Smelting cost prediction method and system based on data analysis

Abstract

The invention relates to the technical field of data analysis, and discloses a smelting cost prediction method and system based on data analysis, and the method comprises the following steps: collecting multi-source data of the same metal material and the metal smelted by the same production line in the smelting process, the multi-source data comprising material data, energy consumption data, production operation data and actual marginal cost increase rate; preprocessing the multi-source data, wherein the preprocessing comprises data cleaning, missing value filling, abnormal value detection and data normalization; static features and dynamic features related to the smelting cost are extracted based on the preprocessed data. According to the invention, by fusing the multi-source data and respectively extracting the static and dynamic characteristics by using the convolutional neural network and the recurrent neural network, the stable and changing cost influence factors in the smelting process can be comprehensively captured, so that the comprehensiveness and accuracy of cost prediction are remarkably improved; and the defect of insufficient consideration of time sequence dynamic factors in a traditional method is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and in particular to a smelting cost prediction method and system based on data analysis. BACKGROUND

[0002] Current smelting cost prediction methods mostly rely on traditional statistical models or analysis of a single data source, and are difficult to comprehensively reflect the dynamic influence of multiple factors such as materials, energy consumption and operation in the production process, and are not adaptive to time sequence changes and working condition fluctuations, resulting in limited prediction accuracy and weak generalization ability, and failing to meet the needs of fine cost control of modern smelting enterprises.

[0003] An existing patent discloses a steelmaking cost prediction method and system (CN120410580A). The disclosed technology in the existing patent mostly relies on manual experience or a single data source, and is difficult to comprehensively and dynamically reflect the influence of multiple factors such as materials, energy consumption and operation in the production process on the cost; ignores the key role of time sequence changes and working condition fluctuations on the cost, resulting in limited prediction accuracy and weak generalization ability; at the same time, the existing model is mostly a "black box" operation, lacks the fusion of physical and chemical mechanisms of the smelting process, and cannot realize fine, adaptive and feedback cost control. SUMMARY

[0004] The present application provides a smelting cost prediction method and system based on data analysis to solve the existing technical problems, and solves the problem of insufficient consideration of time sequence dynamic factors in traditional methods.

[0005] To solve the above technical problems, according to one aspect of the present application, more specifically a smelting cost prediction method based on data analysis, comprising the following steps: S1, collecting multi-source data of the same metal material and the same production line smelted metal in the smelting process, the multi-source data including material data, energy consumption data, production operation data and actual marginal cost increase rate; S2, preprocessing the multi-source data, including data cleaning, missing value filling, abnormal value detection and data normalization; S3, extracting static features and dynamic features related to smelting cost based on the preprocessed data, using a convolutional neural network to extract static features and using a recurrent neural network to extract dynamic features; S4, constructing a smelting cost prediction model of the same metal material and the same production line according to the historical feature data stored in the database, and optimizing the hyperparameters of the smelting cost prediction model using an optimization algorithm; S5, based on the optimized smelting cost prediction model, predicting the marginal cost increase rate of the current smelted metal according to the additional material loss coefficient, the additional energy consumption efficiency coefficient and the additional production operation cost rate of the current smelted metal. S6, feeding back the marginal cost increase rate of the current smelting metal prediction and the subsequent actual generated marginal cost increase rate, and then optimizing the hyperparameters of the smelting cost prediction model again through an optimization algorithm.

[0006] Further, the static features represent features that are relatively stable or do not change over time during the smelting process, including the basic material loss coefficient extracted based on the material data, the basic energy consumption efficiency coefficient extracted based on the energy consumption data, the basic production operation cost rate extracted based on the production operation data, and the fixed correlation between these coefficients and the actual marginal cost increase rate; The dynamic features represent features that change over time during the smelting process, including real-time material loss coefficient changes, real-time energy consumption efficiency coefficient changes, and real-time production operation cost rate changes extracted based on time series of material data, energy consumption data, and production operation data, and dynamic impact sequences of these changes on the actual marginal cost increase rate.

