Carbon anode kneading quality control system based on artificial intelligence

Abstract

The invention discloses a carbon anode mixing and kneading quality control system based on artificial intelligence, particularly relates to the field of carbon anode production, and comprises a raw material information acquisition module, a mixing and kneading process monitoring module, a data preprocessing module, a quality analysis and prediction module, a control instruction generation module, an execution module and a feedback adjustment module. According to the method, information of raw materials and the mixing and kneading process is comprehensively collected, all-around monitoring of the mixing and kneading quality is achieved, rich data support is provided for accurate evaluation and prediction of the mixing and kneading quality, the quality prediction model is established through the artificial intelligence algorithm, the quality index of the carbon anode can be accurately predicted, potential quality problems can be found in advance, and the quality prediction efficiency is improved. A scientific basis is provided for production decision making, dynamic control over the mixing and kneading process is achieved based on a control instruction generation and feedback adjustment mechanism of a prediction result, mixing and kneading parameters can be adjusted in time, the mixing and kneading quality of the carbon anode is effectively improved, the rejection rate is reduced, and the production efficiency and the product quality are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon anode production, and more specifically, to an artificial intelligence-based carbon anode kneading quality control system. Background Art

[0002] As a key component of aluminum electrolytic cells, the quality of carbon anodes has a direct and significant impact on the efficiency, energy consumption, and product quality of aluminum electrolysis production. The kneading process is the core link in the production of carbon anodes. It aims to evenly mix and plasticize raw materials such as petroleum coke and asphalt to form a paste with good molding properties. However, the current quality control of the carbon anode kneading process faces many challenges. Traditional control methods mainly rely on manual experience and simple process parameter settings. It is difficult to accurately control the complex factors in the kneading process. For example, fluctuations in raw material quality, changes in ambient temperature and humidity, etc. will have a significant impact on the kneading quality, resulting in unstable product quality, high scrap rates, and increased production costs.

[0003] Furthermore, existing monitoring methods often only capture limited parameter information, failing to fully and in real time reflect the dynamic changes in the kneading process. This makes it difficult to predict and identify quality issues in advance, thus impacting the continuity and stability of production. Therefore, developing a system that can accurately monitor and control the kneading quality of carbon anodes in real time is of great practical significance. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a carbon anode kneading quality control system based on artificial intelligence, which solves the problems raised in the above-mentioned background technology through the following scheme.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: an artificial intelligence-based carbon anode kneading quality control system, comprising a raw material information acquisition module, a kneading process monitoring module, a data preprocessing module, a quality analysis and prediction module, a control instruction generation module, an execution module, and a feedback adjustment module;

[0006] The raw material information collection module is used to collect basic information of various raw materials, generate raw material information text and transmit the raw material information text to the data preprocessing module;

[0007] The kneading process monitoring module includes a temperature monitoring unit, a time monitoring unit, a speed monitoring unit, and a torque monitoring unit, and is used to monitor the temperature, kneading time, speed of the kneading equipment, and torque changes during the kneading process in real time to obtain kneading process parameter text and environmental parameter text, and transmit the kneading process parameter text and environmental parameter text to the data preprocessing module;

[0008] The data preprocessing module includes a data cleaning unit and a data standardization unit, which are used to preprocess the collected raw material information text, kneading process parameter text and environmental parameter text, and transmit the processed text to the quality analysis and prediction module;

[0009] The quality analysis and prediction module is used to input the processed raw material information text, kneading process parameter text, and environmental parameter text into the long short-term memory network model and the random forest model, and perform weighted fusion on the prediction results of the long short-term memory network model and the random forest model to obtain the final kneading quality prediction result;

[0010] The control instruction generation module is used to compare the prediction results of the quality analysis and prediction module with the preset quality standard range. If the prediction results do not meet the preset standard, a corresponding control instruction is generated based on the deviation analysis and transmitted to the actuator of the kneading equipment to adjust the parameters of the kneading process;

[0011] The execution module is used to adjust the kneading equipment in real time according to the control instructions generated by the control decision module to ensure that the kneading process is carried out according to the predetermined process parameters;

[0012] The feedback adjustment module: after the execution module executes the control instruction, it continuously collects the actual data of the kneading process and the quality detection data of the carbon anode, feeds these data back to the quality analysis and prediction module, optimizes and adjusts the quality prediction model and control strategy, and realizes dynamic control of the kneading quality.

