Performance prediction method and system based on aluminum product element proportioning and process parameters

Abstract

The application belongs to the technical field of data processing, and particularly relates to a performance prediction method and system based on element proportioning and process parameters of aluminum products. The method obtains a sequence of historical element proportioning and process parameters of a target aluminum product and pre-processes the sequence, combines a graph algorithm to construct a process proportioning node network, generates a proportioning-performance positive prediction function and a performance-proportioning reverse reasoning function by using multivariate regression and factor analysis algorithms, obtains an initial parameter sequence from the parameter disassembly matching model and the node network according to real-time performance requirements, obtains a predicted performance and a secondary parameter sequence through forward and reverse function operations, adjusts the parameter and the performance deviation by means of the reinforcement learning and the simulation algorithm through forward and reverse cycles until the deviation meets a threshold value, and outputs an optimal element proportioning and process parameter sequence. The application realizes accurate prediction of the performance of the aluminum product and intelligent optimization of the process parameters, and improves production efficiency and product quality.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to a method and system for performance prediction based on the element ratio and process parameters of aluminum products. Background Technology

[0002] Aluminum products are widely used in aerospace, automotive and other fields due to their low density and high strength. Their performance is closely related to the element ratio sequence and process parameter sequence. In the prior art, such as the patent application with publication number CN104899412A, a method for predicting the mechanical properties of aluminum alloy castings is disclosed. This method establishes the relationship between mechanical properties and porosity and secondary dendrite spacing. Then, it constructs a mathematical model of the formation of micropores and secondary dendrites during the solidification process of the casting. Finally, before casting, the values ​​of micropores and secondary dendrites are predicted by computer, and the mechanical properties of aluminum alloy castings are predicted based on the relationship between porosity, secondary dendrite spacing and mechanical properties. When the predicted mechanical properties fail to meet the standards, the casting process is improved to make the mechanical properties meet the design requirements, thereby achieving the purpose of non-destructive testing. Patent application CN120354727A discloses a method and system for predicting the performance of anode aluminum foil based on a stacked model. The method obtains the target parameters in the production process of anode aluminum foil, inputs the target parameters in the production process of the anode aluminum foil to be tested into the anode aluminum foil performance prediction model, and obtains the performance prediction results of the anode aluminum foil to be tested. However, existing technologies mostly focus on single performance prediction or only through limited parameter optimization, lacking a systematic integration of the element ratio sequence and process parameter sequence of aluminum products. It is difficult to achieve accurate performance prediction under the synergistic effect of multiple parameters, especially in terms of dynamic performance prediction under complex working conditions and full-process parameter optimization, which cannot meet the demand for high performance and high reliability of aluminum products in industrial production. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a performance prediction method and system based on the elemental ratios and process parameters of aluminum products. This method obtains and preprocesses historical elemental ratios and process parameters of the target aluminum product, constructs a process ratio node network using graph algorithms, and generates a ratio-performance forward prediction function and a performance-ratio backward inference function using multiple regression and factor analysis algorithms. Based on real-time performance requirements, an initial parameter sequence is obtained through parameter decomposition and matching models and the node network. Predicted performance and a secondary parameter sequence are obtained through forward and backward function operations. Then, based on the parameter and performance deviation, reinforcement learning and simulation algorithms are used for cyclical adjustment until the deviation meets a threshold, outputting the optimal elemental ratio and process parameter sequence. This application achieves accurate prediction of aluminum product performance and intelligent optimization of process parameters, improving production efficiency and product quality.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] Performance prediction methods based on aluminum product element ratios and process parameters include:

[0006] Obtain the historical element ratio sequence, process parameter sequence, process production control parameter sequence, and corresponding performance evaluation index parameter sequence of the target product, and combine them with graph algorithms to obtain the process ratio node network of the target product.

[0007] Based on the combination of element ratio sequence, process parameter sequence, process production control parameter sequence and performance evaluation index parameter sequence, multiple regression algorithm and factor analysis algorithm are used to obtain the ratio-performance positive prediction function and the performance-ratio inverse inference function.

[0008] To obtain the real-time performance requirements of the target product, the parameter decomposition and matching model with the built-in performance-ratio inverse inference function is combined with the target product process ratio node network to obtain the matched initial element ratio sequence and process parameter sequence. The corresponding predicted performance and the corresponding inverse inference quadratic element ratio sequence and process parameter sequence are obtained through the performance prediction model with the built-in ratio-performance forward prediction function and the performance-ratio inverse inference function.

[0009] Based on the parameter deviations corresponding to the initial element ratio sequence and process parameter sequence, and the secondary element ratio sequence and process parameter sequence, as well as the parameter deviations between real-time performance requirements and predicted performance, reinforcement learning and simulation algorithms are combined to perform forward and reverse cyclic adjustments until the parameter deviations and performance prediction deviations simultaneously meet the corresponding preset thresholds, and the element ratio and process parameter sequences that meet the threshold conditions are output.

[0010] Specifically, the construction process of the target product process ratio node network includes:

[0011] Denoising and label alignment preprocessing for the same target product are performed based on the target product's historical element ratio sequence, process parameter sequence, process production control parameter sequence, and corresponding performance evaluation index parameter sequence.

[0012] Based on the preprocessed historical element ratio sequence, process parameter sequence, process production control parameter sequence, and corresponding performance evaluation index parameter sequence, the factor analysis algorithm is used to obtain the element ratio and process parameter influence factors and influence weights corresponding to each process production control parameter, and the process production control parameter influence factors and corresponding influence weights corresponding to each performance evaluation index parameter.

[0013] Specifically, the construction process of the target product process ratio node network also includes:

[0014] Based on the element ratios and process parameter influence factors and influence weights corresponding to each process production control parameter, and the process production control parameter influence factors and influence weights corresponding to each performance evaluation index parameter, combined with the principal component algorithm and the preset contribution threshold, the main contribution factor and auxiliary contribution factor and the corresponding contribution coefficients corresponding to each performance evaluation index parameter are obtained.

[0015] The primary contribution factors include a first primary contribution factor and a second primary contribution factor; the first primary contribution factor is a process production control parameter for each target product whose contribution coefficient is greater than a contribution threshold; the second primary contribution factor is an element ratio or process parameter for each target product whose contribution coefficient is greater than a contribution threshold; the auxiliary contribution factors include a first auxiliary contribution factor and a second auxiliary contribution factor; the first auxiliary contribution factor is a process production control parameter for each target product whose contribution coefficient is less than or equal to a contribution threshold; the second auxiliary contribution factor is an element ratio or process parameter for each target product whose contribution coefficient is less than or equal to a contribution threshold.

[0016] Specifically, the construction process of the target product process ratio node network also includes:

[0017] Based on the control variable method and correlation algorithm, the process production control parameter sequence and the corresponding performance evaluation index parameter sequence are analyzed to obtain the first influence change rate of the corresponding performance evaluation index parameter when any first main contribution factor or first auxiliary contribution factor changes by a preset unit amount within the same time period.

[0018] Similarly, based on the control variable method and correlation algorithm, the element ratio sequence, process parameter sequence and corresponding process production control parameter sequence are analyzed to obtain the second influence change rate of the first main contribution factor or the first auxiliary contribution factor when the second main contribution factor changes by a preset unit amount, and the third influence change rate of the first main contribution factor or the first auxiliary contribution factor when the second auxiliary contribution factor changes by a preset unit amount.

