Sintered ore quality prediction system and method based on BP neural network
Through the combination of BP neural network with multiple sensors, real-time prediction and optimization of sintered ore quality is achieved, which solves the problems of low prediction accuracy and poor timeliness of traditional methods, and improves the production efficiency and market competitiveness of steel enterprises.
Patent Information
- Application Number
- CN202411063044.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional sintered ore quality prediction methods rely on manual experience and cumbersome laboratory analysis. The prediction accuracy is low and the timeliness is poor, making it difficult to meet the needs of steel enterprises to improve production efficiency and reduce costs.
Using BP neural network combined with thermocouple sensors, electromagnetic flowmeters, X-ray fluorescence spectrometers and other equipment, through data acquisition, preprocessing, model construction and training, real-time prediction and monitoring of sintered ore quality, adjust production parameters, and improve sintered ore quality and output.
It has achieved rapid and accurate prediction and optimization of sintered ore quality, improved production efficiency, reduced energy consumption and environmental pollution, reduced production costs, and enhanced the market competitiveness and social value of steel enterprises.
Smart Images

Figure CN118982283A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of sintered ore quality prediction, in particular to a sintered ore quality prediction system and method based on BP neural network. Background Art
[0002] Sintered ore is an important raw material for steel smelting, and its quality directly affects the efficiency and cost of blast furnace smelting. Traditional sintered ore quality prediction methods often rely on manual experience and cumbersome laboratory analysis, and have problems such as low prediction accuracy and poor timeliness. BP neural network, as a powerful machine learning method, can automatically learn and identify the complex relationship between sintered ore quality and production factors, and achieve fast and accurate prediction. Through the research of this project, a more intelligent and efficient sintered ore quality prediction model can be developed to provide strong support for the production decision-making of steel enterprises and promote the innovation and optimization of sintered ore production technology.
[0003] With the intensification of market competition and the improvement of environmental protection requirements, steel companies need to continuously improve production efficiency and reduce production costs to maintain their competitive advantage. Through the sinter quality prediction model based on BP neural network, real-time prediction and monitoring of sinter quality can be achieved, helping operators to adjust production parameters in time and optimize the sintering process. This can not only improve the quality and output of sinter, but also reduce energy consumption and environmental pollution, and reduce production costs. By introducing artificial intelligence technologies such as BP neural network, intelligent control and optimization of the sinter production process can be achieved, and the automation and intelligence level of the production process can be improved. This can not only improve the production efficiency and quality of steel enterprises, but also create more commercial value and social value for enterprises.
[0004] In summary, the research significance and necessity of the sintered ore quality prediction model project based on BP neural network are very significant. Summary of the invention
[0005] The purpose of the present invention is to provide a sintered ore quality prediction system and method based on BP neural network, by applying the BP neural network, a classic machine learning algorithm, to the field of sintered ore quality prediction to timely discover problems in the production process, and adjust production parameters accordingly, thereby improving the quality and output of sintered ore and reducing production costs.
[0006] To achieve the above-mentioned purpose, the present invention provides the following technical scheme: a sintered ore quality prediction system based on BP neural network, characterized in that it includes a data acquisition module, a data preprocessing module, a model building and training module, a model evaluation and updating module and a prediction output module; the data acquisition module collects data on combustion temperature, flow, combustion time and FeO and CaO content through thermocouple sensors, electromagnetic flowmeters, timers and X-ray fluorescence spectrometers, and transmits the data to the data preprocessing module; the data cleaning unit in the data preprocessing module removes abnormal, repeated and erroneous data, and supplements missing data through a linear interpolation method; the data standardization unit imports the data into the model building and training module after Min-Max standardization processing. ; This module sets the input and output layer nodes, sets the hidden layer, and divides the data into training set (70%), validation set (20%) and test set (10%). It randomly initializes the weights and biases, and adjusts the parameters by calculating the error back propagation until the validation set error reaches the standard and the training is ended. The model is then imported into the evaluation and update module; This module evaluates the model with the test set data. If it does not meet the expectations, it returns to the training module for adjustment. If it meets the expectations, it is evaluated regularly. The trained model is used in the prediction output module to quickly calculate the prediction results using the collected data, and the quality indicators are displayed over time through a line graph. The quality differences of different batches are compared with a bar chart, and the composition ratio of the quality indicators is presented with a pie chart. At the same time, the quality is evaluated, suggestions are provided, and timely warnings are issued when problems occur.