[0013] Preferably, the raw material information text includes the particle size distribution of petroleum coke (D 10 、D 50 、D 90 ), where D 10 Indicates the particle size corresponding to when the cumulative particle size distribution percentage reaches 10%, D 50 is the median particle size, D 90 Indicates the particle size corresponding to the cumulative particle size distribution percentage reaching 90%. Ash content of petroleum coke (A pc ), volatile matter content (V pc ), sulfur content (S pc ). Softening point of asphalt (T s ), needle penetration (P), coking value (C v ).

[0014] Preferably, the kneading process parameters include: kneading temperature (T), in degrees Celsius (°C), measured by a high-precision thermocouple. Kneading time (t), in minutes (min), recorded by a timing device. Stirring speed (n), in revolutions per minute (r / min), obtained by a speed sensor. Pressure (p), in megapascals (MPa), monitored by a pressure sensor. Torque (M), in Newton meters (N·m), measured by a torque sensor;

[0015] The environmental parameters include: ambient temperature (T e ) and humidity (H e ), and collected using temperature and humidity sensors.

[0016] Preferably, the data cleaning unit is used to remove noise, outliers and missing values ​​in the collected data. For missing values, linear interpolation or mean filling method can be used for processing; for outliers, they are identified and eliminated through a preset threshold range.

[0017] Preferably, the data standardization unit processes the data using a Z-score standardization method. For each parameter x, the standardized value x norm The calculation formula is:

[0018]

[0019] Among them, μ is the mean of the parameter and σ is the standard deviation of the parameter.

[0020] Preferably, the long short-term memory network model is used to process time series data in the kneading process, such as changes in temperature and torque over time. The LSTM model can capture long-term dependencies in the data. Its input is the time series data after preprocessing and feature extraction, and its output is the predicted value of the kneading quality index at a certain moment in the future. The state update formula of the LSTM unit is as follows:

[0021] Input gate (i t ):i t =σ(W ii x t +W hi h t-1 +b i )

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

[0023] Output gate (ot ):o t =σ(W io x t +W ho h t-1 +b o )

[0024] Cell state update (C t ):

[0025] C t =f t ☉C t-1 +i t ☉tanh(W ic x t +W hc h t-1 +b c )

[0026] Hidden state update (h t ):h t =o t ☉tanh(C t )

[0027] Among them, x t is the input at the current moment, h t-1 is the hidden state of the previous moment, C t-1 is the cell state at the previous moment, W is the weight matrix, b is the bias vector, σ is the sigmoid function, and ⊙ represents element-by-element multiplication.

[0028] Preferably, the random forest model is used to process non-time series data, such as raw material parameters and environmental parameters, to predict kneading quality indicators;

[0029] The steps of the random forest model are as follows:

[0030] Decision tree generation: Each decision tree in a random forest is trained on a subset obtained through bootstrapping. Bootstrapping involves randomly extracting samples from the original dataset with replacement to form a new subset of the same size as the original dataset. For each decision tree, when splitting at each node, rather than considering all features, a subset of features is randomly selected from the original dataset. The optimal features and split points are then selected from these features to split the node. Assuming there are m features in the original dataset, the size of the feature subset selected is usually m.

[0031] Feature selection criteria: In the process of node splitting in the decision tree, Gini impurity is used as the criterion for feature selection. For a data set D, the calculation formula for its Gini impurity is:

[0032]

[0033] Among them, k is the number of categories in the data set. For regression problems, the continuous quality index can be divided into several intervals, which are regarded as different "categories". i is the proportion of samples of the i-th category in the data set D;

[0034] For a feature A, suppose it has v different values ​​{a1, a2, ..., a v}, divide the data set D according to the value of feature A, and obtain v subsets {D1,D2,…,D v}, then the Gini index of feature A for data set D is:

[0035]

[0036] At each node, select the feature with the smallest Gini index and the corresponding split point for splitting.

