Artificial intelligence intelligent coal blending method based on deep learning

Abstract

The application discloses an AI intelligent coal blending method based on deep learning, and relates to the technical field of intelligent coal blending.The application solves the technical problems of low efficiency of feature correlation identification and screening and poor adaptability of model architecture scene.The application innovatively distinguishes linear and nonlinear correlation features through a scatter diagram+residual diagram system, removes redundancy by using a Pearson coefficient for linear features and mutual information method for nonlinear features, avoids the limitations of traditional single statistical quantity feature screening, mines implicit correlations through composite feature generation, makes the feature dimension more concise and the correlation with the coal blending target more direct, effectively improves the model training efficiency and prediction accuracy, customizes the model for different scenes, and realizes precise coverage of light, dynamic and complex scenes compared with the existing single model adaptation to all scenes, thereby greatly improving the applicability and optimization effect of the coal blending scheme in different scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent coal blending, in particular to an AI intelligent coal blending method based on deep learning. BACKGROUND

[0002] Coal blending refers to a process of mixing and blending two or more than two single coals with different properties according to certain proportions to obtain a mixed coal with better performance and more in line with the use standard. The coal blending scheme is the core of coking production, and its core function is to balance the yield, cost and by-product income and maintain the production stability under the premise of ensuring the coke quality. The coal blending scheme has a decisive influence on the subsequent industrial long production, the quantity and quality of products. Therefore, AI coal blending has important research significance.

[0003] The current coal blending process lacks intelligence, relies on manual experience, has a high error probability, and the waste of resources caused by unreasonable coal blending schemes occurs from time to time. The existing intelligent systems cannot handle special situations and have poor anti-interference ability, and many other problems. These problems not only restrict the stability of coke quality and the competitiveness of enterprises, therefore, optimizing the coal blending process through intelligent technology has become an important upgrading direction of the coking industry. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides an AI intelligent coal blending method based on deep learning, which solves the problems of inefficient feature association recognition and screening and poor scene adaptability of model architecture.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: an AI intelligent coal blending method based on deep learning, which specifically includes the following steps:

[0006] Collect coal quality basic data, production process data and target and constraint data, and pretreat the production process data and target and constraint data, and the pretreatment includes processing missing values and rationality verification;

[0007] Classify the core input features into correlation features and non-correlation features, wherein the correlation features are subjected to redundant feature elimination by Pearson coefficient, the non-correlation features are subjected to redundant feature elimination by mutual information method, and the reserved features are combined to generate composite features;

[0008] Select different model architectures based on different scenes, divide the data set and select the loss function based on the target quantity, adjust the hyperparameters through the optimization algorithm, verify the model precision by using the test set, and if it does not meet the standard, return to step one to re-optimize;

[0009] Collect real-time coal quality data and load constraint conditions, input the trained model to generate candidate schemes, push to the decision layer for selection, and trigger model retraining or parameter adjustment according to the feedback of the decision layer.

[0010] As a further scheme of the present application, the coal quality basic data includes industrial analysis indicators, elemental analysis indicators, and process characteristic indicators, and is collected in a mode combining offline testing and online monitoring.

[0011] The production process data covers equipment working conditions, historical coal blending records, and actual production effects, and is captured in real time through a manufacturing execution system and a distributed control system.

[0012] The target and constraint data includes target indicators and constraint conditions, the target indicators cover cost dimensions, quality dimensions, and environmental protection dimensions, and the constraint conditions include coal type availability, equipment limitations, and process limitations.

[0013] As a further scheme of the present application, linear interpolation is used to process missing values.

[0014] After sorting the target features in order according to the collection time to obtain an ordered sequence, the missing points are located, the missing values are estimated using a linear formula based on the known values before and after the missing points and the corresponding times.

[0015] The rationality check uses the 3σ principle, after verifying the normal distribution of the target features, the mean μ and the standard deviation σ are calculated, Lower = μ-3σ, Upper = μ+3σ, and the abnormal values exceeding the threshold are marked and processed.

[0016] As a further scheme of the present application, the classification of correlation features and non-correlation features is as follows:

[0017] A scatter plot is drawn with feature A as the x-axis and feature B as the y-axis, if the scatter points are distributed along a straight line, it tends to be linearly correlated, if they are distributed in a curved or non-linear regularity, it tends to be nonlinearly correlated.