[0007] Further, in step S3, the specific steps of extracting static features using a convolutional neural network are as follows: (1) Align and splice the fields related to static attributes in the preprocessed material data, energy consumption data, and production operation data to form a multi-dimensional static data matrix, and reconfigure it into a two-dimensional or three-dimensional tensor format that meets the input requirements of the convolutional neural network; (2) Input the constructed input tensor into the convolutional neural network, which contains multiple convolutional layers, each using different size convolution kernels for multi-channel convolution operations to capture local static feature patterns at different scales; (3) Apply an activation function to the feature map output by the convolutional layer to introduce a nonlinear transformation; then use a pooling layer to downsample the feature map, preserving significant features while reducing data dimensions and enhancing feature translation invariance; (4) Concatenate and fuse the feature maps after multi-layer convolution and pooling processing, further abstract high-level static feature representations through a fully connected layer, and flatten the high-level static feature representations into a one-dimensional static feature vector; (5) Output the one-dimensional static feature vector obtained finally as the extracted static features, which are used for subsequent construction of the smelting cost prediction model.

[0008] Further, in step S3, the specific steps of extracting dynamic features using a recurrent neural network are as follows: (1) From the pretreated multi-source data, the time sequence variables related to dynamic attributes in the material data, energy consumption data and production operation data are extracted in time sequence, a plurality of parallel time series data streams are constructed, and a three-dimensional tensor format suitable for recurrent neural network processing is organized, and the dimensions represent the number of samples, time steps and feature dimensions respectively; (2) The constructed time series tensor is input into the recurrent neural network, the network adopts long short-term memory network or gated recurrent unit structure, and the input sequence is processed in time steps, and the long-term dependence relationship and time sequence dynamic mode in the data are captured through the internal gating mechanism; (3) At each time step, the recurrent neural network calculates the current hidden state according to the current input and the hidden state at the previous time, so as to encode the context information of the entire time sequence into the final hidden state sequence; (4) The hidden states of all time steps output by the recurrent neural network are aggregated, and the attention mechanism or the hidden state of the last time step is used as the summary representation of the entire sequence to generate a comprehensive dynamic feature vector containing time sequence dynamic information; (5) The finally generated dynamic feature vector is output as the extracted dynamic feature, which is used together with the static feature vector for the construction of the subsequent smelting cost prediction model.

[0009] Further, in step S4, the specific steps of constructing the smelting cost prediction model are: S401, collect the historical feature data of the same metal and the same production line of the smelted metal; S402, control the variables of the additional increased energy consumption efficiency coefficient and the additional increased production operation cost rate, collect the additional increased material loss coefficient and the marginal cost increase rate of the smelted metal, and mark the marginal cost increase rate of the smelted metal at this time as marginal cost increase rate A; construct a loss prediction function according to the dynamic feature relationship between the marginal cost increase rate A and the additional increased material loss coefficient; S403, control the variables of the additional increased material loss coefficient and the additional increased energy consumption efficiency coefficient, collect the additional increased production operation cost rate and the marginal cost increase rate of the smelted metal, and mark the marginal cost increase rate of the smelted metal at this time as marginal cost increase rate B; construct an operation cost prediction function according to the dynamic feature relationship between the marginal cost increase rate B and the additional increased production operation cost rate; S404, control the additional material loss coefficient, the variable of the additional production operation cost rate, collect the marginal cost increase rate of the additional energy consumption efficiency coefficient and the smelted metal, and mark the marginal cost increase rate of the smelted metal at this time as the marginal cost increase rate C; and construct an energy consumption efficiency prediction function according to the dynamic characteristic relationship between the marginal cost increase rate C and the additional energy consumption efficiency coefficient; S405, based on the dynamic characteristic relationship between the loss prediction function, the operation cost prediction function and the energy consumption efficiency prediction function, a smelting cost prediction model under the same metal material and the same production line is constructed.

[0010] Further, the smelting cost prediction model is integrated and constructed based on the dynamic characteristic relationship between the loss prediction function, the operation cost prediction function and the energy consumption efficiency prediction function, and is used for comprehensively reflecting the prediction model of the marginal cost increase rate of the smelted metal under the same metal material and the same production line with the additional material loss coefficient, the additional energy consumption efficiency coefficient and the additional production operation cost rate.

[0011] Further, in the step S4, the optimization algorithm is used to optimize the hyperparameters of the smelting cost prediction model, which is based on historical characteristic data, automatically searches in the preset hyperparameter search space by selecting the optimization algorithm, and determines the optimal hyperparameter combination according to the model performance index.

[0012] Further, in the step S6, the hyperparameters of the smelting cost prediction model are feedback optimized again by the optimization algorithm, which is based on the error data between the predicted marginal cost increase rate and the subsequent actual marginal cost increase rate, and the hyperparameters of the smelting cost prediction model are re-optimized and updated by the optimization algorithm.