[0007] The data acquisition module includes a thermocouple sensor, an electromagnetic flowmeter and a timer, which respectively record the combustion temperature, flow rate and combustion time of the sintered ore. The characteristic of the data acquisition module is that it also includes a sampling device installed on the sintered ore conveyor belt, which starts sampling at a fixed time, grinds and crushes the sample, and then measures its FeO and CaO content data through an X-ray fluorescence spectrometer; through the interface established with the data preprocessing module, the transmission of operating parameters and quality test data is realized.
[0008] The data preprocessing module includes a data cleaning unit and a data standardization unit. The data cleaning unit collects data transmitted by the data acquisition module through the interface for data cleaning; sets thresholds to judge the rationality of combustion temperature, flow rate, combustion time and FeO and CaO content, removes data points that obviously deviate from the normal range, and uses linear interpolation method to supplement the missing data. The supplemented data is imported into the data standardization unit. The data standardization unit uses Min-Max standardization processing for each feature. The calculation formula is X'=(X-Xmin) / (Xmax-Xmin), where X is the original data value, Xmin is the minimum value in the data, and Xmax is the maximum value in the data. The obtained data is imported into the model construction and training module; The model construction and training module includes setting FeO, CaO content, combustion temperature, flow rate and combustion time as input layer nodes, and the strength, grade and particle size of sintered ore as output layer nodes, and setting 4 hidden layer nodes at the same time; sorting the data, wherein a training set of 70% is set for model learning; a validation set of 20% is used for adjusting hyperparameters during the training process; a test set of 10% is used for finally evaluating the generalization ability of the model, and does not participate in the adjustment of the model during the training process; randomly initialize the weight and bias of the input data of the BP neural network, input the sorted data into the data through the network to obtain the output prediction value, and calculate the error by comparing with the actual output value (the real measured value of the strength, grade and particle size), and according to the error, back propagate from the output layer to the input layer, adjust the weight and bias of the network, and repeat the above steps continuously. When the error of the validation set drops below a preset threshold, the training is terminated, and the trained model is imported into the model evaluation and update module.
[0009] The model evaluation and update module imports the BP neural network model that has been completed in the training phase and meets the end conditions into the evaluation and update module, reads the pre-divided test set data, including input features (FeO content, CaO content, combustion temperature, flow rate, combustion time) and corresponding true output values (strength, grade, particle size), uses the imported model to predict the input data of the test set, and obtains the predicted output value (the predicted value of the strength, grade and particle size of the sintered ore), compares the predicted value with the true output value of the test set, calculates the average of the square of the difference between the calculated predicted value and the actual value, and analyzes and evaluates the performance of the model based on the calculated evaluation index value; if the model performance does not meet expectations, it is necessary to return to the model construction and training module to readjust the training parameters, and regularly use newly collected data to evaluate the existing model for the model that meets expectations; if it is found that the model performance has declined and the evaluation index has significantly deteriorated, the model is retrained with a new data set, and the weight and bias are adjusted to optimize the model performance.
[0010] The prediction output module uses the data collected by the data acquisition module and the trained BP neural network model to quickly calculate and obtain the prediction results of the sintered ore quality, and uses a line graph to display the changing trend of the quality indicators over time, a bar graph to compare the quality differences of different batches, and a pie chart to present the composition ratio of the quality indicators. At the same time, it displays the estimated values of the key indicators of the strength, grade and particle size of the sintered ore. The quality of the sintered ore can be evaluated through the prediction results, such as judging whether it meets specific quality standards or customer needs, and providing real-time adjustment suggestions for the production process. When the prediction results show that quality problems may occur, an early warning signal is issued to remind the operator to take corresponding measures.
[0011] Establish a database to classify and store all data during the collection, preprocessing, training and evaluation processes to ensure the security, integrity and traceability of the data.