[0019] The target product is set as a first-level node, the performance evaluation index parameters are set as second-level nodes, the first primary contribution factor and the first secondary contribution factor are set as the first primary third-level node and the first secondary third-level node, respectively, and the second primary contribution factor and the second secondary contribution factor are set as the second primary fourth-level node and the second secondary fourth-level node.

[0020] Specifically, the construction process of the target product process ratio node network also includes:

[0021] Based on the relationship between the performance evaluation index parameters corresponding to the target product, a primary index connection is constructed. Based on the first primary contribution factor corresponding to each performance evaluation index parameter, a secondary primary contribution connection sequence is constructed. The contribution coefficient of the first secondary contribution factor and the first influence change rate are used to construct a secondary secondary contribution sequence.

[0022] Based on the contribution coefficient of any second primary contribution factor to any first primary contribution factor or first secondary contribution factor and the rate of change of the second influence, a three-level primary contribution connection is constructed; at the same time, based on the contribution coefficient of any second secondary contribution factor to any first primary contribution factor or first secondary contribution factor and the rate of change of the third influence, a three-level secondary contribution connection is constructed.

[0023] Based on primary nodes, secondary nodes, first primary tertiary nodes and first secondary tertiary nodes, second primary quaternary nodes and second secondary quaternary nodes, primary indicator connections, secondary primary contribution connection sequences and secondary secondary contribution sequences, and tertiary primary contribution connections and tertiary secondary contribution connections, a target product process ratio node network is constructed using a graph algorithm.

[0024] Specifically, the construction process of the target product process ratio node network also includes:

[0025] Based on the target product process ratio node network, for each target product, a product process chain is constructed using blockchain algorithms, consisting of first-level nodes, second-level nodes, first primary third-level nodes and first secondary third-level nodes, second primary fourth-level nodes and second secondary fourth-level nodes, and their corresponding connections.

[0026] The process control parameters, element ratios, and process parameters stored in the first primary tertiary node, the first secondary tertiary node, the second primary quaternary node, and the second secondary quaternary node in each product process chain are analyzed. The process control parameters, element ratios, and process parameters stored in the first primary tertiary node, the first secondary tertiary node, the second primary quaternary node, and the second secondary quaternary node are sorted according to their time priority to obtain the product process classification parameter chain.

[0027] Specifically, the construction process of the ratio-performance forward prediction function and the performance-ratio backward inference function includes:

[0028] Based on the contribution coefficients between each performance evaluation index parameter in the product process grading parameter chain corresponding to each target product and the corresponding first main contributing factor or first auxiliary contributing factor, the contribution coefficients between the second main contributing factor and the corresponding first main contributing factor or first auxiliary contributing factor, and the contribution coefficients between the second auxiliary contributing factor and the corresponding first main contributing factor or first auxiliary contributing factor, the ratio-performance positive prediction function corresponding to each performance evaluation index parameter of the target product is obtained through a multiple regression algorithm.

[0029] Based on the first influence change rate of each performance evaluation index parameter in the product process classification parameter chain with the corresponding first main contributing factor or first auxiliary contributing factor, the second influence change rate of each second main contributing factor with the corresponding first main contributing factor or first auxiliary contributing factor, and the third influence change rate of each second auxiliary contributing factor with the corresponding first main contributing factor or first auxiliary contributing factor within the same time period, an initial performance-ratio inverse inference function is constructed by combining a multiple regression algorithm.

[0030] The first influence change rate, the second influence change rate, the third influence change rate, the first main contribution factor and the first auxiliary contribution factor and the second main contribution factor and the second auxiliary contribution factor corresponding to the ratio-performance positive prediction function, as well as the corresponding weight parameters and the weight sequence corresponding to the initial performance-ratio reverse inference function, are used as the input sequence of the genetic algorithm.

[0031] Specifically, the construction process of the ratio-performance forward prediction function and the performance-ratio backward inference function also includes:

[0032] The ratio-performance forward prediction function is used as the forward fitness function, and the performance-ratio backward inference function is used as the backward fitness function. The parameter deviation of the performance corresponding to the forward fitness function and the backward fitness function is used as the target fitness function. At the same time, based on the constraints in the process production control parameter sequence, the constraints of the process parameters, and the corresponding correlation constraints between the process parameters and the process production control parameters, the constraint condition sequence in the genetic algorithm is constructed.

[0033] Based on the input sequence, forward fitness function, reverse fitness function, target fitness function and constraint condition sequence of genetic algorithm, the dynamic multi-objective evolutionary algorithm optimized by genetic algorithm is combined with simulation algorithm and preset training period to simulate training and solve, and obtain performance deviation, forward and reverse ratio and process parameter value deviation.

[0034] If the deviation between the input ratio and process parameters corresponding to the positive fitness function and the ratio and process parameters predicted by the negative fitness function does not meet the corresponding preset deviation threshold, and the deviation between the predicted performance obtained by the positive fitness function and the input performance corresponding to the negative fitness function also does not meet the corresponding preset deviation threshold, the first execution action information is triggered.

[0035] Based on the parameters and performance deviations in the first execution action information, a reinforcement learning algorithm combined with a deep search algorithm and a target product process ratio node network is used to perform secondary ratio, process parameter search, and corresponding performance search. The searched secondary ratio, process parameter search, and corresponding performance search are used to adjust the first influence change rate, second influence change rate, third influence change rate, first main contribution factor, first auxiliary contribution factor, second main contribution factor, and second auxiliary contribution factor, as well as the corresponding weight parameters. Forward and reverse training is then performed again to solve and adjust until the element ratio, process parameter deviation, and performance deviation simultaneously meet the corresponding thresholds. The trained forward fitness function, reverse fitness function, and corresponding element ratio, process parameter, and performance evaluation index parameters are obtained.

[0036] A performance prediction system based on aluminum product element ratios and process parameters includes: a data processing module and a function construction module;

[0037] The data processing module is used to obtain the historical element ratio sequence, process parameter sequence, process production control parameter sequence and corresponding performance evaluation index parameter sequence of the target product, and combine them with graph algorithm to obtain the process ratio node network of the target product.

[0038] The function construction module, based on the element ratio sequence, process parameter sequence, process production control parameter sequence, and performance evaluation index parameter sequence, combined with multiple regression algorithm and factor analysis algorithm, obtains the ratio-performance positive prediction function and the performance-ratio inverse inference function.

[0039] Specifically, the system also includes a matching prediction module and a reinforcement adjustment module;

[0040] The matching prediction module is used to obtain the real-time performance requirements of the target product. By combining the parameter decomposition matching model with the target product process ratio node network with the built-in performance-ratio inverse inference function, the initial element ratio sequence and process parameter sequence of the matching are obtained. By the performance prediction model with the built-in ratio-performance forward prediction function and the performance-ratio inverse inference function, the corresponding predicted performance and the corresponding inverse inference quadratic element ratio sequence and process parameter sequence are obtained.