[0018] Then, verification is performed through a residual plot, a scatter plot is drawn with feature A as the x-axis and the residual as the y-axis, if the residual is randomly distributed on both sides of y=0, it is linearly correlated, if the residual shows a systematic trend, it is nonlinearly correlated.

[0019] As a further scheme of the present application, when the Pearson coefficient is used for redundant feature elimination, all continuous features are obtained and standardized, a Pearson correlation coefficient matrix between the features is constructed, the correlation coefficients are calculated and compared with a high correlation threshold, feature pairs with correlation coefficients exceeding the threshold are screened, the feature with the highest correlation with the target indicator is retained, and the high correlation threshold is set by business requirements and has a value of 0.8.

[0020] As a further scheme of the present application, when the redundant features are removed according to the mutual information method, the continuous features and the target features are discretized, the mutual information values of the continuous features and the target features are calculated, the first N features are selected as screening features in descending order of the mutual information values, the mutual information values between the screening features are calculated again, the redundant features with mutual information values exceeding a threshold value are removed to obtain reserved features, and the reserved features are normalized to the interval [0, 1].

[0021] As a further scheme of the present application, the scene selection model architecture is specifically based on:

[0022] For single-target optimization, a basic regression model mainly based on a multilayer perceptron is selected, the number of input layer neurons is consistent with the feature dimension, 3-5 hidden layers are set, and the number of neurons in each layer is decreased by 2 / 3 of the input layer, and the output layer is one neuron;

[0023] For the scene of large coal quality fluctuation, an LSTM or GRU time series model is selected, the input is a sliding window time series feature, 2-3 hidden layers are set, and 64-128 neurons are set in each layer, and the output is a dynamic ratio of 1-3 hours in the future;

[0024] For high value-added scenes such as coking and coal chemical industry, a Transformer model or a hybrid intelligent model is selected, the hybrid intelligent model includes a CNN+MLP+linear programming, a model combining deep learning and traditional algorithms.

[0025] As a further scheme of the present application, the data set is divided into a training set, a validation set and a test set in the order of time according to a ratio of 7:2:1, and a loss function is selected based on the number of targets;

[0026] In the single-target case, mean square error is selected to calculate deviation, and in the multi-target case, a weighted loss function is selected and combined with an algorithm to generate a Pareto optimal solution.

[0027] As a further scheme of the present application, the Pearson correlation coefficient matrix between features is constructed in the following manner:

[0028] All continuous features are obtained and standardized, denoted as X, and X=1, 2, …, Y, wherein Y represents the type of continuous features, a Pearson correlation coefficient matrix R between features is constructed, wherein r XY represents the Pearson correlation coefficient of feature X and feature Y, and the value range is [-1, 1], b is the number of samples, xa a represents the observation value of feature X in the a-th sample, ya represents the observation value of feature Y in the a-th sample, represents the sample mean of feature X, r Y represents the sample mean of feature Y.

[0029] As a further scheme of the present application, the way of calculating the mutual information value of the continuous feature and the target feature is:

[0030] The continuous feature and the target feature are acquired, and the continuous feature and the target feature are discretized to convert them into discrete variables, and then the mutual information value of the continuous feature X and the target feature Y is calculated according to the formula The mutual information value of the continuous feature X and the target feature Y is calculated, wherein P(X, Y) is the joint probability, P(X) is the marginal probability of X taking x, and P(Y) is the marginal probability of Y taking y.

[0031] The present application provides an AI intelligent coal blending method based on deep learning.

[0032] The present application adopts a multi-source data acquisition mode of offline testing + online monitoring, combines fine pretreatment of linear interpolation and 3σ principle, retains high precision of offline data, and plays the real-time nature of online data; through time sequence alignment and abnormal value linkage equipment state checking, the problems of single data dimension and poor quality of traditional coal blending data are solved, and more reliable input is provided for the model.

[0033] The present application innovatively distinguishes linear and nonlinear correlation features through scatter plot + residual plot system, removes redundancy for linear features using Pearson coefficient and nonlinear features using mutual information method, avoids the limitations of traditional single statistical quantity feature screening, and effectively improves model training efficiency and prediction accuracy by mining implicit correlation through composite features.