[0012] A sinter quality prediction method based on BP neural network comprises the following steps: S1, data collection: The data of combustion temperature, flow rate and combustion time are collected through thermocouple sensors, electromagnetic flowmeters, timers and X-ray fluorescence spectrometers. In addition, there is a sampling device installed on the sintering ore conveyor belt, which starts sampling at a fixed time. After grinding and crushing the samples, the FeO and CaO content data are measured by X-ray fluorescence spectrometer. These relevant data are interfaced with the data preprocessing module to realize the transmission of operation parameters and quality test data; S2, data preprocessing: The data cleaning unit removes abnormal, repeated and erroneous data, and supplements missing data through linear interpolation. The data standardization unit uses Min-Max standardization to process data, and the calculation formula is X'=(X-Xmin) / (Xmax-Xmin); the processed data is imported into the model construction and training module; S3, model construction and training: set the input and output layer nodes, set 4 hidden layers. Organize the data into training set (70%), validation set (20%) and test set (10%); randomly initialize weights and biases, calculate the error back propagation and adjust the parameters until the validation set error reaches the standard and the training ends; S4, model evaluation and update: Use the test set data to evaluate the model, and calculate the average of the square of the difference between the predicted value and the actual value; if it does not meet expectations, return to the training module for adjustment, and if it meets expectations, evaluate it regularly. If performance is found to have degraded, retrain with a new data set; S5, prediction output: Use the trained model to quickly calculate the prediction results of the collected data, use line graphs to show the changing trend of quality indicators over time, use bar graphs to compare the quality differences of different batches, and use pie charts to show the composition ratio of quality indicators, evaluate quality, provide suggestions, and warn of problems; The data collection step also collects historical data and real-time data in the sintering ore production process by cooperating with enterprises.
[0013] In the actual production process, the operating status and prediction results of the model are monitored in real time. If it is found that there is a large deviation between the prediction results of the model and the actual situation, an alarm will be issued in time and manual intervention will be carried out. At the same time, these new data will be fed back to the model to continuously optimize the model.
[0014] During the data collection phase, thermocouple sensors, electromagnetic flowmeters and timers are installed at key locations in the sintering production process to collect data at specific frequencies. At the same time, the sampling device installed on the sintering ore conveyor belt is started every 30 minutes. After fine grinding and crushing of the collected samples, the FeO and CaO content data are accurately measured by X-ray fluorescence spectrometer. These devices and instruments transmit the collected operating parameters and quality test data to the data preprocessing module in real time and stably through a specially established interface. Entering the data preprocessing phase, the data cleaning unit will operate strictly in accordance with the pre-set rules and thresholds. Data points exceeding the threshold will be judged as abnormal and eliminated. Duplicate data will be identified and removed. Missing data will be supplemented by linear interpolation of temperature data at adjacent moments. The data standardization unit uses the Min-Max standardization method to process the cleaned data and map it to the interval of 0 to 1 to prepare for subsequent model training. In the model building and training module, FeO, CaO content, combustion temperature, flow rate and combustion time are explicitly used as input layer nodes, and the strength, grade and particle size of sintered ore are used as output layer nodes. The large amount of data collected is divided into 70% groups as training sets, 20% groups as validation sets, and 10% groups as test sets. When the weights and biases are randomly initialized, the sorted data is then input into the network, the output prediction value is obtained by calculation, and it is compared with the actual output value (real measured value of strength, grade, and particle size) to calculate the error. According to the error, backpropagation is carried out from the output layer to the input layer, and the weights and biases of the network are continuously adjusted. This process is repeated until the error of the validation set drops below the preset threshold to end the training. In the model evaluation and update module, the pre-divided test set data is used, including input features (FeO content, CaO content, combustion temperature, flow rate, combustion time) and corresponding real output values (strength, grade, particle size). Use the trained model to predict the input data of the test set and obtain the predicted output value (the predicted value of the strength, grade and particle size of the sintered ore). Compare the predicted value with the actual output value of the test set, and calculate the average of the square of the difference between the predicted value and the actual value. If the model performance does not meet expectations, it is necessary to return to the model building and training module to readjust the training parameters by adjusting the number of hidden layer nodes, learning rate, etc. For models that meet expectations, use newly collected data to evaluate the existing model every two weeks. If it is found that the model performance has declined and the evaluation indicators have significantly deteriorated, use a new data set to retrain the model, adjust the weights and biases, and optimize the model performance. In the prediction output module, the trained model is used to quickly calculate the newly collected data to obtain accurate sintered ore quality prediction results to accurately evaluate the quality of the sintered ore, determine whether it meets specific quality standards or customer needs, and provide real-time and effective adjustment suggestions for the production process.When the prediction results show that there may be quality problems, early warning signals are used to remind operators to take appropriate measures quickly. In the actual production process, a dedicated monitoring system is used to monitor the operating status and prediction results of the model in real time and continuously.