[0037] Random forest prediction: For a new sample, input it to each decision tree in the random forest and get the prediction result of each decision tree. For regression problems, the final prediction result of the random forest is the average of the prediction results of all decision trees. Assume that there are N decision trees in the random forest, and the prediction result of the jth decision tree for the sample is The final prediction result of random forest is for:

[0038]

[0039] Preferably, the weighted fusion formula is:

[0040]

[0041] in, is the final predicted value, is the predicted value of the LSTM model, is the predicted value of the RF model, w1 and w2 are weight coefficients, and w1+w2=1. The weight coefficients are optimized and determined by the cross-validation method;

[0042] The predicted mixing quality indicators include the density (ρ), compressive strength (Rc), and resistivity (ρe) of the carbon anode.

[0043] Preferably, the quality standard range refers to setting a reasonable target value and an allowable fluctuation range for each kneading quality indicator, and the preset range is obtained by laboratory experiments and daily practice;

[0044] The deviation analysis refers to calculating the deviation between the quality prediction value and the target value. The deviation calculation formula is as follows:

[0045] in is the predicted value, and y0 is the target value.

[0046] Preferably, the execution module specifically performs the following actions:

[0047] Adjust the power of the heating device to control the kneading temperature, adjust the speed of the stirring motor to change the stirring speed, control the opening of the feed valve, and adjust the feed amount of raw materials.

[0048] Technical effects and advantages of the present invention:

[0049] 1. A comprehensive monitoring and control system has been established. By comprehensively collecting information on raw materials and the kneading process, all-round monitoring of kneading quality has been achieved, laying a solid data foundation for the subsequent accurate evaluation and prediction of kneading quality. In terms of raw material information collection, parameters such as the particle size distribution, ash content, volatile matter, and sulfur content of petroleum coke, as well as the softening point, needle penetration, and coking value of asphalt are accurately measured. Slight differences in these raw material parameters will affect the kneading quality. Comprehensive collection of this information can more accurately grasp the characteristics of the raw materials. At the same time, during the kneading process, high-precision sensors are used to collect key process parameters such as kneading temperature, time, stirring speed, pressure, and torque, as well as environmental parameters such as ambient temperature and humidity in real time. By comprehensively collecting this multi-dimensional information covering raw materials, processes, and the environment, a large and rich data set has been formed, providing sufficient data support for the accurate evaluation and prediction of kneading quality, making the analysis of kneading quality no longer one-sided and limited;

[0050] 2. The present invention uses advanced artificial intelligence algorithms to construct a quality prediction model. The long short-term memory network and random forest algorithm in deep learning are introduced into the kneading quality prediction. The LSTM model, with its powerful processing ability for time series data, can capture the changing trend of parameters such as temperature and torque over time during the kneading process and mine the long-term dependencies in the data; while the RF model is good at processing non-time series data and efficiently analyzing raw material parameters and environmental parameters. By combining the two, the quality prediction model can accurately predict the density, compressive strength, resistivity and other quality indicators of carbon anodes. This precise prediction capability enables potential quality problems in the production process to be discovered in advance. For example, before the product is formed, problems such as uneven density or insufficient compressive strength can be predicted. This provides a scientific basis for production decisions. Production personnel can adjust production plans and process parameters in a timely manner according to the prediction results to avoid producing a large number of substandard products and reduce production costs.

[0051] 3. The present invention establishes a control instruction generation and feedback adjustment mechanism based on the prediction results, achieving dynamic control of the kneading process. When the quality prediction model predicts that quality indicators may deviate from expectations, the control instruction generation module will quickly generate corresponding control instructions based on preset rules and algorithms. This dynamic control mechanism can timely adjust kneading parameters, effectively improving the kneading quality of carbon anodes and significantly reducing the scrap rate. The reduced scrap rate means increased production efficiency, reduced waste of raw materials and energy, and improved product quality, enhancing product competitiveness in the market. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the overall structure of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0054] As attached Figure 1 The carbon anode kneading quality control system shown in the figure based on artificial intelligence includes a raw material information acquisition module, a kneading process monitoring module, a data preprocessing module, a quality analysis and prediction module, a control instruction generation module, an execution module and a feedback adjustment module.

[0055] The raw material information collection module is used to collect basic information of various raw materials, generate raw material information text and transmit the raw material information text to the data preprocessing module;

[0056] It should be further explained that the raw material information text includes the particle size distribution of petroleum coke (D 10 、D 50 、D 90 ), where D 10 Indicates the particle size corresponding to when the cumulative particle size distribution percentage reaches 10%, D 50 is the median particle size, D 90 Indicates the particle size corresponding to the cumulative particle size distribution percentage reaching 90%, the ash content of petroleum coke (A pc ), volatile matter content (V pc ), sulfur content (S pc ). Softening point of asphalt (T s ), needle penetration (P), coking value (C v ).