[0034] The present application customizes models for different scenarios, uses lightweight MLP for single-target scenarios to meet the efficient needs of small and medium-sized enterprises, uses LSTM / GRU for coal quality fluctuation scenarios to capture time sequence dynamics, and uses mixed models such as Transformer or CNN+MLP+linear programming for high value-added scenarios to process multi-modal data and multi-target constraints. Compared with the existing single model that adapts to all scenarios, the present application realizes precise coverage of lightweight-dynamic-complex scenarios, and greatly improves the applicability and optimization effect of the coal blending scheme in different scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The present application provides a step method diagram. DETAILED DESCRIPTION

[0036] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of the present application.

[0037] First embodiment:

[0038] Please refer to Figure 1 The present application provides an AI intelligent coal blending method based on deep learning, which specifically comprises the following steps:

[0039] Step one, collect coal quality basic data, specifically including industrial analysis indexes such as moisture M, ash A, volatile matter V, fixed carbon FC; element analysis indexes such as carbon C, hydrogen H, oxygen O, nitrogen N, sulfur S; process characteristic indexes such as caking index G, gel layer thickness Y, ash melting point ST, and calorific value Qnet, var, and collect through the combination mode of offline testing + online monitoring. Specifically, the offline testing is carried out laboratory detection according to the national standard method, the online monitoring is realized through near infrared spectrometer, laser particle size instrument and other equipment to realize second-level data updating, and it is taken as the core input feature. At the same time, production process data and target and constraint data are collected. Specifically, the production process data covers equipment working conditions such as coking furnace temperature, gasification furnace pressure; historical coal blending records, actual production effects such as coke strength M40 / CSR, boiler thermal efficiency, gasification furnace carbon conversion rate, which are real-time grabbed through manufacturing execution system MES and distributed control system DCS. The data sampling frequency is usually 1-5 minutes / time. The target and constraint data includes target indexes and constraint conditions, wherein the target indexes include cost dimension, quality dimension and environmental protection dimension, and the constraint conditions include coal type availability, equipment limitation and process limitation. The target and constraint data sources include indexes issued by production planning department, supply chain inventory system and equipment parameter manual. At the same time, the production process data and target and constraint data are preprocessed, and the preprocessing includes missing value processing and rationality checking;

[0040] The missing value is processed by linear interpolation. The target features are sorted in time sequence to obtain an ordered sequence X0=[x1, x2, …, x i-1 , NaN, x i+1 , …, x n ]. Then the missing point and the missing value position i are obtained, and the previous known value x i-1 and the next known value x i+1 corresponding to the time t-1 and t+1 are found. The missing value is estimated by linear formula .

[0041] The rationality check is processed by the 3σ principle, Shapiro-Wilk test or Q-Q plot analysis is performed on the target feature to verify whether it conforms to the normal distribution, it is assumed that the normal distribution is met, then the mean μ and the standard deviation σ of the feature are calculated, and the abnormal threshold is set, Lower = μ-3σ, Upper = μ+3σ, the data less than Lower or greater than Upper is marked as an abnormal value, and is removed after confirmation of the original record; if it is a false alarm of the device, it can be replaced with normal data at the adjacent time.

[0042] Step two, obtain the core input features, and classify them into correlation features and non-correlation features, and the specific classification method is as follows:

[0043] Take feature A as the x-axis and feature B as the y-axis, draw the scatter distribution of all samples, observe the aggregation trend of the points, if the scatter points are roughly distributed along a straight line, it tends to be linearly correlated, if the scatter points show a curve trend or non-linear regularity distribution, such as U-shaped, S-shaped, logarithmic, it tends to be non-linearly correlated;

[0044] Then verify whether there is a non-linear correlation through the residual plot, draw the scatter plot with feature A as the x-axis and the residual as the y-axis, if the residual is randomly distributed on both sides of y=0, it is linearly correlated, if the residual shows a systematic trend, such as the residual increases with A showing a positive, negative, or positive curve, it is non-linearly correlated;

[0045] The Pearson coefficient is used for redundant feature elimination for correlation features, and the mutual information method is used for redundant feature elimination for non-correlation features;