[0015] Compared with the prior art, the present invention has the following beneficial effects: Applying the classic machine learning algorithm, BP neural network, to the field of sinter quality prediction is an innovation in the algorithm application scenario. The sinter quality prediction model based on BP neural network can be used as a part of an intelligent decision support system to provide real-time quality prediction and optimization suggestions for the sintering process.
[0016] Through data collection, including combustion temperature, flow rate, combustion time, FeO and CaO content, combined with scientific data preprocessing and reasonable model construction and training, the quality of sintered ore can be predicted more accurately.
[0017] The forecast results are presented in a variety of intuitive charts, including line charts, bar charts and pie charts, providing clear and easy-to-understand information support for production decisions, allowing managers to quickly understand quality trends and differences and make reasonable decisions.
[0018] Establishing a database to classify and store data from each link ensures the security, integrity and traceability of the data, which is conducive to subsequent data analysis and improvement, and accumulates valuable data resources for long-term production optimization and quality improvement.
[0019] The model's operating status and prediction results are monitored in real time, and the model is continuously optimized by feeding back new data so that it can adapt to changes in the production process, solving the problem that traditional models are fixed and difficult to cope with changes in production conditions.
[0020] Timely warning and adjustment can help maintain the stability of sintered ore quality and meet specific quality standards and customer needs. Stable quality can improve the company's market competitiveness and reputation. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 This is a system diagram of a sintered ore quality prediction system and method based on a BP neural network of the present invention; Figure 2 This is a step diagram of a sintered ore quality prediction system and method based on BP neural network of the present invention; Specific implementation methods
[0022] The technical solutions in the embodiments of the present invention will be fully described below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0023] As shown in the figure, a sinter quality prediction system based on BP neural network is characterized by comprising a data acquisition module, a data preprocessing module, a model building and training module, a model evaluation and updating module and a prediction output module; the data acquisition module collects data on combustion temperature, flow rate, combustion time and FeO and CaO content through thermocouple sensors, electromagnetic flowmeters, timers and X-ray fluorescence spectrometers, and transmits the data to the data preprocessing module; the data cleaning unit in the data preprocessing module removes abnormal, repeated and erroneous data, and supplements missing data through linear interpolation method; the data standardization unit imports the data into the model building and training module after Min-Max standardization processing; the module sets the input Enter the output layer node, set the hidden layer, and divide the data into training set (70%), validation set (20%) and test set (10%). Randomly initialize the weights and biases, adjust the parameters by calculating the error back propagation, and end the training when the validation set error reaches the standard. Then import the model into the evaluation and update module; this module evaluates the model with the test set data. If it does not meet the expectations, it returns to the training module for adjustment. If it meets the expectations, it is evaluated regularly. The trained model is used in the prediction output module to quickly calculate the prediction results using the collected data, and the changing trend of quality indicators over time is displayed through a line graph, and the quality differences of different batches are compared with a bar graph. The composition ratio of quality indicators is presented with a pie chart. At the same time, the quality is evaluated, suggestions are provided, and timely warnings are issued when problems occur.