[0041] The enhancement adjustment module, based on the parameter deviations corresponding to the initial element ratio sequence, the process parameter sequence, and the secondary element ratio sequence and the process parameter sequence, as well as the parameter deviations between the real-time performance requirements and the predicted performance, combines reinforcement learning and simulation algorithms to perform forward and reverse cyclic approximation adjustments until the parameter deviations and performance prediction deviations simultaneously meet the corresponding preset thresholds, and outputs the element ratio and process parameter sequences that meet the threshold conditions.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] This invention addresses the shortcomings of existing technologies by constructing a target product process ratio node network. It integrates historical element ratios and process parameter sequences to build a multi-level process ratio node network and hierarchical parameter chain. Factor analysis quantifies the influence weights of primary and secondary contributing factors and the rate of change of cross-level parameter influence. Combined with a genetic algorithm, it collaboratively optimizes the ratio-performance forward prediction function and the performance-ratio backward inference function, forming a bidirectional closed-loop training mechanism. When the forward and backward deviations exceed a threshold, reinforcement learning dynamically adjusts the node network parameter weights and retrains the system, ultimately achieving dual convergence of process parameters and performance requirements. This application achieves accurate prediction of aluminum product performance and intelligent optimization of process parameters, avoiding the blindness of traditional trial-and-error methods, improving production efficiency and product quality, and enhancing the adaptability and reliability of process adjustments. Attached Figure Description

[0044] Figure 1 This is a flowchart of the performance prediction method based on aluminum product element ratio and process parameters of the present invention;

[0045] Figure 2 This is a block diagram of the performance prediction system based on the element ratio and process parameters of aluminum products according to the present invention. Detailed Implementation

[0046] Example 1

[0047] Please see Figure 1 The present invention provides an embodiment of a performance prediction method based on the elemental ratio and process parameters of aluminum products, comprising the following steps:

[0048] S1. Obtain the historical element ratio sequence, process parameter sequence, process production control parameter sequence, and corresponding performance evaluation index parameter sequence of the target product, and combine them with graph algorithm to obtain the process ratio node network of the target product; further, the target product in this embodiment is an aluminum product;

[0049] S2. Based on the element ratio sequence, process parameter sequence, process production control parameter sequence and performance evaluation index parameter sequence, combined with multiple regression algorithm and factor analysis algorithm, obtain the ratio-performance positive prediction function and the performance-ratio inverse inference function.

[0050] S3. Obtain the real-time performance requirements of the target product. By combining the parameter decomposition and matching model with the target product's process ratio node network, a matching initial element ratio sequence and process parameter sequence are obtained. Then, by combining the performance prediction model with the performance ratio-performance forward prediction function and the performance-ratio backward inference function, the corresponding predicted performance and the corresponding backward inference quadratic element ratio sequence and process parameter sequence are obtained. The parameter decomposition and matching model is preferably constructed by combining a pre-trained BERT model with a matching algorithm.

[0051] S4. Based on the parameter deviations corresponding to the initial element ratio sequence and process parameter sequence, and the secondary element ratio sequence and process parameter sequence, as well as the parameter deviations between real-time performance requirements and predicted performance, reinforcement learning and simulation algorithms are combined to perform forward and reverse cyclic adjustments until the parameter deviations and performance prediction deviations simultaneously meet the corresponding preset thresholds, and the element ratio and process parameter sequences that meet the threshold conditions are output.

[0052] It should be further explained that the construction process of the target product process ratio node network in this embodiment includes:

[0053] Denoising and label alignment preprocessing for the same target product are performed based on the target product's historical element ratio sequence, process parameter sequence, process production control parameter sequence, and corresponding performance evaluation index parameter sequence.

[0054] It should be further noted that the elemental ratio sequence in this embodiment includes, but is not limited to, the main elements in the aluminum product and the mole fraction or mass fraction of the refining agent; the process parameter sequence includes, but is not limited to, temperature parameters such as casting temperature, mold temperature, solution temperature, and aging temperature, as well as time parameters such as solution time, aging time, and cooling rate; the process production control parameter sequence involves pressure parameters, equipment operating parameters, and environmental parameters; the performance evaluation index parameter sequence includes mechanical property parameters, physical property parameters, and process performance parameters; the main elements include, but are not limited to, Al, Cu, and Mg; the mechanical property parameters include, but are not limited to, tensile strength, yield strength, and elongation; the physical property parameters include, but are not limited to, thermal conductivity and electrical conductivity; and the process performance parameters include, but are not limited to, casting fluidity and hot cracking tendency.

[0055] Based on the preprocessed historical element ratio sequence, process parameter sequence, process production control parameter sequence, and corresponding performance evaluation index parameter sequence, the factor analysis algorithm is used to obtain the element ratio and process parameter influence factors and influence weights corresponding to each process production control parameter, and the process production control parameter influence factors and corresponding influence weights corresponding to each performance evaluation index parameter.

[0056] Based on the element ratios and process parameter influence factors and influence weights corresponding to each process production control parameter, and the process production control parameter influence factors and influence weights corresponding to each performance evaluation index parameter, combined with the principal component algorithm and the preset contribution threshold, the main contribution factor and auxiliary contribution factor and the corresponding contribution coefficients corresponding to each performance evaluation index parameter are obtained.

[0057] The primary contribution factors include a first primary contribution factor and a second primary contribution factor; the first primary contribution factor is a process control parameter for each target product whose contribution coefficient is greater than a contribution threshold; the second primary contribution factor is an element ratio or process parameter for each target product whose contribution coefficient is greater than a contribution threshold; the secondary contribution factors include a first secondary contribution factor and a second secondary contribution factor; the first secondary contribution factor is a process control parameter for each target product whose contribution coefficient is less than or equal to a contribution threshold; the second secondary contribution factor is an element ratio or process parameter for each target product whose contribution coefficient is less than or equal to a contribution threshold.

[0058] It should be further explained that, in this embodiment, one specific method for obtaining the main contribution factor, secondary contribution factor, and corresponding contribution coefficient for each performance evaluation index parameter is as follows:

[0059] First, using a matrix dimension alignment algorithm, the matrix of factors influencing the performance evaluation indicators of process control parameters and the matrix of factors influencing the process control parameters of element ratios are transposed and feature-mapped. A tensor fusion algorithm is then used to construct a third-order multidimensional influence factor correlation matrix. Based on this matrix, a covariance matrix is ​​generated between parameters using a covariance matrix calculation module. Then, eigenvalues ​​and eigenvectors are solved using an eigenvalue decomposition algorithm. The number of principal components is determined by combining the cumulative variance contribution rate threshold, and then the standardized loading matrix of each influence factor on the principal components is extracted. Based on the principal component loading matrix and the variance contribution rate of each principal component, a weighted aggregation algorithm (using the variance contribution rate as the weighting coefficient) is used to perform multi-principal component analysis on each influence factor. The loads are weighted and summed to generate a comprehensive contribution coefficient vector. Based on a preset contribution threshold, a two-way conditional filtering algorithm is used to classify the process control parameters and element ratios or process parameters: process control parameters with contribution coefficients greater than the threshold are marked as the first main contribution factor using a Boolean index algorithm, and element ratios or process parameters with contribution coefficients greater than the threshold are marked as the second main contribution factor. At the same time, process control parameters and element ratios or process parameters with contribution coefficients less than or equal to the threshold are classified as the first auxiliary contribution factor and the second auxiliary contribution factor, respectively. Finally, a matrix concatenation algorithm is used to generate a four-dimensional classification result matrix containing performance evaluation indicators, main contribution factors, auxiliary contribution factors, and corresponding contribution coefficients.