[0046] For the analysis of the Pearson coefficient, all continuous features are obtained and standardized, denoted as X, and X=1, 2, …, Y, where Y represents the types of continuous features, such as ash content A, volatile content V, and heat output Q, a Pearson correlation coefficient matrix R between features is built, where r XY represents the Pearson correlation coefficient between feature X and feature Y, and the value range is [-1, 1], b is the sample number, such as the number of coal batches and the number of historical coal blending schemes in the coal blending data, x a represents the observation value of feature X in the a-th sample, such as the ash content and volatile content of the a-th batch of coal, ya represents the observation value of feature Y in the a-th sample, such as the heat output and coke strength M40 of the a-th batch of coal, represents the sample mean of feature X, such as the average ash content of all batches of coal, r YThe sample mean representing the feature Y, such as the average calorific value of all batches of coal, is calculated, and the calculated Pearson correlation coefficient is compared with a high correlation threshold value, and the specific value of the high correlation threshold value is set by business requirements, generally taking a value of 0.8, and the feature pairs with a correlation coefficient exceeding the threshold value are screened, and all high-correlation feature pairs are obtained in this way, and the feature with the highest correlation with the target index is retained,

[0047] For the analysis of the mutual information method, continuous features and target features are obtained, and the continuous features and the target index are discretized, and are converted into discrete variables, and then the mutual information value of the continuous features X and the target features Y is calculated according to the formula The mutual information value of the continuous features X and the target features Y is calculated, where P(X, Y) is the joint probability, P(X) is the marginal probability of X taking x, and P(Y) is the marginal probability of Y taking y, and then the mutual information values MI are sorted from large to small, and the first N features are selected as the screened features. For the screened features, the mutual information values between each other are calculated again, and the redundant features with a mutual information value MI exceeding the threshold value are removed to obtain the retained features, and the retained features are converted to the [1, 1] interval, and then the retained features are combined to generate composite features.

[0048] Step three, different model architectures are selected based on different scenarios. For single-target optimization, a basic regression model is selected, mainly using a multilayer perceptron to learn simple nonlinear relationships through an input layer-3-5 hidden layers-output layer. The specific number of input layer neurons is consistent with the feature dimension, such as 8-15 coal quality features corresponding to 8-15 input neurons, and the number of neurons in the hidden layer is set to 3-5, with the number of neurons in each layer decreasing by 2 / 3 of the input layer, such as 10 dimensions, 6-8 in the first layer, and 4-5 in the second layer to avoid overfitting, and the output layer has one neuron. For large coal quality fluctuations, a time series model is selected, and LSTM / GRU is used to capture coal quality time series changes, and the output is a dynamic blending scheme. The input is a sliding window time series feature, such as the coal quality data of the previous 6 hours + the current inventory, and the window length is set according to the fluctuation period, usually 4-12 hours. The hidden layer has 2-3 layers with 64-128 neurons in each layer, and the output layer is the dynamic blending of the next 1-3 hours. For high-value-added scenarios such as coking and coal chemical industry, multi-modal or complex models are selected, including Transformer model; hybrid intelligent model, combining CNN+MLP+linear programming, deep learning+traditional algorithm, and then the data set is divided into training set, validation set and test set in the order of time according to the ratio of 7:2:1;

[0049] Then different loss functions are selected based on different target quantities, for single target, mean square error is selected to calculate deviation, for multi-target, a weighted loss function is selected to generate Pareto optimal solution, learning rate, number of hidden layer neurons and number of iterations are adjusted through grid, random or Bayesian optimization, the specific number of iterations is 1000-5000 rounds, to ensure model convergence, test set is used to test accuracy, if not up to standard, return to data preprocessing link for optimization.