[0013] The smelting cost prediction system based on data analysis comprises a data acquisition module, a data preprocessing module, a feature extraction module, a model construction and optimization module, a cost prediction module and a feedback optimization module.

[0014] The smelting cost prediction method and system based on data analysis provided by the application have the following effects compared with the prior art: 1, the application can fully capture the stable and changing cost influencing factors in the smelting process by fusing multi-source data and using convolutional neural network and recurrent neural network to extract static and dynamic features respectively, thereby significantly improving the comprehensiveness and accuracy of cost prediction, and overcoming the defects of insufficient consideration of time sequence dynamic factors in traditional methods.

[0015] 2. This invention constructs prediction functions for material loss, energy efficiency, and the relationship between production operation costs and marginal cost growth rate using the controlled variable method. Based on these functions, a comprehensive prediction model is built, enabling the model to accurately reflect the independent and interactive effects of various factors on costs under different production conditions. This achieves refined modeling and prediction of cost changes for specific metals and production lines.

[0016] 3. By introducing a feedback optimization mechanism based on prediction error, this invention enables the smelting cost prediction model to continuously adjust its hyperparameters according to actual production data, thereby adapting to changes in production conditions, effectively maintaining the prediction accuracy and stability of the model in long-term operation, and improving the practicality and reliability of the model.

[0017] 4. By constructing a complete system that includes data acquisition, preprocessing, feature extraction, model building and optimization, cost prediction and feedback optimization, this invention achieves modularization and automation of the cost prediction process, greatly improving prediction efficiency and system operability, and providing reliable technical support for enterprises to implement refined cost management in actual production. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 The marginal cost growth rate predicted in this invention A diagram showing the relationship between 'a' and 'a'; Figure 3 The marginal cost growth rate predicted in this invention A diagram showing the relationship between m and m; Figure 4 The marginal cost growth rate predicted in this invention A diagram showing the relationship between s and s. Detailed Implementation

[0019] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1

[0021] like Figure 1 As shown, according to one aspect of the present invention, a data analysis-based method for predicting smelting costs is provided, comprising the following steps: collecting multi-source data on the same metal material and the same metal smelted on the same production line during the smelting process, wherein the multi-source data includes material data, energy consumption data, production operation data, and actual marginal cost growth rate; and preprocessing the multi-source data, including data cleaning, missing value imputation, outlier detection, and data normalization.

[0022] Among them, the static characteristics represent the characteristics relatively stable or not changing with time in the smelting process, including the basic material loss coefficient extracted based on material data, the basic energy consumption efficiency coefficient extracted based on energy consumption data, the basic production operation cost rate extracted based on production operation data, and the fixed correlation between these coefficients and the actual marginal cost increase rate; The dynamic characteristics represent the characteristics changing with time in the smelting process, including the real-time material loss coefficient change, the real-time energy consumption efficiency coefficient change, the real-time production operation cost rate change extracted based on the time series of material data, energy consumption data and production operation data, and the dynamic influence sequence of these changes on the actual marginal cost increase rate.

[0023] Based on the preprocessed data, the static characteristics and dynamic characteristics related to the smelting cost are extracted, the convolutional neural network is used to extract the static characteristics, and the recurrent neural network is used to extract the dynamic characteristics. Then, the specific steps of using the convolutional neural network to extract the static characteristics are as follows: (1) Align and splice the fields related to static attributes in the preprocessed material data, energy consumption data and production operation data, form a multi-dimensional static data matrix, and reconfigure it into a two-dimensional or three-dimensional tensor format conforming to the input requirements of the convolutional neural network; (2) Input the constructed input tensor into the convolutional neural network, which contains multiple convolutional layers, each of which uses different size convolution kernels for multi-channel convolution operation to capture local static feature patterns at different scales; (3) Apply an activation function to the feature map output by the convolutional layer to introduce a nonlinear transformation; then use a pooling layer to downsample the feature map, retaining significant features while reducing data dimensions and enhancing feature translation invariance; (4) Splice and fuse the feature maps after multi-layer convolution and pooling processing, further abstract high-level static feature representations through a fully connected layer, and flatten the high-level static feature representations into a one-dimensional static feature vector; (5) Output the one-dimensional static feature vector finally obtained as the extracted static characteristics, which are used for subsequent construction of smelting cost prediction model.