[0057] The kneading process monitoring module includes a temperature monitoring unit, a time monitoring unit, a speed monitoring unit, and a torque monitoring unit, which are used to monitor the temperature, kneading time, speed of the kneading equipment, and torque changes in the kneading process in real time to obtain kneading process parameter text and environmental parameter text, and transmit the kneading process parameter text and environmental parameter text to the data preprocessing module.

[0058] It should be further explained that the kneading process parameters include: kneading temperature (T), in degrees Celsius (°C), measured by a high-precision thermocouple; kneading time (t), in minutes (min), recorded by a timing device; stirring speed (n), in revolutions per minute (r / min), obtained by a speed sensor; pressure (p), in megapascals (MPa), monitored by a pressure sensor; torque (M), in Newton meters (N·m), measured by a torque sensor;

[0059] The environmental parameters include: ambient temperature (T e ) and humidity (H e ), and collected using temperature and humidity sensors.

[0060] The data preprocessing module includes a data cleaning unit and a data standardization unit, which are used to preprocess the collected raw material information text, kneading process parameter text and environmental parameter text, and transmit the processed text to the quality analysis and prediction module;

[0061] The data cleaning unit is used to remove noise, outliers and missing values ​​in the collected data. For missing values, linear interpolation or mean filling method can be used for processing; for outliers, they are identified and eliminated within a preset threshold range;

[0062] The data standardization unit processes the data using the Z-score standardization method. For each parameter x, its standardized value x norm The calculation formula is:

[0063]

[0064] Among them, μ is the mean of the parameter, σ is the standard deviation of the parameter;

[0065] The quality analysis and prediction module is used to input the processed raw material information text, kneading process parameter text, and environmental parameter text into the long short-term memory network model and the random forest model, and perform weighted fusion on the prediction results of the long short-term memory network model and the random forest model to obtain the final kneading quality prediction result;

[0066] The long short-term memory network model is used to process time series data in the kneading process, such as the changes in temperature and torque over time. The LSTM model can capture long-term dependencies in the data. Its input is the time series data after preprocessing and feature extraction, and its output is the predicted value of the kneading quality index at a certain moment in the future. The state update formula of the LSTM unit is as follows:

[0067] Input gate (i t ):i t =σ(W ii x t +W hi h t-1 +b i )

[0068] Forget gate (f t ):f t =σ(W if x t +W hf h t-1 +b f )

[0069] Output gate (o t ):o t =σ(W io x t +W ho h t-1 +b o )

[0070] Cell state update (C t ):

[0071] C t =f t ☉C t-1 +i t ☉tanh(W ic x t +W hc h t-1 +b c )

[0072] Hidden state update (h t ):h t =o t ☉tanh(C t )

[0073] Among them, x t is the input at the current moment, h t-1 is the hidden state of the previous moment, C t-1 is the cell state at the previous moment, W is the weight matrix, b is the bias vector, σ is the sigmoid function, and ⊙ represents element-by-element multiplication;

[0074] The random forest model is used to process non-time series data, such as raw material parameters and environmental parameters, to predict kneading quality indicators;

[0075] The steps of the random forest model are as follows:

[0076] Decision tree generation: Each decision tree in a random forest is trained based on a subset obtained through bootstrap sampling. Bootstrap sampling refers to randomly extracting samples from the original dataset with replacement to form a new subset of the same size as the original dataset. For each decision tree, when splitting at each node, rather than considering all features, a subset of features is randomly selected from the original dataset. The optimal features and splitting points are then selected from these features to split the node. Assuming that there are m features in the original dataset, the size of the feature subset selected is usually m.

[0077] Feature selection criteria: In the process of node splitting of the decision tree, Gini impurity is used as the criterion for feature selection. For a data set D, the calculation formula of its Gini impurity is:

[0078]

[0079] Among them, k is the number of categories in the data set. For regression problems, the continuous quality index can be divided into several intervals, which are regarded as different "categories". i is the proportion of samples of the i-th category in the data set D;

[0080] For a feature A, suppose it has v different values ​​{a1, a2, ..., a v}, divide the data set D according to the value of feature A, and obtain v subsets {D1,D2,…,D v}, then the Gini index of feature A for data set D is:

[0081]

[0082] At each node, select the feature with the smallest Gini index and the corresponding split point for splitting.