[0060] Based on the control variable method and correlation algorithm, the process production control parameter sequence and the corresponding performance evaluation index parameter sequence are analyzed to obtain the first influence change rate of the corresponding performance evaluation index parameter when any first main contribution factor or first auxiliary contribution factor changes by a preset unit amount within the same time period.

[0061] Similarly, based on the control variable method and correlation algorithm, the element ratio sequence, process parameter sequence and corresponding process production control parameter sequence are analyzed to obtain the second influence change rate of the first main contribution factor or the first auxiliary contribution factor when the second main contribution factor changes by a preset unit amount, and the third influence change rate of the first main contribution factor or the first auxiliary contribution factor when the second auxiliary contribution factor changes by a preset unit amount.

[0062] It should be further explained that the specific process for obtaining the first, second, and third influence change rates in this embodiment includes:

[0063] First, the data is divided into subsequences with a 50% overlap using a sliding window time segmentation algorithm (the window length is an integer multiple of the historical data period average). Based on the control variable method, other parameter values ​​are predicted and kept constant using a Kalman filter. Piecewise linear interpolation adjustment with a preset unit amount is performed on the target first main contribution factor or first secondary contribution factor. The parameter-performance curve is fitted using a local weighted scatter smoothing algorithm (bandwidth parameter determined by cross-validation). The gradient value of the curve at each data point is calculated using cubic spline interpolation. After smoothing with a Savitzky-Golay filter, the first influence change rate is obtained. Similarly, based on the element ratio sequence, process parameter sequence, and corresponding process production control parameter sequence, a dynamic time warping algorithm is used to perform multi-factor optimization. The parameters are aligned along the time axis. During the control variable process, a Hidden Markov Model (HMM) is applied to predict the states of other parameters while keeping them constant. A pre-set unit impulse response test is performed on the target second primary contributing factor. The direct effect coefficients between parameters are calculated using a partial least squares path modeling algorithm. The confidence intervals of the coefficients are estimated using Monte Carlo simulation (1000 sampling times) to obtain the second influence rate of change. For the second secondary contributing factor, a time series preprocessing is performed using a differential integrated moving average autoregressive model. A vector error correction model is constructed to capture the long-term equilibrium relationship between parameters. The cumulative effect value of the change in the first primary or secondary contributing factor caused by a unit change is calculated based on the generalized impulse response function. After verifying stability through Bootstrap resampling, the third influence rate of change is obtained. The preferred dynamic time warping algorithm is the DTW-Barycenter Averaging method. The ARIMA order in the differential integrated moving average autoregressive model is determined by the AIC information criterion. The cumulative response period of the generalized impulse response function is set to 12 time steps.

[0064] The target product is set as a first-level node, the performance evaluation index parameters are set as second-level nodes, the first primary contribution factor and the first secondary contribution factor are set as the first primary third-level node and the first secondary third-level node, respectively, and the second primary contribution factor and the second secondary contribution factor are set as the second primary fourth-level node and the second secondary fourth-level node.

[0065] Based on the relationship between the performance evaluation index parameters corresponding to the target product, a primary index connection is constructed. Based on the first primary contribution factor corresponding to each performance evaluation index parameter, a secondary primary contribution connection sequence is constructed. The contribution coefficient of the first secondary contribution factor and the first influence change rate are used to construct a secondary secondary contribution sequence.

[0066] Based on the contribution coefficient of any second primary contribution factor to any first primary contribution factor or first secondary contribution factor and the rate of change of the second influence, a three-level primary contribution connection is constructed; at the same time, based on the contribution coefficient of any second secondary contribution factor to any first primary contribution factor or first secondary contribution factor and the rate of change of the third influence, a three-level secondary contribution connection is constructed.

[0067] Based on primary nodes, secondary nodes, first primary tertiary nodes and first secondary tertiary nodes, second primary quaternary nodes and second secondary quaternary nodes, primary indicator connections, secondary primary contribution connection sequences and secondary secondary contribution sequences, and tertiary primary contribution connections and tertiary secondary contribution connections, a target product process ratio node network is constructed using a graph algorithm.

[0068] It should be further explained that, in this embodiment, a specific implementation of constructing the target product process ratio node network using graph algorithms is as follows:

[0069] First, an adjacency matrix construction algorithm is used to convert the first-level indicator connections, second-level primary or secondary contribution sequences, and third-level primary or secondary contribution connections into weighted directed graph adjacency matrices, where the edge weights are the product of the corresponding contribution coefficients and the rate of change of influence. Based on the adjacency matrix, the PageRank algorithm is used to calculate the importance score of each node to identify key performance indicators and core contribution factors. The initial adjacency matrix is ​​then sparsified using the minimum spanning tree algorithm, retaining the connections with the highest weights to reduce graph complexity. Based on the sparsified adjacency matrix, a community detection algorithm is used to identify the functional modules and clustering structures among parameters, forming a process ratio sub-network. Based on the sub-network partitioning results, a hierarchical clustering algorithm is used to construct the hierarchical structure of nodes, grouping nodes with similar contributions... The node organization in the contribution pattern is a tree hierarchy; the influence path between any two nodes is calculated using the shortest path algorithm to identify the indirect interaction relationships between parameters; based on the path analysis results, the control capability and influence of each node in the network are evaluated using centrality analysis algorithms (such as betweenness centrality and proximity centrality); finally, the nodes and connection relationships are mapped to a low-dimensional vector space using a graph embedding algorithm to form a structured target product process ratio node network, where the node position reflects the importance of the parameters, and the vector distance represents the correlation strength between the parameters; the minimum spanning tree algorithm is preferably Kruskal's algorithm; the community detection algorithm is preferably Louvain's algorithm; the graph embedding algorithm is preferably DeepWalk; and the shortest path algorithm is preferably Dijkstra's algorithm.

[0070] Based on the target product process ratio node network, for each target product, a product process chain is constructed using blockchain algorithms, consisting of first-level nodes, second-level nodes, first primary third-level nodes and first secondary third-level nodes, second primary fourth-level nodes and second secondary fourth-level nodes, and their corresponding connections.

[0071] The process control parameters, element ratios, and process parameters stored in the first primary tertiary node, the first secondary tertiary node, the second primary quaternary node, and the second secondary quaternary node in each product process chain are analyzed. The process control parameters, element ratios, and process parameters stored in the first primary tertiary node, the first secondary tertiary node, the second primary quaternary node, and the second secondary quaternary node are sorted according to their time priority to obtain the product process classification parameter chain.

[0072] It should be further explained that one implementation method for constructing the product process hierarchical parameter chain in this embodiment is as follows:

[0073] Based on the nodes and connections at each level in the target product process ratio node network, a blockchain distributed ledger construction algorithm is first used to encode the parameter data and connection weights of the first-level to fourth-level nodes into a standardized block structure: the block header includes a version number, the hash value of the previous block, the Merkle root hash, and a timestamp; the block body stores metadata such as node attributes, contribution coefficients, and impact change rates. A consensus algorithm is used to verify the legality of the block data, and a hash pointer linking algorithm is used to connect the blocks in chronological order to construct an immutable product process chain. For the control parameters and ratio process parameters stored at each level of the process chain, a timestamp is generated by extracting the parameters using a timestamp parsing algorithm. Based on a primary and secondary factor priority determination algorithm, combined with a stable sorting algorithm (using merge sort to ensure that the primary and secondary order of parameters at the same time remains unchanged), the parameters are sorted according to the rule of primary and secondary level priority > timestamp order. Finally, a product process hierarchical parameter chain sorted by both primary and secondary level and time priority is obtained through a parameter hierarchical aggregation algorithm.