[0024] The data acquisition module includes a thermocouple sensor, an electromagnetic flowmeter and a timer, which respectively record the combustion temperature, flow rate and combustion time of the sintered ore. The characteristic of the data acquisition module is that it also includes a sampling device installed on the sintered ore conveyor belt, which starts sampling at a fixed time, grinds and crushes the sample, and then measures its FeO and CaO content data through an X-ray fluorescence spectrometer; through the interface established with the data preprocessing module, the transmission of operating parameters and quality test data is realized.
[0025] The data preprocessing module includes a data cleaning unit and a data standardization unit. The data cleaning unit collects data transmitted by the data acquisition module through the interface for data cleaning; sets thresholds to judge the rationality of combustion temperature, flow rate, combustion time and FeO and CaO content, removes data points that obviously deviate from the normal range, and uses linear interpolation method to supplement the missing data. The supplemented data is imported into the data standardization unit. The data standardization unit uses Min-Max standardization processing for each feature. The calculation formula is X'=(X-Xmin) / (Xmax-Xmin), where X is the original data value, Xmin is the minimum value in the data, and Xmax is the maximum value in the data. The obtained data is imported into the model construction and training module; The model construction and training module includes setting FeO, CaO content, combustion temperature, flow rate and combustion time as input layer nodes, and the strength, grade and particle size of sintered ore as output layer nodes, and setting 4 hidden layer nodes at the same time; sorting the data, wherein a training set of 70% is set for model learning; a validation set of 20% is used for adjusting hyperparameters during the training process; a test set of 10% is used for finally evaluating the generalization ability of the model, and does not participate in the adjustment of the model during the training process; randomly initialize the weight and bias of the input data of the BP neural network, input the sorted data into the data through the network to obtain the output prediction value, and calculate the error by comparing with the actual output value (the real measured value of the strength, grade and particle size), and according to the error, back propagate from the output layer to the input layer, adjust the weight and bias of the network, and repeat the above steps continuously. When the error of the validation set drops below a preset threshold, the training is terminated, and the trained model is imported into the model evaluation and update module.
[0026] The model evaluation and update module imports the BP neural network model that has been completed in the training phase and meets the end conditions into the evaluation and update module, reads the pre-divided test set data, including input features (FeO content, CaO content, combustion temperature, flow rate, combustion time) and corresponding true output values (strength, grade, particle size), uses the imported model to predict the input data of the test set, and obtains the predicted output value (the predicted value of the strength, grade and particle size of the sintered ore), compares the predicted value with the true output value of the test set, calculates the average of the square of the difference between the calculated predicted value and the actual value, and analyzes and evaluates the performance of the model based on the calculated evaluation index value; if the model performance does not meet expectations, it is necessary to return to the model construction and training module to readjust the training parameters, and regularly use newly collected data to evaluate the existing model for the model that meets expectations; if it is found that the model performance has declined and the evaluation index has significantly deteriorated, the model is retrained with a new data set, and the weight and bias are adjusted to optimize the model performance.
[0027] The prediction output module uses the data collected by the data acquisition module and the trained BP neural network model to quickly calculate and obtain the prediction results of the sintered ore quality, and uses a line graph to display the changing trend of the quality indicators over time, a bar graph to compare the quality differences of different batches, and a pie chart to present the composition ratio of the quality indicators. At the same time, it displays the estimated values of the key indicators of the strength, grade and particle size of the sintered ore. The quality of the sintered ore can be evaluated through the prediction results, such as judging whether it meets specific quality standards or customer needs, and providing real-time adjustment suggestions for the production process. When the prediction results show that quality problems may occur, an early warning signal is issued to remind the operator to take corresponding measures.
[0028] Establish a database to classify and store all data during the collection, preprocessing, training and evaluation processes to ensure the security, integrity and traceability of the data.