[0024] Then, the specific steps of using the recurrent neural network to extract the dynamic characteristics are as follows: (1) From the preprocessed multi-source data, extract time series variables related to dynamic attributes in material data, energy consumption data and production operation data in time sequence, construct multiple parallel time series data streams, and organize them into a three-dimensional tensor format suitable for processing by the recurrent neural network, whose dimensions represent the number of samples, time steps and feature dimensions, respectively; (2) Input the constructed time series tensor into a recurrent neural network, which adopts a long short-term memory network or a gated recurrent unit structure, and sequentially processes the input sequence according to the time step, and captures the long-term dependence and time sequence dynamic mode in the data through the internal gating mechanism; (3) At each time step, the recurrent neural network calculates the current hidden state according to the current input and the hidden state at the previous time, so as to encode the context information of the entire time series into the final hidden state sequence; (4) Aggregate the hidden states of all time steps output by the recurrent neural network, and use the attention mechanism or the hidden state of the last time step as the summary representation of the entire sequence to generate a comprehensive dynamic feature vector containing time sequence dynamic information; (5) The finally generated dynamic feature vector is output as the extracted dynamic feature, which is used together with the static feature vector for the construction of the subsequent smelting cost prediction model.

[0025] The expression of the current additional material loss coefficient formula for smelting metal is: ; In the above formula, a represents the additional material loss coefficient, which ranges from 0 to 15%; represents the actual material consumption; represents the planned material consumption. Among them, the additional material loss coefficient a takes a positive number, and the reduced material loss to reduce the cost is not considered.

[0026] The expression of the current additional energy consumption efficiency coefficient formula for smelting metal is: ; In the above formula, s represents the additional energy consumption efficiency coefficient, which ranges from 0 to 10%; represents the actual energy consumption; represents the planned energy consumption. Among them, the additional energy consumption efficiency coefficient s takes a positive number, and the reduced energy consumption efficiency coefficient to reduce the cost is not considered.

[0027] The expression of the current additional production operation cost rate formula for smelting metal is: ; In the above formula, m represents the additional production operation cost rate, which ranges from 0 to 10%; represents the additional production operation cost; represents the basic production operation cost. Among them, the additional production operation cost rate m takes a positive number, and the reduced production operation cost rate to reduce the cost is not considered.

[0028] The expression of the actual calculated marginal cost increase rate in the current smelting metal follow-up statistics is: ; In the above formula, g represents the actual marginal cost increase rate; represents the total cost of the current production batch; represents the reference total cost; represents the additional output of the current batch compared to the reference.

[0029] By collecting multi-source data (including material, energy consumption, production operation data and actual marginal cost increase rate) in the smelting process and preprocessing (such as data cleaning, normalization), the static features (such as basic material loss coefficient) are extracted by using convolutional neural network and the dynamic features (such as real-time energy consumption efficiency change) are extracted by using recurrent neural network, so as to comprehensively capture the stable and time-varying factors affecting the cost; This hybrid feature extraction method uses the local pattern recognition ability of convolutional neural network and the time sequence dependence capture ability of recurrent neural network, enhances the richness and robustness of feature representation, provides more accurate data basis for subsequent cost prediction, and at the same time ensures the data quality through preprocessing step, reduces the interference of noise and outliers.

[0030] Embodiment 2

[0031] As shown in Figure 1 , a smelting cost prediction model of the same metal material and the same production line is constructed according to the historical feature data stored in the database, and an optimization algorithm is used to optimize the hyperparameters of the smelting cost prediction model; then, the specific steps of constructing the smelting cost prediction model are: S401, collecting historical feature data of the same metal (for example, copper) and the same production line smelting metal (for example, copper bar); S402, controlling the variables of the additional increased energy consumption efficiency coefficient s and the additional increased production operation cost rate m, collecting the additional increased material loss coefficient a and the marginal cost increase rate of the smelted metal, and marking the marginal cost increase rate of the smelted metal at this time as the marginal cost increase rate A; constructing a loss prediction function according to the dynamic feature relationship between the marginal cost increase rate A and the additional increased material loss coefficient; For example, 100 copper bar sample data are collected, the values of s and m in the 100 sample data are controlled as variables, but a is not controlled, and if the marginal cost increase rate g (marked data 1) actually calculated in the follow-up statistics of the copper bar is collected.

