warning method based on the SARIMA model is highly consistent with the actual pushing current peak data and has high accuracy, which can meet the requirements of industrial production.

[0141] The above technical solutions of the present invention are only preferred embodiments of the present invention, and the patent scope of the present invention is not limited thereto. All equivalent structural changes made by using the contents of the present invention specification and drawings under the technical concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for early warning of difficult focus pushing, characterized in that: The following steps are involved: S100: preprocessing the historical data of the coke pushing current of the coke oven to obtain preprocessed historical data of the coke pushing current; S200: Analyze the pre-processed push-focus current historical data based on the requirements of the SARIMA model, and then determine the model parameters of the SARIMA model to obtain a prediction model; the model parameters include the order p of the autoregressive term, the order d of the difference, the order q of the moving average term, the order P of the seasonal autoregressive term, the order D of the seasonal difference, the order Q of the seasonal moving average term and the length s of the seasonal cycle; S300: performing a denormalization process on the output result based on the prediction model to obtain a predicted value of the push current; S400: Based on the predicted value of the pushing current, combined with historical data and preset rules, a difficult pushing warning is implemented.

2. The early warning method according to claim 1, characterized in that: The step S100 includes: Perform at least one of missing value interpolation, outlier removal, and normalization on the historical data of the push-focus current.

3. The early warning method according to claim 1, characterized in that: The step of interpolating missing values ​​of the historical data of the push-focus current comprises: The missing values ​​in the historical data of the push-focus current are processed by using a mean filling method; The step of normalizing the historical data of the push-focus current comprises: The Min-Max normalization method is used to normalize the historical data of the push-coke current after missing value interpolation and outlier removal. The normalization formula is as follows: Among them, x N is the expected value of the normalized historical data of the push current; min is the minimum value in the historical data of the push-focus current; x max is the maximum value in the historical data of the push-focus current; x is any data in the historical data of the push-focus current.

4. The early warning method according to claim 1, characterized in that: The step S200 includes: S210: converting the pre-processed push-coke current historical data into corresponding stable first time series and second time series through seasonal difference transformation and non-seasonal difference transformation respectively, and obtaining the difference order d and seasonal difference order D of the first time series and the second time series respectively based on the number of seasonal difference transformation and the number of non-seasonal difference transformation; S220: Determine the optimal values ​​of the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term, and the order Q of the seasonal moving average term by using the Bayesian Information Criterion BIC parameter optimization algorithm; S230: Perform STL decomposition on the second time series, and determine the value of the focusing current period S in combination with the irregular periodicity of the focusing operation in actual production.

5. The early warning method according to claim 4, characterized in that: The step of determining the optimal values ​​of the order p of the autoregressive term, the order q of the moving average term, the order P of the seasonal autoregressive term, and the order Q of the seasonal moving average term by using the Bayesian Information Criterion BIC parameter optimization algorithm comprises: The calculation formula of BIC is as follows: BIC=-2*ln(L)+k*ln(n) Among them, L is the maximum likelihood value of the model, k is the number of parameters of the model, and n is the sample size of the data.

6. The early warning method according to claim 4, characterized in that: The step of performing STL decomposition on the second time series and determining the value of the coke pushing current period S in combination with the irregular periodicity of the actual coke pushing operation comprises: S231: Detrending the second time series: in, is the second time series after detrending, Y t is the second time series, is the result of the ith iteration of the trend component; i is the number of iterations; S232: Smoothing each subsequence using the local weighted regression LOWESS algorithm, and then extending it by one period each before and after, to obtain a new subsequence S233: Calculate the trend component of the atomic sequence using low-pass filtering Remove the periodic differences and then calculate the seasonal component of the new time series by the additivity of the time series The calculation formula is: S234: Remove the seasonal component in the subsequence to obtain the seasonal component of the second time series and trend component The calculation formula is: S235: Determine whether it converges. If not, repeat the process of steps S231 to S234. If converged, t , the seasonal component of the second time series Trend Component Get the residual component of the time series 7. The early warning method according to claim 6, characterized in that: According to the second time series Y t , the seasonal component of the second time series Trend Component Get the residual component of the time series The calculation formula is:

8. The early warning method according to claim 4, characterized in that: Also included between steps S200 and S300: Use standard residuals to perform residual normality test on the SARIMA model to determine whether the residuals obey normal distribution; The Durbin-Watson method is used to test the residual correlation of the SARIMA model to determine whether there is a first-order correlation in the residuals; After the SARIMA model passes the residual normality test and the residual correlation test, step S300 is performed.

9. The early warning method according to any one of claims 1 to 8, characterized in that: The preset rules are constructed by using expert experience analysis method.

Images