[0029] A sinter quality prediction method based on BP neural network comprises the following steps: S1, data collection: The data of combustion temperature, flow rate and combustion time are collected through thermocouple sensors, electromagnetic flowmeters, timers and X-ray fluorescence spectrometers. In addition, there is a sampling device installed on the sintering ore conveyor belt, which starts sampling at a fixed time. After grinding and crushing the samples, the FeO and CaO content data are measured by X-ray fluorescence spectrometer. These relevant data are interfaced with the data preprocessing module to realize the transmission of operation parameters and quality test data; S2, data preprocessing: The data cleaning unit removes abnormal, repeated and erroneous data, and supplements missing data through linear interpolation. The data standardization unit uses Min-Max standardization to process data, and the calculation formula is X'=(X-Xmin) / (Xmax-Xmin); the processed data is imported into the model construction and training module; S3, model construction and training: set the input and output layer nodes, set 4 hidden layers. Organize the data into training set (70%), validation set (20%) and test set (10%); randomly initialize weights and biases, calculate the error back propagation and adjust the parameters until the validation set error reaches the standard and the training ends; S4, model evaluation and update: Use the test set data to evaluate the model, and calculate the average of the square of the difference between the predicted value and the actual value; if it does not meet expectations, return to the training module for adjustment, and if it meets expectations, evaluate it regularly. If performance is found to have degraded, retrain with a new data set; S5, prediction output: Use the trained model to quickly calculate the prediction results of the collected data, use line graphs to show the changing trend of quality indicators over time, use bar graphs to compare the quality differences of different batches, and use pie charts to show the composition ratio of quality indicators, evaluate quality, provide suggestions, and warn of problems; The data collection step also collects historical data and real-time data in the sintering ore production process by cooperating with enterprises.
[0030] In the actual production process, the operating status and prediction results of the model are monitored in real time. If it is found that there is a large deviation between the prediction results of the model and the actual situation, an alarm will be issued in time and manual intervention will be carried out. At the same time, these new data will be fed back to the model to continuously optimize the model.
[0031] During the data collection phase, thermocouple sensors, electromagnetic flowmeters and timers are installed at key locations in the sintering production process to collect data at specific frequencies. At the same time, the sampling device installed on the sintering ore conveyor belt is started every 30 minutes. After fine grinding and crushing of the collected samples, the FeO and CaO content data are accurately measured by X-ray fluorescence spectrometer. These devices and instruments transmit the collected operating parameters and quality test data to the data preprocessing module in real time and stably through a specially established interface. Entering the data preprocessing phase, the data cleaning unit will operate strictly in accordance with the pre-set rules and thresholds. Data points exceeding the threshold will be judged as abnormal and eliminated. Duplicate data will be identified and removed. Missing data will be supplemented by linear interpolation of temperature data at adjacent moments. The data standardization unit uses the Min-Max standardization method to process the cleaned data and map it to the interval of 0 to 1 to prepare for subsequent model training. In the model building and training module, FeO, CaO content, combustion temperature, flow rate and combustion time are explicitly used as input layer nodes, and the strength, grade and particle size of sintered ore are used as output layer nodes. The large amount of data collected is divided into 70% groups as training sets, 20% groups as validation sets, and 10% groups as test sets. When the weights and biases are randomly initialized, the sorted data is then input into the network, the output prediction value is obtained by calculation, and it is compared with the actual output value (real measured value of strength, grade, and particle size) to calculate the error. According to the error, backpropagation is carried out from the output layer to the input layer, and the weights and biases of the network are continuously adjusted. This process is repeated until the error of the validation set drops below the preset threshold to end the training. In the model evaluation and update module, the pre-divided test set data is used, including input features (FeO content, CaO content, combustion temperature, flow rate, combustion time) and corresponding real output values (strength, grade, particle size). Use the trained model to predict the input data of the test set and obtain the predicted output value (the predicted value of the strength, grade and particle size of the sintered ore). Compare the predicted value with the actual output value of the test set, and calculate the average of the square of the difference between the predicted value and the actual value. If the model performance does not meet expectations, it is necessary to return to the model building and training module to readjust the training parameters by adjusting the number of hidden layer nodes, learning rate, etc. For models that meet expectations, use newly collected data to evaluate the existing model every two weeks. If it is found that the model performance has declined and the evaluation indicators have significantly deteriorated, use a new data set to retrain the model, adjust the weights and biases, and optimize the model performance. In the prediction output module, the trained model is used to quickly calculate the newly collected data to obtain accurate sintered ore quality prediction results to accurately evaluate the quality of the sintered ore, determine whether it meets specific quality standards or customer needs, and provide real-time and effective adjustment suggestions for the production process.When the prediction results show that there may be quality problems, early warning signals are used to remind operators to take appropriate measures quickly. In the actual production process, a dedicated monitoring system is used to monitor the operating status and prediction results of the model in real time and continuously.