[0032] Then, a mathematical model is established between the predicted marginal cost increase rate and a (the red dots in the figure are the distribution of the 100 collected samples), then: (Formula 1); In the above Formula 1, , for controlling and the constant tends to approximate the marked data 1. And the data in the figure can determine , When, the constant tends to approximate the marked data 1 in Formula 1.

[0033] S403, control the variables of the additional material loss coefficient a and the additional energy consumption efficiency coefficient s, collect the marginal cost increase rate of the smelted metal and the additional production operation cost rate m at this time, and mark the marginal cost increase rate of the smelted metal at this time as the marginal cost increase rate B; and construct an operation cost prediction function according to the dynamic characteristic relationship between the marginal cost increase rate B and the additional production operation cost rate; For example, the data of 100 copper bar samples are collected, the values of a and s in the 100 sample data are controlled, but m is not controlled, and the actual calculated marginal cost increase rate g (marked data 2) is collected when the copper bar is subsequently counted.

[0034] Then, a mathematical model is established between the predicted marginal cost increase rate and m (the blue points in the figure are the distribution of the collected 100 samples), so: (Formula 2); In the above Formula 2, , , for controlling and the constant tends to approximate the marked data 2. And the data in the figure can determine , , When, the constant tends to approximate the marked data 2 in Formula 2.

[0035] S404, control the variables of the additional material loss coefficient a and the additional production operation cost rate m, collect the marginal cost increase rate of the smelted metal and the additional energy consumption efficiency coefficient s, and mark the marginal cost increase rate of the smelted metal at this time as the marginal cost increase rate C; and construct an energy consumption efficiency prediction function according to the dynamic characteristic relationship between the marginal cost increase rate C and the additional energy consumption efficiency coefficient; For example, the data of 100 copper bar samples are collected, the values of a and m in the 100 sample data are controlled, but s is not controlled, and the actual calculated marginal cost increase rate g (marked data 3) is collected when the copper bar is subsequently counted.

[0036] So, the marginal cost increase rate of prediction A mathematical model is established for the relationship between s (the red dots in the figure are the distribution of the 100 samples collected), so: (Formula 3); In the above formula 3, , For control The constant tends to approximate the marked data 3. And from the data in the figure, it can be determined that , When, The constant in formula 3 tends to approximate the marked data 3.

[0037] S405, based on the dynamic characteristic relationship between the loss prediction function, the operating cost prediction function and the energy consumption efficiency prediction function, the smelting cost prediction model of the same metal material and the same production line is constructed (1) Similarly, the data of 100 copper bar samples are collected, and only the additional energy consumption efficiency coefficient s is controlled to collect the actual marginal cost increase rate g (marked data 4). A mathematical model is established for the dynamic characteristic relationship between the actual marginal cost increase rate g (marked data 4) and a, m, and combined with the characteristic relationship of the above formula 1 and formula 2, so: G= (Formula 1) x (Formula 2); (2) Similarly, the data of 100 copper bar samples are collected, and only the additional energy consumption efficiency coefficient m is controlled to collect the actual marginal cost increase rate g (marked data 5). A mathematical model is established for the dynamic characteristic relationship between the actual marginal cost increase rate g (marked data 5) and a, s, and combined with the characteristic relationship of the above formula 1 and formula 3, so: G= (Formula 1) x (Formula 3); (3) Similarly, the data of 100 copper bar samples are collected, and only the additional energy consumption efficiency coefficient a is controlled to collect the actual marginal cost increase rate g (marked data 6). A mathematical model is established for the dynamic characteristic relationship between the actual marginal cost increase rate g (marked data 6) and m, s, and combined with the characteristic relationship of the above formula 2 and formula 3, so: G= (Formula 2) x (Formula 3); Then according to the above derivation, it can be known that the mathematical expression of the cost prediction model of copper and the same production line smelting copper bar is: ; In the above formula, G represents the predicted marginal cost increase rate when smelting copper bars of the same production line. Moreover, the optimization of the hyperparameters of the smelting cost prediction model is based on historical characteristic data, and the optimal combination is searched automatically in the preset hyperparameter search space by selecting an optimization algorithm, and the optimal hyperparameter combination is determined according to the model performance index. The specific steps are as follows: (1) According to the structural characteristics of the smelting cost prediction model, determine the set of hyperparameters to be optimized and their value ranges, including but not limited to learning rate, batch size, number of hidden layer nodes, convolution kernel size, network layer number, and dropout rate; (2) Use Bayesian optimization, particle swarm optimization, or genetic algorithm as the hyperparameter optimization method to establish the mapping relationship between the hyperparameter combination and the model prediction accuracy; (3) Randomly generate an initial hyperparameter combination in the hyperparameter search space, train and validate the smelting cost prediction model using historical characteristic data, and record the model performance index corresponding to each hyperparameter combination; (4) Based on the evaluated hyperparameter combinations and their performance results, generate new candidate hyperparameter combinations through the optimization algorithm, evaluate the performance of the candidate combinations using cross-validation, and update the mapping relationship between the hyperparameters and the performance; (5) When the maximum number of iterations is reached or the model performance meets the preset threshold, terminate the optimization process, and select the hyperparameter combination with the highest marginal cost increase rate prediction accuracy on the validation set as the final optimization result; (6) Configure the optimized hyperparameters to the smelting cost prediction model to complete the hyperparameter optimization process of the model.