[0074] In this embodiment, the Merkle root hash is generated by hashing and aggregating node data using the Merkle tree algorithm; the consensus algorithm is preferably the Practical Byzantine Fault Tolerance (PBFT) algorithm, which is suitable for industrial scenarios; in this embodiment, the parameter hierarchical aggregation algorithm first groups nodes according to the third / fourth level, arranges them in ascending order of timestamp within the same level, and arranges parameters in descending order of contribution coefficient at the same time; based on the primary and secondary factor priority determination algorithm, the parameter priority of the first primary third-level node is set to be higher than that of the first secondary third-level node, and the priority of the second primary fourth-level node is higher than that of the second secondary fourth-level node.

[0075] This embodiment utilizes a target product process ratio node network and supporting parameter management system constructed through the integration of multi-dimensional technologies, enabling full-process technology empowerment from data preprocessing to parameter optimization. In the data preprocessing stage, denoising and label alignment algorithms enhance the reliability of multi-source data such as historical element ratios and process parameters, laying the foundation for subsequent analysis. By combining factor analysis and principal component analysis algorithms, a multi-dimensional influence factor correlation matrix is ​​constructed and principal component loadings are extracted, effectively identifying primary and secondary factors that significantly contribute to performance evaluation indicators, reducing data dimensionality while enhancing model interpretability. Based on the synergy of the control variable method with algorithms such as sliding window and Kalman filtering, the impact rate of unit parameter changes on performance indicators is accurately quantified in time-segmented data. Dynamic time warping and partial least squares path modeling techniques ensure accurate calculation of multi-parameter time axis alignment and indirect effects, while the introduction of vector error correction models and generalized impulse response functions captures the long-term equilibrium relationship and cumulative effect between parameters, improving the comprehensiveness of the influence analysis. Secondly, the application of graph algorithms constructs a hierarchical process matching node network: the adjacency matrix transforms parameter associations into a weighted directed graph, the PageRank algorithm identifies key nodes, the minimum spanning tree and community detection algorithms achieve network sparsity and functional module division, hierarchical clustering and shortest path algorithms construct the parameter hierarchy and reveal indirect influence paths, and finally, the graph embedding algorithm maps the high-dimensional network into a low-dimensional vector space, realizing the visualization and structured expression of parameter associations; the introduction of blockchain technology, through distributed ledger and PBFT consensus algorithm, ensures the immutability and traceability of parameter data in the process chain, and the Merkle tree hash aggregation and timestamp mechanism ensure data integrity and temporal correctness. The product process grading parameter chain achieves dual sorting of parameters by primary and secondary levels and time sequence through the priority determination of primary and secondary factors and a stable sorting algorithm. The parameter hierarchical aggregation strategy further optimizes the arrangement of parameters at the same time according to the contribution coefficient. In summary, this embodiment forms a process management system with data reliability, model interpretability and parameter traceability through the synergy of technologies such as data cleaning, factor extraction, influence quantification, graph network construction and blockchain notarization. It not only realizes the visual modeling of complex relationships between parameters, but also improves the controllability and traceability of the production process through hierarchical sorting and blockchain notarization.

[0076] It should be further explained that the construction process of the ratio-performance forward prediction function and the performance-ratio backward inference function in this embodiment includes:

[0077] Based on the contribution coefficients between each performance evaluation index parameter in the product process grading parameter chain corresponding to each target product and the corresponding first main contributing factor or first auxiliary contributing factor, the contribution coefficients between the second main contributing factor and the corresponding first main contributing factor or first auxiliary contributing factor, and the contribution coefficients between the second auxiliary contributing factor and the corresponding first main contributing factor or first auxiliary contributing factor, the ratio-performance positive prediction function corresponding to each performance evaluation index parameter of the target product is obtained through a multiple regression algorithm.

[0078] It should be further explained that one implementation method for constructing the ratio-performance positive prediction function corresponding to each performance evaluation index of the target product in this embodiment is as follows:

[0079] Based on the contribution coefficients between performance evaluation indicators and contribution factors at each level in the product process grading parameter chain corresponding to the target product, a third-order tensor (performance indicator, primary and secondary factors, contribution) is first constructed according to the hierarchical relationship between the contribution coefficients of the performance evaluation indicators and the first or secondary contribution factors and the second or secondary contribution factors using a multidimensional data tensor reconstruction algorithm. This tensor is then converted into a two-dimensional feature matrix using a tensor expansion algorithm. Based on this matrix, the data in each column is normalized using a quantile standardization algorithm to eliminate dimensional differences. Then, abnormal contribution coefficient samples are identified and removed using an adaptive sliding window correlation coefficient matrix calculation and a local outlier detection algorithm. Finally, a dynamic VIF threshold adjustment is performed. The algorithm (adaptively adjusting the threshold according to the feature dimension) combines principal component analysis dimensionality reduction technology to combine and reconstruct highly collinear contribution factors, generating an orthogonalized feature matrix. Based on the processed feature matrix, a regularized multiple linear regression framework is adopted, and the regression coefficients are iteratively optimized through stochastic gradient descent algorithm, while early stopping is used to avoid overfitting. The regularization parameters and model hyperparameters are jointly optimized through nested cross-validation algorithm. Based on the optimal regression coefficient vector and standardized parameters, the multiple regression equation is transformed into a proportion-performance positive prediction function that supports batch input through function vectorization encapsulation algorithm, realizing the nonlinear mapping modeling from the contribution coefficients of each level of contribution factors to performance indicators.

[0080] Based on the first influence change rate of each performance evaluation index parameter in the product process classification parameter chain with the corresponding first main contributing factor or first auxiliary contributing factor, the second influence change rate of each second main contributing factor with the corresponding first main contributing factor or first auxiliary contributing factor, and the third influence change rate of each second auxiliary contributing factor with the corresponding first main contributing factor or first auxiliary contributing factor within the same time period, an initial performance-ratio inverse inference function is constructed by combining a multiple regression algorithm.

[0081] The first influence change rate, the second influence change rate, the third influence change rate, the first main contribution factor and the first auxiliary contribution factor and the second main contribution factor and the second auxiliary contribution factor corresponding to the ratio-performance positive prediction function, as well as the corresponding weight parameters and the weight sequence corresponding to the initial performance-ratio reverse inference function, are used as the input sequence of the genetic algorithm.

[0082] It should be further explained that, in this embodiment, based on the performance evaluation indicators and contribution factor data of each level within the product process grading parameter chain during the same time period, the first, second, and third impact change rates are synchronized by timestamp using a time series alignment algorithm. The first or second main or secondary contribution factor and corresponding weight parameters are then extracted from the ratio-performance forward prediction function using an impact factor extraction algorithm. The weight sequence is then obtained from the initial performance-ratio reverse inference function using a weight sequence extraction algorithm. Based on the above multi-source data, the impact change rate, contribution factor, and weight parameter are converted into numerical vectors of the same dimension using a data type unification algorithm. The vectors are then standardized using a normalization algorithm. Finally, the processed impact change rate vector, contribution factor vector, weight parameter vector, and reverse weight sequence are concatenated into a multi-dimensional feature matrix using a feature concatenation algorithm in a preset order. Based on the genetic algorithm chromosome encoding rules, the multi-dimensional feature matrix is ​​converted into a fixed-length binary string using a binary encoding algorithm. An initial fitness value is associated with each encoding string using a fitness function mapping algorithm. Finally, a genetic algorithm input sequence containing the impact change rate, main and secondary contribution factors, weight parameters, and reverse weight sequence is obtained.