Claims
1. A sinter quality prediction system based on BP neural network, characterized in that It includes a data acquisition module, a data preprocessing module, a model building and training module, a model evaluation and updating module and a prediction output module; the data acquisition module collects data on combustion temperature, flow, combustion time and FeO and CaO content through thermocouple sensors, electromagnetic flowmeters, timers and X-ray fluorescence spectrometers, and transmits the data to the data preprocessing module; the data cleaning unit in the data preprocessing module removes abnormal, repeated and erroneous data, and supplements missing data through a linear interpolation method; the data standardization unit uses Min-Max standardization processing and then imports the data into the model building and training module; the module sets the input and output layer nodes, sets the hidden layer, The data is divided into 70% for training, 20% for validation and 10% for testing. The weights and biases are randomly initialized, and the parameters are adjusted by calculating the error back propagation until the validation set error reaches the standard and the training is ended. The model is then imported into the evaluation and update module. This module uses the test set data to evaluate the model. If it does not meet expectations, it is returned to the training module for adjustment. If it meets expectations, it is evaluated regularly. The trained model is used in the prediction output module to quickly calculate the prediction results using the collected data. The changing trend of quality indicators over time is displayed through a line graph, the quality differences of different batches are compared with a bar graph, and the composition ratio of quality indicators is presented with a pie chart. At the same time, the quality is evaluated, suggestions are provided, and timely warnings are issued when problems occur.
2. The sintered ore quality prediction system based on BP neural network according to claim 1 is characterized in that: The data acquisition module includes a thermocouple sensor, an electromagnetic flowmeter and a timer, which respectively record the combustion temperature, flow rate and combustion time of the sintered ore. The characteristic of the data acquisition module is that it also includes a sampling device installed on the sintered ore conveyor belt, which starts sampling at a fixed time, grinds and crushes the sample, and then measures its FeO and CaO content data through an X-ray fluorescence spectrometer; through the interface established with the data preprocessing module, the transmission of operating parameters and quality test data is realized.
3. The sintered ore quality prediction system based on BP neural network according to claim 1, characterized in that: The data preprocessing module includes a data cleaning unit and a data standardization unit. The data cleaning unit collects data transmitted by the data acquisition module through the interface for data cleaning; sets thresholds to judge the rationality of combustion temperature, flow rate, combustion time and FeO, CaO content, removes data points that obviously deviate from the normal range, and uses a linear interpolation method to supplement the missing data. The supplemented data is imported into the data standardization unit. The data standardization unit uses Min-Max standardization processing for each feature. The calculation formula is X'=(X-Xmin) / (Xmax-Xmin), where X is the original data value, Xmin is the minimum value in the data, and Xmax is the maximum value in the data. The obtained data is imported into the model construction and training module.
4. The sintered ore quality prediction system based on BP neural network according to claim 1, characterized in that: The model building and training module includes setting FeO, CaO content, combustion temperature, flow rate and combustion time as input layer nodes, the strength, grade and particle size of sintered ore as output layer nodes, and setting 4 hidden layer nodes; sorting the data, wherein 70% of the training set is set for model learning; The validation set is 20%, which is used to adjust the hyperparameters during the training process. The test set is 10%, which is used to finally evaluate the generalization ability of the model and does not participate in the adjustment of the model during the training process. The weights and biases of the input data of the BP neural network are randomly initialized, and the sorted data is input into the network to obtain the output prediction value. The error is calculated by comparing it with the actual measured values of the output value intensity, grade, and granularity. According to the error, it is back-propagated from the output layer to the input layer to adjust the weights and bias of the network. The above steps are repeated continuously. When the error of the validation set drops below the preset threshold, the training is terminated and the trained model is imported into the model evaluation and update module.