[0038] By controlling variable method (such as fixing energy consumption efficiency coefficient and production operation cost rate, only changing material loss coefficient), the prediction functions (such as loss prediction function, energy consumption efficiency prediction function) between material loss, energy consumption efficiency and production operation cost and marginal cost increase rate are constructed respectively, and based on the dynamic characteristic relationship of these functions, a comprehensive smelting cost prediction model is integrated; using optimization algorithm (such as Bayesian optimization) to automatically search the optimal combination in the preset hyperparameter space, so that the model can more accurately fit the historical data, thereby improving the prediction accuracy and generalization ability, and at the same time adapting to the specific conditions of different metal materials and production lines, realizing personalized cost modeling.

[0039] Example 3

[0040] As shown in Figure 1 , based on the optimized smelting cost prediction model, the additional material loss coefficient, the additional energy consumption efficiency coefficient, and the additional production operation cost rate of the current smelting metal are used to predict the marginal cost increase rate of the current smelting metal; The marginal cost increase rate of the current smelting metal prediction is fed back to the actual marginal cost increase rate generated subsequently, and then the hyperparameters of the smelting cost prediction model are optimized again through an optimization algorithm. The hyperparameters of the smelting cost prediction model are re-optimized and updated through an optimization algorithm based on the error data between the predicted marginal cost increase rate and the actual marginal cost increase rate generated subsequently. The specific steps are as follows: (1) Compare the marginal cost increase rate of the current smelting metal prediction with the actual marginal cost increase rate generated in the subsequent production process, obtain the model prediction error data, and combine the error data with the corresponding production batch data to form a feedback data set; (2) Based on the feedback data set, calculate the key performance indicators of the smelting cost prediction model under the current production conditions, and compare them with the historical performance before model optimization to determine whether the model performance has significantly degraded; (3) When the model performance degradation exceeds the preset threshold or reaches a fixed production batch interval, automatically trigger the hyperparameter feedback optimization process; (4) Use the feedback data set as a new validation set, combine historical feature data, and use the same optimization algorithm as step S4 to redefine the search space around the original optimal hyperparameters and perform iterative optimization of the hyperparameters; (5) Update the new hyperparameters obtained after feedback optimization to the smelting cost prediction model to replace the original hyperparameter configuration, and complete the online adaptive optimization of the model; (6) Repeat the above feedback optimization steps to enable the smelting cost prediction model to continuously adapt to changes in production conditions and maintain prediction accuracy.

[0041] Based on the optimized model, real-time cost prediction is performed, and by comparing the error data of the predicted marginal cost increase rate and the actual value, the feedback optimization mechanism is triggered, and the optimization algorithm (such as genetic algorithm) is used to re-optimize and update the hyperparameters; this dynamic feedback optimization process enables the model to adapt to changes in production conditions (such as equipment aging or process adjustment), continuously adjusts the model parameters to maintain prediction accuracy, thereby realizing online learning and self-evolution of the model, reducing prediction bias caused by environmental changes, and improving the long-term stability and practicality of the model.

[0042] Example 4

[0043] The smelting cost prediction system based on data analysis includes a data acquisition module, a data preprocessing module, a feature extraction module, a model construction and optimization module, a cost prediction module, and a feedback optimization module.