[0083] The ratio-performance forward prediction function is used as the forward fitness function, and the performance-ratio backward inference function is used as the backward fitness function. The parameter deviation of the performance corresponding to the forward fitness function and the backward fitness function is used as the target fitness function. At the same time, based on the constraints in the process production control parameter sequence, the constraints of the process parameters, and the corresponding correlation constraints between the process parameters and the process production control parameters, the constraint condition sequence in the genetic algorithm is constructed.

[0084] It should be noted that the process of constructing the constraint condition sequence in the genetic algorithm in this embodiment includes:

[0085] First, the improved Apriori algorithm is used to mine frequent co-occurrence patterns among parameters. After aligning the time axes of multiple parameters using a dynamic time warping algorithm, the time-delay correlation between parameters is obtained through cross-correlation coefficient matrix calculation, resulting in a dynamic association rule set between process parameters and production control parameters. Through a constraint formalization transformation engine, the association rules (e.g., casting temperature must be higher than mold temperature and the difference must be within a preset range) are transformed into inequality constraints with time windows. A basic constraint library is constructed by combining physical constraints of process parameters (e.g., solution treatment time) and equipment constraints of production control parameters (e.g., stirring speed ≤ maximum stirring speed). Based on the target product process proportion node network... The algorithm iterates through the causal dependency paths between parameters using a depth-first search algorithm (e.g., the setting of aging temperature needs to refer to historical values ​​of solution temperature) to generate a parameter hierarchical dependency constraint chain. The priority weight of each constraint is calculated using fuzzy hierarchical analysis (the weight of the main contribution factor constraint is higher than that of the auxiliary factor). Finally, the algorithm extracts the historical legal value range of parameters (the parameter range verified by multi-version consensus) from the blockchain product process chain through a smart contract. The constraint condition serialization engine encodes the dynamic association constraints, physical device constraints, hierarchical dependency constraints, and historical legal value range in descending order of priority into a constraint condition sequence that can be executed by the genetic algorithm, thereby realizing multi-dimensional constraints on the parameter search space.

[0086] Based on the input sequence, forward fitness function, reverse fitness function, target fitness function and constraint condition sequence of genetic algorithm, the dynamic multi-objective evolutionary algorithm optimized by genetic algorithm is combined with simulation algorithm and preset training period to simulate training and solve, and obtain performance deviation, forward and reverse ratio and process parameter value deviation.

[0087] If the deviation between the input ratio and process parameters corresponding to the positive fitness function and the ratio and process parameters predicted by the negative fitness function does not meet the corresponding preset deviation threshold, and the deviation between the predicted performance obtained by the positive fitness function and the input performance corresponding to the negative fitness function also does not meet the corresponding preset deviation threshold, the first execution action information is triggered.

[0088] It should be further explained that the specific process of simulation training and solving the problem by combining the dynamic multi-objective evolutionary algorithm optimized by the genetic algorithm with the simulation algorithm and the preset training period in this embodiment includes:

[0089] Based on the genetic algorithm input sequence, forward and reverse fitness functions, target fitness function, and constraint condition sequence, the input sequence is first converted into a chromosome population using an adaptive binary encoding algorithm (the encoding length is dynamically adjusted according to the parameter dimension). Parent individuals are selected from the current population using a tournament selection algorithm, and offspring are generated using an arithmetic crossover algorithm and a polynomial mutation algorithm. During the iteration process, individuals are constrained using a hybrid algorithm of feasibility rules and penalty functions (the penalty coefficient adapts to the degree of constraint violation) based on the constraint condition sequence. For individuals that exceed the physical range of process parameters or violate related constraints, a parameter repair algorithm based on the target product process ratio node network is used for repair. Using forward and backward fitness functions, a batch matrix operation algorithm is used to calculate the predicted performance and backward inference parameters of an individual, and the fitness value of the individual is calculated based on the target fitness function. A simulated annealing algorithm based on the Metropolis criterion is introduced to calculate the acceptance probability of new individuals after each genetic operation, accepting inferior solutions with a certain probability to avoid local optima. Combined with a preset training period and dual control through an iteration counter and early stopping mechanism, training stops when the maximum number of iterations is reached or the fitness converges. Finally, a deviation matrix calculation algorithm is used to obtain the performance prediction deviation and the deviation of the forward and backward mixing process parameters. If the deviation does not simultaneously meet the threshold, the first execution action is triggered. In this embodiment, the parameter repair algorithm searches for feasible alternative values ​​through graph traversal.

[0090] Based on the parameters and performance deviations in the first execution action information, a reinforcement learning algorithm combined with a deep search algorithm and a target product process ratio node network is used to perform secondary ratio, process parameter search, and corresponding performance search. The searched secondary ratio, process parameter search, and corresponding performance search are used to adjust the first influence change rate, second influence change rate, third influence change rate, first main contribution factor, first auxiliary contribution factor, second main contribution factor, and second auxiliary contribution factor, as well as the corresponding weight parameters. Forward and reverse training is then performed again to solve and adjust until the element ratio, process parameter deviation, and performance deviation simultaneously meet the corresponding thresholds. The trained forward fitness function, reverse fitness function, and corresponding element ratio, process parameter, and performance evaluation index parameters are obtained.

[0091] It should be further explained that, in this embodiment, one implementation method of using reinforcement learning algorithms combined with deep search algorithms and a target product process ratio node network to perform secondary ratioing, process parameter search, and corresponding performance search is as follows:

[0092] Based on the parameters and performance deviations in the first execution action information, a state space construction algorithm is first used to map the parameter deviations and performance deviations into reinforcement learning state vectors. Based on the topology of the target product process ratio node network, a node embedding algorithm is used to map the ratio, process parameters, and performance indicators into low-dimensional vector representations. Based on this state space, a deep deterministic policy gradient network is constructed using a policy network initialization algorithm, taking the state vectors as input and outputting the adjustment amounts of the ratio and process parameters. In each training cycle, parameter adjustment actions are generated using the Ornstein-Uhlenbeck Process (stochastic process). Based on the connection relationships of the target product process ratio node network, a graph traversal algorithm is used to calculate the propagation path of the adjustment actions in the node network and predict the impact of parameter adjustments on nodes at each level. Using the forward fitness function and the backward fitness function, a performance prediction and backward inference algorithm is used to calculate the performance indicators corresponding to the new parameters and the backward inference parameters to obtain a new state vector. Based on the difference between the new and old state vectors, an immediate reward (positively correlated with the reduction in deviation) is calculated using a reward function design algorithm, and the state-action-reward sequence is stored using an experience replay mechanism. The policy network parameters are updated using a policy gradient algorithm, and simultaneously... The target network is updated using a target network stabilization algorithm. Combined with a deep search algorithm, after each action, a Monte Carlo tree search algorithm is used to perform a multi-step look-ahead search of the parameter space to evaluate the value of different adjustment paths. Based on the search results, a path optimization algorithm is used to select the optimal adjustment path and update the proportions and process parameters. Using the updated parameters, the change rates of the first, second, and third influences are recalculated using an influence factor calculation algorithm, and the weight parameters of the primary and secondary contribution factors are adjusted using a weight learning algorithm. This process is repeated until the deviations in proportions and process parameters, along with the performance deviations, simultaneously meet the thresholds, ultimately obtaining the trained positive and negative fitness functions and their corresponding parameters and performance indicators.