5. The sintered ore quality prediction system based on BP neural network according to claim 1 is characterized in that: The model evaluation and update module imports the BP neural network model that has been completed in the training phase and meets the end conditions into the evaluation and update module, reads the pre-divided test set data, including input features: FeO content, CaO content, combustion temperature, flow rate, combustion time, and corresponding real output values: strength, grade, and particle size, uses the imported model to predict the input data of the test set, and obtains the predicted output value of the strength, grade and particle size of the sintered ore, compares the predicted value with the real output value of the test set, calculates the average of the square of the difference between the calculated predicted value and the actual value, and analyzes and evaluates the performance of the model based on the calculated evaluation index value; if the model performance does not meet expectations, it is necessary to return to the model construction and training module to readjust the training parameters, and regularly use newly collected data to evaluate the existing model for the model that meets expectations; if it is found that the model performance has declined and the evaluation index has obviously deteriorated, the model is retrained with a new data set, and the weight and bias are adjusted to optimize the model performance.
6. The sintered ore quality prediction system based on BP neural network according to claim 1, characterized in that: The prediction output module uses the data collected by the data acquisition module and the trained BP neural network model to quickly calculate and obtain the prediction results of the sintered ore quality, and uses a line graph to display the changing trend of the quality indicators over time, a bar graph to compare the quality differences of different batches, and a pie chart to present the composition ratio of the quality indicators. At the same time, it displays the estimated values of the key indicators of the strength, grade and particle size of the sintered ore. The quality of the sintered ore can be evaluated through the prediction results, such as judging whether it meets specific quality standards or customer needs, and providing real-time adjustment suggestions for the production process. When the prediction results show that quality problems may occur, an early warning signal is issued to remind the operator to take corresponding measures.
7. The sinter quality prediction system based on BP neural network according to claim 1 is characterized in that: Establish a database to classify and store all data during the collection, preprocessing, training and evaluation processes to ensure the security, integrity and traceability of the data.
8. A sinter quality prediction method based on BP neural network, comprising the following steps: S1, data collection: The data of combustion temperature, flow rate and combustion time are collected through thermocouple sensors, electromagnetic flowmeters, timers and X-ray fluorescence spectrometers. In addition, there is a sampling device installed on the sintering ore conveyor belt, which starts sampling at a fixed time. After grinding and crushing the samples, the FeO and CaO content data are measured by X-ray fluorescence spectrometer. These relevant data are interfaced with the data preprocessing module to realize the transmission of operation parameters and quality test data; S2, data preprocessing: The data cleaning unit removes abnormal, duplicate and erroneous data, and supplements missing data through linear interpolation method. The data standardization unit uses Min-Max standardization to process data, and the calculation formula is X'=(X-Xmin) / (Xmax-Xmin); the processed data is imported into the model construction and training module; S3, model construction and training: set the input and output layer nodes, set 4 hidden layers. Organize the data into 70% training set, 20% validation set and 10% test set; randomly initialize weights and biases, calculate the error back propagation and adjust the parameters until the validation set error reaches the standard and the training ends; S4, model evaluation and update: Use the test set data to evaluate the model, and calculate the average of the square of the difference between the predicted value and the actual value; if it does not meet expectations, return to the training module for adjustment, and if it meets expectations, evaluate it regularly. If performance is found to have degraded, retrain with a new data set; S5, prediction output: Use the trained model to quickly calculate the prediction results of the collected data, use a line graph to show the changing trend of quality indicators over time, a bar graph to compare the quality differences between different batches, and a pie chart to show the composition ratio of quality indicators, evaluate quality, provide suggestions, and warn of problems.
9. The method for predicting sintered ore quality based on BP neural network according to claim 8, characterized in that: The data collection step also collects historical data and real-time data in the sintering ore production process by cooperating with enterprises.
10. The method for predicting sintered ore quality based on BP neural network according to claim 8, characterized in that: In the actual production process, the operating status and prediction results of the model are monitored in real time. If it is found that there is a large deviation between the prediction results of the model and the actual situation, an alarm will be issued in time and manual intervention will be carried out. At the same time, these new data will be fed back to the model to continuously optimize the model.