[0044] The data acquisition module is configured to acquire multi-source data of the same metal material and the same production line in the smelting process. The data preprocessing module is configured to perform data cleaning, missing value filling, abnormal value detection and data normalization processing on the multi-source data. The feature extraction module is configured to extract static features and dynamic features related to the smelting cost based on the preprocessed data, including a static feature extraction unit configured to extract static features by using a convolutional neural network and a dynamic feature extraction unit configured to extract dynamic features by using a recurrent neural network. The model construction and optimization module is configured to construct a smelting cost prediction model according to historical feature data, and optimize hyperparameters of the model by using an optimization algorithm. The cost prediction module is configured to predict a marginal cost increase rate of the current smelted metal based on the optimized smelting cost prediction model, according to an additional material loss coefficient, an energy consumption efficiency coefficient and a production operation cost rate of the current smelted metal. The feedback optimization module is configured to compare the predicted marginal cost increase rate with an actual marginal cost increase rate, and perform feedback optimization on the hyperparameters of the smelting cost prediction model by using an optimization algorithm. The entire cost prediction method is systematized, and the data acquisition, preprocessing, feature extraction, model construction and optimization, cost prediction and feedback optimization modules are integrated into one, forming a complete smelting cost prediction system based on data analysis. The system realizes an automatic process from data acquisition to prediction output through modular design, improves the efficiency and operability of cost prediction, and ensures real-time and accuracy through the collaborative work of each module, facilitating deployment and maintenance in an industrial production environment, supporting continuous optimization and expansion, and ultimately helping enterprises to realize fine cost management.

[0045] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the present patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present patent should be subject to the appended claims.

Claims

1. A data analysis-based method for predicting smelting costs, characterized in that, Includes the following steps: S1. Collect multi-source data on the same metal materials and metals smelted on the same production line during the smelting process. This multi-source data includes material data, energy consumption data, production operation data, and actual marginal cost growth rate. S2. Preprocess the multi-source data, including data cleaning, missing value imputation, outlier detection and data normalization; S3. Based on the preprocessed data, extract static and dynamic features related to smelting costs. Use convolutional neural networks to extract static features and recurrent neural networks to extract dynamic features. S4. Construct a smelting cost prediction model for the same metal material and the same production line based on the historical feature data stored in the database, and optimize the hyperparameters of the smelting cost prediction model using an optimization algorithm. S5. Based on the optimized smelting cost prediction model, predict the marginal cost growth rate of the current smelting metal according to the additional material loss coefficient, additional energy efficiency coefficient, and additional production operation cost rate of the current smelting metal. S6. Feedback the current marginal cost growth rate of smelting metal prediction and the subsequent actual marginal cost growth rate, and then use the optimization algorithm to optimize the hyperparameters of the smelting cost prediction model again.

2. The smelting cost prediction method based on data analysis according to claim 1, characterized in that: The static features refer to features that are relatively stable or do not change over time during the smelting process, including the basic material loss coefficient extracted based on material data, the basic energy efficiency coefficient extracted based on energy consumption data, the basic production operation cost rate extracted based on production operation data, and the inherent correlation between these coefficients and the actual marginal cost growth rate. The dynamic features refer to the characteristics that change over time during the smelting process, including real-time changes in material loss coefficient, real-time changes in energy efficiency coefficient, and real-time changes in production operation cost rate extracted from time series based on material data, energy consumption data, and production operation data, as well as the dynamic impact sequence of these changes on the actual marginal cost growth rate.

3. The smelting cost prediction method based on data analysis according to claim 1, characterized in that: In step S3, the specific steps for extracting static features using a convolutional neural network are as follows: (1) Align and splice the fields related to static attributes in the preprocessed material data, energy consumption data and production operation data to form a multidimensional static data matrix, and reconstruct it into a two-dimensional or three-dimensional tensor format that meets the input requirements of convolutional neural networks. (2) Input the constructed input tensor into the convolutional neural network. The network contains multiple convolutional layers. Each convolutional layer uses convolutional kernels of different sizes to perform multi-channel convolution operations in order to capture local static feature patterns at different scales. (3) Apply activation functions to the feature maps output by the convolutional layer to introduce nonlinear transformations; Subsequently, a pooling layer is used to downsample the feature map, preserving salient features while reducing data dimensionality and enhancing the translation invariance of features; (4) The feature maps after multi-layer convolution and pooling are spliced ​​and fused, and a high-level static feature representation is further abstracted through a fully connected layer. This high-level static feature representation is then flattened into a one-dimensional static feature vector. (5) The final one-dimensional static feature vector is used as the extracted static feature output for the construction of the subsequent smelting cost prediction model.