[0093] This embodiment achieves intelligent control across the entire process, from parameter modeling to dynamic optimization, through a multi-algorithm deep fusion process optimization framework. At the function construction level, multi-dimensional data tensor recombination and quantile standardization algorithms are used to transform the hierarchical correlation of contribution factors at all levels into a structured feature matrix. Combined with regularized multivariate regression and nested cross-validation, the nonlinear mapping relationship between performance indicators and matching parameters is effectively captured, improving the generalization ability of the forward prediction function and the parameter inversion accuracy of the backward inference function. The construction process of the genetic algorithm input sequence uses time series alignment and feature splicing techniques to uniformly encode information such as the rate of change of multi-source heterogeneous influences and weight parameters, providing comprehensive data support for subsequent optimization. In the constraint processing and optimization stage, a multi-dimensional constraint condition sequence constructed based on the improved Apriori algorithm and graph traversal technology accurately integrates the dynamic association rules between parameters, physical equipment constraints, and time series dependencies. Combined with dynamic multi-objective evolutionary algorithms and simulated annealing mechanisms, global optimization under complex constraints is achieved, avoiding the defect of traditional algorithms getting trapped in local optima. When the optimization result of the genetic algorithm fails to reach the threshold, a secondary search is performed based on the topology of the process ratio node network using reinforcement learning combined with Monte Carlo tree search. State space mapping and policy network updates are used to dynamically adjust the weights of influencing factors, forming a closed-loop iterative mechanism of prediction, reasoning, optimization, and feedback. This embodiment significantly enhances the robustness and adaptability of process parameter optimization through the synergistic complementarity between algorithms: tensor modeling improves the correlation mining capability of multi-level parameters, the combination of genetic algorithm and simulated annealing ensures global optimization efficiency, the secondary search mechanism of reinforcement learning solves the problem of parameter fine-tuning under complex operating conditions, and blockchain constraints and graph network topology ensure the feasibility of the optimization process and production safety.

[0094] Example 2

[0095] Please see Figure 2 Another embodiment of the present invention provides a performance prediction system based on the element ratio and process parameters of aluminum products, comprising: a data processing module, a function construction module, a matching prediction module, and a strengthening adjustment module;

[0096] The data processing module is used to obtain the historical element ratio sequence, process parameter sequence, process production control parameter sequence and corresponding performance evaluation index parameter sequence of the target product, and combine them with graph algorithm to obtain the process ratio node network of the target product.

[0097] The function construction module, based on the element ratio sequence, process parameter sequence, process production control parameter sequence and performance evaluation index parameter sequence, combined with multiple regression algorithm and factor analysis algorithm, obtains the ratio-performance positive prediction function and the performance-ratio inverse inference function.

[0098] The matching prediction module is used to obtain the real-time performance requirements of the target product. By combining the parameter decomposition matching model with the target product process ratio node network with the built-in performance-ratio inverse inference function, the initial element ratio sequence and process parameter sequence of the matching are obtained. By the performance prediction model with the built-in ratio-performance forward prediction function and the performance-ratio inverse inference function, the corresponding predicted performance and the corresponding inverse inference quadratic element ratio sequence and process parameter sequence are obtained.

[0099] The enhancement adjustment module, based on the parameter deviations corresponding to the initial element ratio sequence and process parameter sequence, and the secondary element ratio sequence and process parameter sequence, as well as the parameter deviations between real-time performance requirements and predicted performance, combines reinforcement learning and simulation algorithms to perform forward and reverse cyclic approximation adjustments until the parameter deviations and performance prediction deviations simultaneously meet the corresponding preset thresholds, and outputs the element ratio and process parameter sequences that meet the threshold conditions.

[0100] Example 3

[0101] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a performance prediction method based on the elemental ratio and process parameters of aluminum products.

[0102] A computer-readable storage medium having computer instructions stored thereon, which, when executed, perform a performance prediction method based on the elemental ratios and process parameters of aluminum products.

[0103] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A method for performance prediction based on elemental composition of aluminum products and process parameters, characterized in that, The application relates to a method for constructing a target product process ratio node network. The element ratio sequence comprises the mole fraction or mass fraction of main elements and refiners in aluminum products; the process parameter sequence comprises casting temperature, mold temperature, solid solution temperature, aging temperature, solid solution time, aging time and cooling rate; the process production control parameter sequence comprises pressure parameters, equipment operation parameters and environmental parameters; the performance evaluation index parameter sequence comprises mechanical property parameters, physical property parameters and process performance parameters; the main elements comprise Al, Cu and Mg; the mechanical property parameters comprise tensile strength, yield strength and elongation; the physical property parameters comprise thermal conductivity and electrical conductivity; and the process performance parameters comprise casting fluidity and hot cracking tendency. Based on the element ratio sequence, the process parameter sequence, the process production control parameter sequence and the performance evaluation index parameter sequence, a ratio-performance positive prediction function and a performance-ratio reverse reasoning function are obtained by combining a multiple regression algorithm and a factor analysis algorithm. Real-time performance requirements of the target product are obtained, and an initial element ratio sequence and a process parameter sequence matched by a parameter disassembly matching model with the performance-ratio reverse reasoning function are obtained by combining the target product process ratio node network; and corresponding predicted performance and secondary element ratio sequence and process parameter sequence obtained by reverse reasoning are obtained by a performance prediction model with the ratio-performance positive prediction function and the performance-ratio reverse reasoning function. Based on the parameter deviation of the initial element ratio sequence and the process parameter sequence and the secondary element ratio sequence and the process parameter sequence and the parameter deviation of the real-time performance requirements and the predicted performance, positive and negative cycle adjustments are carried out by combining reinforcement learning and simulation algorithms until the parameter deviation and the performance prediction deviation simultaneously satisfy corresponding preset threshold values, and the element ratio and the process parameter sequence satisfying the threshold condition are output. The construction process of the target product process ratio node network comprises the following steps:

2. The method for performance prediction based on aluminum product element ratio and process parameter of claim 1, wherein, Based on the target product historical element ratio sequence, the process parameter sequence, the process production control parameter sequence and the corresponding performance evaluation index parameter sequence, denoising and the same target product label alignment preprocessing are carried out. Based on the historical element ratio sequence, the process parameter sequence, the process production control parameter sequence and the corresponding performance evaluation index parameter sequence after preprocessing, element ratio and process parameter influence factors and influence weights corresponding to each process production control parameter and process production control parameter influence factors and corresponding influence weights corresponding to each performance evaluation index parameter are obtained by a factor analysis algorithm. The construction process of the target product process ratio node network further comprises the following steps:

3. The method for predicting performance based on aluminum product element ratio and process parameters according to claim 2, wherein, Based on the element ratio and process parameter influence factors and influence weights corresponding to each process production control parameter and the process production control parameter influence factors and corresponding influence weights corresponding to each performance evaluation index parameter, main contribution factors and auxiliary contribution factors corresponding to each performance evaluation index parameter and corresponding contribution degree coefficients are obtained by combining a principal component algorithm and a preset contribution degree threshold. ​ The main contribution factors include a first main contribution factor and a second main contribution factor; the first main contribution factor is a process production control parameter corresponding to a contribution degree coefficient greater than a contribution degree threshold value for each target product; the second main contribution factor is an element ratio or a process parameter corresponding to a contribution degree coefficient greater than a contribution degree threshold value for each target product; the auxiliary contribution factors include a first auxiliary contribution factor and a second auxiliary contribution factor; the first auxiliary contribution factor is a process production control parameter corresponding to a contribution degree coefficient less than or equal to a contribution degree threshold value for each target product; and the second auxiliary contribution factor is an element ratio or a process parameter corresponding to a contribution degree coefficient less than or equal to a contribution degree threshold value for each target product.