[0083] Random forest prediction: For a new sample, input it into each decision tree in the random forest to obtain the prediction result of each decision tree. For regression problems, the final prediction result of the random forest is the average of the prediction results of all decision trees. Assuming that there are N decision trees in the random forest, the prediction result of the jth decision tree for the sample is The final prediction result of random forest is for:

[0084]

[0085] The weighted fusion formula is:

[0086]

[0087] in, is the final predicted value, is the predicted value of the LSTM model, is the predicted value of the RF model, w1 and w2 are weight coefficients, and w1+w2=1. The weight coefficients are optimized and determined by the cross-validation method;

[0088] The predicted kneading quality indicators include the density (ρ), compressive strength (Rc), and resistivity (ρe) of the carbon anode;

[0089] The control instruction generation module is used to compare the prediction results of the quality analysis and prediction module with the preset quality standard range. If the prediction results do not meet the preset standard, a corresponding control instruction is generated based on the deviation analysis and transmitted to the actuator of the kneading equipment to adjust the parameters of the kneading process;

[0090] It should be further explained that the quality standard range refers to setting a reasonable target value and an allowable fluctuation range for each kneading quality indicator. The preset range is obtained by laboratory experiments and daily practice. For example, the density target value of carbon anode is ρ0, and the allowable fluctuation range is [ρ0-Δρ, ρ0+Δρ];

[0091] The deviation analysis refers to calculating the deviation between the quality prediction value and the target value. The deviation calculation formula is as follows:

[0092] in is the predicted value, y0 is the target value;

[0093] The execution module is used to adjust the kneading equipment in real time according to the control instructions generated by the control decision module to ensure that the kneading process is carried out according to the predetermined process parameters;

[0094] It should be noted that the execution module specifically performs the following actions:

[0095] Adjust the power of the heating device to control the kneading temperature, adjust the speed of the stirring motor to change the stirring speed, control the opening of the feed valve, and adjust the feed amount of raw materials.

[0096] The feedback adjustment module: after the execution module executes the control instruction, it continuously collects the actual data of the kneading process and the quality detection data of the carbon anode, feeds these data back to the quality analysis and prediction module, optimizes and adjusts the quality prediction model and control strategy, and realizes dynamic control of the kneading quality.

[0097] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0098] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A carbon anode kneading quality control system based on artificial intelligence, characterized in that: include: Raw material information collection module, kneading process monitoring module, data preprocessing module, quality analysis and prediction module, control instruction generation module, execution module and feedback adjustment module; The raw material information collection module is used to collect basic information of various raw materials, generate raw material information text and transmit the raw material information text to the data preprocessing module; The kneading process monitoring module includes a temperature monitoring unit, a time monitoring unit, a speed monitoring unit, and a torque monitoring unit, and is used to monitor the temperature, kneading time, speed of the kneading equipment, and torque changes during the kneading process in real time to obtain kneading process parameter text and environmental parameter text, and transmit the kneading process parameter text and environmental parameter text to the data preprocessing module; The data preprocessing module includes a data cleaning unit and a data standardization unit, which are used to preprocess the collected raw material information text, kneading process parameter text and environmental parameter text, and transmit the processed text to the quality analysis and prediction module; The quality analysis and prediction module is used to input the processed raw material information text, kneading process parameter text, and environmental parameter text into the long short-term memory network model and the random forest model, and perform weighted fusion on the prediction results of the long short-term memory network model and the random forest model to obtain the final kneading quality prediction result; The control instruction generation module is used to compare the prediction results of the quality analysis and prediction module with the preset quality standard range. If the prediction results do not meet the preset standard, a corresponding control instruction is generated based on the deviation analysis and transmitted to the actuator of the kneading equipment to adjust the parameters of the kneading process; The execution module is used to adjust the kneading equipment in real time according to the control instructions generated by the control decision module to ensure that the kneading process is carried out according to the predetermined process parameters; The feedback adjustment module: after the execution module executes the control instruction, it continuously collects the actual data of the kneading process and the quality detection data of the carbon anode, feeds these data back to the quality analysis and prediction module, optimizes and adjusts the quality prediction model and control strategy, and realizes dynamic control of the kneading quality.

2. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The raw material information text includes the particle size distribution of petroleum coke (D 10 、D 50 、D 90 ), where D 10 Indicates the particle size corresponding to when the cumulative particle size distribution percentage reaches 10%, D 50 is the median particle size, D 90 Indicates the particle size corresponding to the cumulative particle size distribution percentage reaching 90%, the ash content of petroleum coke (A pc ), volatile matter content (V pc ), sulfur content (S pc ), softening point of asphalt (T s ), needle penetration (P), coking value (C v ).

3. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The kneading process parameters include: kneading temperature (T), in degrees Celsius (°C), measured by a high-precision thermocouple, kneading time (t), in minutes (min), recorded by a timing device, stirring speed (n), in revolutions per minute (r / min), obtained by a speed sensor, pressure (p), in megapascals (MPa), monitored by a pressure sensor, and torque (M), in Newton meters (N·m), measured by a torque sensor; The environmental parameters include: ambient temperature (T e ) and humidity (H e ), and collected using temperature and humidity sensors.

4. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The state update formula of the LSTM unit is as follows: Input gate (i t ):i t =σ(W ii x t +W hi h t-1 +b i ) Forget gate (f t ):f t =σ(W if x t +W hf h t-1 +b f ) Output gate (o t ):o t =σ(W io x t +W ho h t-1 +b o ) Cell state update (C t ): C t =f t ⊙C t-1 +i t ⊙tanh(W ic x t +W hc h t-1 +b c ) Hidden state update (h t ):h t =o t ☉tanh(C t ) Among them, x t is the input at the current moment, h t-1 is the hidden state of the previous moment, C t-1 is the cell state at the previous moment, W is the weight matrix, b is the bias vector, σ is the sigmoid function, and ⊙ represents element-by-element multiplication.

5. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The steps of the random forest model are as follows: Decision tree generation: Each decision tree in a random forest is trained based on a subset obtained through bootstrap sampling. Bootstrap sampling refers to randomly extracting samples from the original dataset with replacement to form a new subset of the same size as the original dataset. For each decision tree, when splitting at each node, rather than considering all features, a subset of features is randomly selected from the original dataset. The optimal features and splitting points are then selected from these features to split the node. Assuming that there are m features in the original dataset, the size of the feature subset selected is usually m. Feature selection criteria: In the process of node splitting of the decision tree, Gini impurity is used as the criterion for feature selection. For a data set D, the calculation formula of its Gini impurity is: Where k is the number of categories in the data set. For regression problems, the continuous quality index can be divided into several intervals, which are regarded as different "categories". i is the proportion of samples of the i-th category in the data set D; For a feature A, suppose it has v different values ​​{a1, a2, ..., a v }, divide the data set D according to the value of feature A, and obtain v subsets {D1,D2,…,D v }, then the Gini index of feature A for data set D is: At each node, select the feature with the smallest Gini index and the corresponding split point for splitting. Random forest prediction: For a new sample, input it into each decision tree in the random forest to obtain the prediction result of each decision tree. For regression problems, the final prediction result of the random forest is the average of the prediction results of all decision trees. Assuming that there are N decision trees in the random forest, the prediction result of the jth decision tree for the sample is The final prediction result of random forest is for:

6. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The weighted fusion formula is: in, is the final predicted value, is the predicted value of the LSTM model, is the predicted value of the RF model, w1 and w2 are weight coefficients, and w1+w2=1. The weight coefficients are optimized and determined by the cross-validation method; The predicted mixing quality indicators include the density (ρ), compressive strength (Rc), and resistivity (ρe) of the carbon anode.

7. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The quality standard range refers to setting a reasonable target value and allowable fluctuation range for each kneading quality indicator; The deviation analysis refers to calculating the deviation between the quality prediction value and the target value. The deviation calculation formula is as follows: in is the predicted value, and y0 is the target value.

8. The artificial intelligence-based carbon anode kneading quality control system according to claim 1, characterized in that: The execution module specifically performs the following actions: Adjust the power of the heating device to control the kneading temperature, adjust the speed of the stirring motor to change the stirring speed, control the opening of the feed valve, and adjust the feed amount of raw materials.

Images