[0050] Step four, real-time data is collected by online near-infrared spectrometer (error ≤0.5% for detecting ash, volatile matter and moisture) and X-ray fluorescence instrument (error ≤0.02% for detecting sulfur and phosphorus), instantaneous value is generated every 10 seconds, 30-second sliding average is aggregated every 30 seconds, and inventory, procurement, environmental protection and other constraints are loaded from ERP system, and are converted into mathematical conditions recognizable by the model as constraint conditions, when there is a constraint conflict, priority mechanism is triggered, first priority: environmental protection and safety constraints (must be met, such as SO2 emission exceeding limit, the scheme is directly excluded); second priority: equipment constraints (such as coal mill load cannot be exceeded, otherwise the ratio is adjusted); third priority: cost, inventory and other flexible constraints (can be optimized after meeting the first and second priorities);

[0051] Real-time coal quality characteristics and constraint conditions are input into the trained model, the model calculates the optimal proportion of each coal type using the learned nonlinear relationship, and simultaneously predicts target indicators, in response to the dynamic change of target priority, the model generates 3-5 candidate schemes by adjusting the loss function weight, each candidate scheme includes the ratio of different coal types and the corresponding target indicator prediction value, such as the lowest cost scheme, the best quality scheme, the environmental compliance scheme, etc., then the candidate schemes are pushed to the production decision layer, the decision layer selects or adjusts the final coal blending scheme from the candidate schemes according to actual production demand, market dynamics and strategic target, if the decision layer is not satisfied with the candidate schemes, feedback mechanism can be triggered to require the model to retrain or adjust parameters to generate new candidate schemes that meet the requirements better.

[0052] Second embodiment:

[0053] As a second embodiment of the present application, it is implemented on the basis of the first embodiment, and differs from the first embodiment in that:

[0054] The missing values are processed by box plot method, the distribution interval is divided by the quartiles of the data, the data exceeding 1.5 times of the interquartile range is judged as abnormal value, which is not affected by extreme value, the target characteristics are sorted in time sequence to obtain an ordered sequence X0=[x1, x2, …, x i-1 , NaN, x i+1 , …, x nAt the same time, the lower quartile Q1 (25% quantile), the median Q2 (50% quantile), and the upper quartile Q3 (75% quantile) are calculated, and then the interquartile range IQR=Q3-Q1 is calculated;

[0055] Then, the abnormal threshold is determined according to the calculated interquartile range, wherein the lower boundary is Q1-1.5*IQR, and the upper boundary is Q3+1.5*IQR, the data beyond the boundary is marked as an abnormal value, and if it is an isolated case, it is directly eliminated; if it is a continuous abnormality, it is necessary to check whether the equipment is faulty.

[0056] The third embodiment is:

[0057] As the third embodiment of the present application, the first embodiment and the second embodiment are combined and implemented.