4. The smelting cost prediction method based on data analysis according to claim 1, characterized in that: In step S3, the specific steps for extracting dynamic features using a recurrent neural network are as follows: (1) From the preprocessed multi-source data, extract time-series variables related to dynamic attributes from material data, energy consumption data and production operation data in time order, construct multiple parallel time series data streams, and organize them into a three-dimensional tensor format suitable for recurrent neural network processing, whose dimensions represent the number of samples, time step and feature dimension respectively. (2) Input the constructed time series tensor into the recurrent neural network. The network adopts a long short-term memory network or a gated recurrent unit structure, processes the input sequence step by step, and captures the long-term dependencies and time-series dynamic patterns in the data through its internal gating mechanism. (3) At each time step, the recurrent neural network calculates the current hidden state based on the current input and the hidden state of the previous time step, thereby encoding the context information of the entire time series into the final hidden state sequence; (4) Aggregate the hidden states of all time steps of the recurrent neural network output, and use the attention mechanism or the hidden state of the last time step as a summary representation of the entire sequence to generate a comprehensive dynamic feature vector containing temporal dynamic information. (5) The final generated dynamic feature vector is used as the extracted dynamic feature output, and together with the static feature vector, it is used to construct the subsequent smelting cost prediction model.

5. The smelting cost prediction method based on data analysis according to claim 1, characterized in that: In step S4, the specific steps for constructing the smelting cost prediction model are as follows: S401. Collect historical characteristic data of the same metal and the metal smelted on the same production line; S402. Control the variables of the additional energy efficiency coefficient and the additional production operation cost rate, collect the additional material loss coefficient and the marginal cost increase rate of the smelted metal, and mark the marginal cost increase rate of the smelted metal at this time as the marginal cost increase rate A; construct a loss prediction function based on the dynamic characteristic relationship between the marginal cost increase rate A and the additional material loss coefficient. S403. Control the variables of additional material loss coefficient and additional energy efficiency coefficient, collect the additional production operation cost rate and the marginal cost growth rate of the smelted metal, and mark the marginal cost growth rate of the smelted metal at this time as the marginal cost growth rate B; construct an operation cost prediction function based on the dynamic characteristic relationship between the marginal cost growth rate B and the additional production operation cost rate. S404. Control the variables of additional material loss coefficient and additional production operation cost rate, collect the additional energy consumption efficiency coefficient and the marginal cost growth rate of the smelted metal, and mark the marginal cost growth rate of the smelted metal at this time as the marginal cost growth rate C; construct an energy consumption efficiency prediction function based on the dynamic characteristic relationship between the marginal cost growth rate C and the additional energy consumption efficiency coefficient. S405. Based on the dynamic characteristic relationship between the loss prediction function, the operating cost prediction function, and the energy efficiency prediction function, a smelting cost prediction model for the same metal material and the same production line is constructed.

6. The smelting cost prediction method based on data analysis according to claim 5, characterized in that: The smelting cost prediction model is a prediction model that is constructed by integrating the dynamic characteristic relationships between the loss prediction function, the operating cost prediction function, and the energy efficiency prediction function. It is used to comprehensively reflect the marginal cost increase rate of the metal smelted on the same metal material and the same production line due to the additional material loss coefficient, the additional energy efficiency coefficient, and the additional production operating cost rate.

7. The smelting cost prediction method based on data analysis according to claim 1, characterized in that: In step S4, the optimization of the hyperparameters of the smelting cost prediction model using an optimization algorithm is based on historical feature data. The optimization algorithm is selected to automatically search within the preset hyperparameter search space, and the optimal hyperparameter combination is determined according to the model performance index.

8. The smelting cost prediction method based on data analysis according to claim 1, characterized in that: In step S6, the hyperparameters of the smelting cost prediction model are optimized again by the optimization algorithm based on the error data between the predicted marginal cost growth rate and the subsequent actual marginal cost growth rate. The hyperparameters of the smelting cost prediction model are then re-optimized and updated by the optimization algorithm.

9. A smelting cost prediction system based on data analysis, characterized in that, The data analysis-based smelting cost prediction method according to any one of claims 1-8, wherein the data analysis-based smelting cost prediction system comprises: a data acquisition module, a data preprocessing module, a feature extraction module, a model building and optimization module, a cost prediction module, and a feedback optimization module.

Images