[0058] Part of the data in the above formula is calculated by taking its numerical value, and is not calculated by substituting the parameter unit, and the contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0059] The above embodiments are only used to illustrate the technical method of the present application and are not limited. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. An AI intelligent coal blending method based on deep learning, characterized in that, The method specifically comprises the following steps: Collecting coal quality basic data, production process data, and target and constraint data, preprocessing the production process data and the target and constraint data, and the preprocessing including processing missing values and reasonableness checking; Classifying the core input features into correlation features and non-correlation features, wherein the correlation features are subjected to redundant feature elimination by using a Pearson coefficient, the non-correlation features are subjected to redundant feature elimination by using a mutual information method, the retained features are combined to generate composite features, and the specific processing manner is as follows: Taking the feature A as the x-axis and the feature B as the y-axis to draw a scatter plot, if the scatter points are distributed along a straight line, the linear correlation is inclined, if the scatter points are distributed in a curved or non-linear regularity, the non-linear correlation is inclined; Then, verification is performed by using a residual plot, taking the feature A as the x-axis and the residual as the y-axis to draw a scatter plot, if the residuals are randomly distributed on both sides of y=0, the linear correlation is linear, and if the residuals show a systematic trend, the non-linear correlation is non-linear; Different model architectures are selected based on different scenarios, the data set is divided, the loss function is selected based on the number of targets, the hyperparameters are adjusted by using an optimization algorithm, the model accuracy is verified by using a test set, if the accuracy is not up to standard, the step one is returned to re-optimize, and the specific processing manner is as follows: For single-target optimization, a basic regression model mainly using a multilayer perceptron is selected, the number of input layer neurons is consistent with the feature dimension, 3-5 layers of hidden layers are set, the number of neurons in each layer is decreased by 2 / 3 of the input layer, and the output layer has one neuron; For a coal quality fluctuation scenario, an LSTM or GRU time series model is selected, the input is a sliding window time series feature, 2-3 layers of hidden layers are set, and each layer has 64-128 neurons, and the output is a dynamic proportioning for 1-3 hours in the future; For high value-added scenarios such as coking and coal chemical industry, a Transformer model or a hybrid intelligent model is selected, the hybrid intelligent model includes a CNN+MLP+linear programming, a model combining deep learning and traditional algorithms; Real-time coal quality data are collected and constraint conditions are loaded, a trained model is input to generate a candidate scheme, the candidate scheme is pushed to a decision layer for selection, and the model is retrained or the parameters are adjusted according to the feedback of the decision layer. 2.The deep learning-based AI intelligent coal blending method according to claim 1, characterized in that, The coal quality basic data include industrial analysis indexes, element analysis indexes, and process characteristic indexes, and are collected in a mode combining offline testing and online monitoring; The production process data cover equipment working conditions, historical coal blending records, and actual production effects, and are real-time grabbed through a manufacturing execution system and a distributed control system; The target and constraint data include target indexes and constraint conditions, the target indexes cover cost dimensions, quality dimensions, and environmental protection dimensions, and the constraint conditions include coal availability, equipment limitations, and process limitations. 3.The deep learning-based AI intelligent coal blending method according to claim 1, characterized in that, The missing values are processed by using linear interpolation: After the target features are sorted according to collection time to obtain an ordered sequence, the missing points are located, the missing values are estimated by using a linear formula according to the known values before and after the missing points and the corresponding time; The reasonableness checking is performed by using a 3σ principle, after the target features are verified to be normally distributed, the mean μ and the standard deviation σ are calculated, Lower=μ-3σ, Upper=μ+3σ, the abnormal values exceeding the threshold are marked and processed. 4.The deep learning-based AI intelligent coal blending method according to claim 1, characterized in that, For the Pearson coefficient, all continuous features are obtained and standardized, the Pearson correlation coefficient matrix between features is constructed, the correlation coefficient is calculated and compared with the high correlation threshold, the feature pairs with correlation coefficient exceeding the threshold are screened, the feature with the highest correlation with the target index is retained, and the high correlation threshold is set by business demand, with a value of 0.

8. 5.The deep learning-based AI intelligent coal blending method according to claim 1, wherein, For the mutual information method, the continuous features and target features are discretized, the mutual information value of the continuous features and the target features is calculated, the top N features are selected as screening features according to the mutual information value from large to small, the mutual information value between the screening features is calculated again, the redundant features with mutual information value exceeding the threshold are removed to obtain the retained features, and the retained features are normalized to the interval [0, 1]. 6.The deep learning-based AI intelligent coal blending method according to claim 1, characterized in that, The data set is divided into training set, validation set and test set in time sequence 7:2:1, and the loss function is selected based on the target number; In the single-target case, the mean square error is selected to calculate the deviation, and in the multi-target case, the weighted loss function is selected and combined with the algorithm to generate the Pareto optimal solution. 7.The deep learning-based AI intelligent coal blending method according to claim 4, characterized in that, The Pearson correlation coefficient matrix between features is constructed in the following manner: all continuous features are obtained and standardized, denoted as X, and X = 1, 2, …, Y, wherein Y represents the number of continuous features, a Pearson correlation coefficient matrix R between features is constructed, wherein r XY represents the Pearson correlation coefficient between feature X and feature Y, and the value range is [-1, 1], b is the number of samples, xa a represents the observation value of feature X in the a th sample, ya represents the observation value of feature Y in the a th sample, represents the sample mean of feature X, r Y represents the sample mean of feature Y. 8.The deep learning-based AI intelligent coal blending method according to claim 5, characterized in that, The way of calculating the mutual information value of the continuous feature and the target feature is: obtaining the continuous feature and the target feature, and performing discretization processing on the continuous feature and the target feature, converting the continuous feature and the target feature into discrete variables, then calculating the mutual information value of the continuous feature X and the target feature Y according to the formula The mutual information value of the continuous feature X and the target feature Y is calculated, wherein P(X, Y) is a joint probability, P(X) is an edge probability of X taking x, and P(Y) is an edge probability of Y taking y.

Images