4. The method for predicting performance based on aluminum product element ratio and process parameters according to claim 3, characterized in that, The construction process of the target product process ratio node network further includes: Based on the control variable method and the correlation algorithm, the process production control parameter sequence and the corresponding performance evaluation index parameter sequence are analyzed to obtain a first influence change rate of a corresponding performance evaluation index parameter when any first main contribution factor or first auxiliary contribution factor corresponding to each performance evaluation index parameter changes by a preset unit amount within the same time period; Similarly, based on the control variable method and the correlation algorithm, the element ratio sequence, the process parameter sequence and the corresponding process production control parameter sequence are analyzed to obtain a second influence change rate of a corresponding first main contribution factor or first auxiliary contribution factor when a second main contribution factor changes by a preset unit amount, and a third influence change rate of a corresponding first main contribution factor or first auxiliary contribution factor when a second auxiliary contribution factor changes by a preset unit amount; The target product is set as a first-level node, the performance evaluation index parameter is set as a second-level node, the first main contribution factor and the first auxiliary contribution factor are set as a first main third-level node and a first auxiliary third-level node respectively, and the second main contribution factor and the second auxiliary contribution factor are set as a second main fourth-level node and a second auxiliary fourth-level node.

5. The method for performance prediction based on aluminum product element ratio and process parameter of claim 4, wherein, The construction process of the target product process ratio node network further includes: Based on the relationship between the target product and the performance evaluation index parameter, a first-level index connection is constructed, a second-level main contribution connection sequence is constructed based on the first main contribution factor corresponding to each performance evaluation index parameter, and a second-level auxiliary contribution sequence is constructed by using the contribution degree coefficient and the first influence change rate of the first auxiliary contribution factor; Based on the contribution degree coefficient and the second influence change rate of any second main contribution factor to any first main contribution factor or first auxiliary contribution factor, a third-level main contribution connection is constructed; and based on the contribution degree coefficient and the third influence change rate of any second auxiliary contribution factor to any first main contribution factor or first auxiliary contribution factor, a third-level auxiliary contribution connection is constructed; Based on the first-level node, the second-level node, the first main third-level node and the first auxiliary third-level node, the second main fourth-level node and the second auxiliary fourth-level node, the first-level index connection, the second-level main contribution connection sequence and the second-level auxiliary contribution sequence, and the third-level main contribution connection and the third-level auxiliary contribution connection, a target product process ratio node network is constructed by using a graph algorithm.

6. The method for predicting performance based on aluminum product element ratio and process parameters according to claim 5, wherein, The construction process of the target product process ratio node network further includes: Based on the first primary node, the first secondary node, the second primary node, the second secondary node and the corresponding connection relationship of each target product corresponding to the process matching node network of the target product, a product process chain is constructed through a block chain algorithm; The process production control parameters and element matching and process parameters saved by the first primary node and the first secondary node, the second primary node and the second secondary node in each product process chain are analyzed, and the process production control parameters and element matching and process parameters saved by the first primary node and the first secondary node and the second primary node and the second secondary node are sorted according to the time priority of the process production control parameters and the corresponding element matching and process parameters, and a product process hierarchical parameter chain is obtained.

7. The method for performance prediction based on aluminum product element ratio and process parameter of claim 6, wherein, The construction process of the matching-performance positive prediction function and the performance-matching reverse reasoning function includes: Based on the contribution degree coefficient between each performance evaluation index parameter in the product process hierarchical parameter chain corresponding to each target product and the corresponding first primary contribution factor or first secondary contribution factor, the contribution degree coefficient between the second primary contribution factor and the corresponding first primary contribution factor or first secondary contribution factor, and the contribution degree coefficient between the second secondary contribution factor and the corresponding first primary contribution factor or first secondary contribution factor, a matching-performance positive prediction function corresponding to each performance evaluation index parameter of the target product is obtained through a multiple regression algorithm; Based on the first influence change rate of each performance evaluation index parameter and the corresponding first primary contribution factor or first secondary contribution factor, the second influence change rate of each second primary contribution factor and the corresponding first primary contribution factor or first secondary contribution factor, and the third influence change rate of each second secondary contribution factor and the corresponding first primary contribution factor or first secondary contribution factor in the product process hierarchical parameter chain under the same time period, an initial performance-matching reverse reasoning function is constructed by combining a multiple regression algorithm; The first influence change rate, the second influence change rate, the third influence change rate, the first primary contribution factor and the first secondary contribution factor and the second primary contribution factor and the second secondary contribution factor corresponding to the matching-performance positive prediction function and the corresponding weight parameters and the weight sequence corresponding to the initial performance-matching reverse reasoning function are used as the input sequence of the genetic algorithm.

8. The method for performance prediction based on aluminum product element ratio and process parameter of claim 7, wherein, The construction process of the matching-performance positive prediction function and the performance-matching reverse reasoning function also includes: The matching-performance positive prediction function is used as a forward fitness function, the performance-matching reverse reasoning function is used as a reverse fitness function, and the minimum parameter deviation of the performance corresponding to the forward fitness function and the reverse fitness function is used as a target fitness function, and a constraint condition sequence in the genetic algorithm is constructed based on the constraints in the process production control parameter sequence and the constraints of the process parameters and the corresponding associated constraints between the process parameters and the process production control parameters; The dynamic multi-objective evolutionary algorithm optimized by the genetic algorithm is simulated and trained to obtain the performance deviation, the forward and reverse matching ratio, and the process parameter value deviation based on the input sequence, the forward fitness function, the reverse fitness function, the target fitness function, and the constraint condition sequence. If the deviation between the input matching ratio and process parameter corresponding to the forward fitness function and the matching ratio and process parameter predicted by the reverse fitness function does not satisfy the corresponding preset deviation threshold, and the deviation between the predicted performance corresponding to the forward fitness function and the input performance corresponding to the reverse fitness function also does not satisfy the corresponding preset deviation threshold, a first execution action information is triggered. Based on the parameters and performance deviation in the first execution action information, secondary matching, process parameter searching, and corresponding performance searching are performed by the reinforcement learning algorithm combined with the deep search algorithm and the target product process matching node network. The first influence change rate, the second influence change rate, the third influence change rate, the first main contribution factor and the first auxiliary contribution factor, the second main contribution factor and the second auxiliary contribution factor, and the corresponding weight parameters are adjusted by the searched secondary matching, process parameter searching, and corresponding performance searching, and the forward and reverse training is performed again to adjust until the element matching ratio and process parameter deviation and the performance deviation satisfy the corresponding threshold at the same time, and the trained forward fitness function, reverse fitness function, and corresponding element matching ratio and process parameter and performance evaluation index parameter are obtained.

9. A system for performance prediction based on elemental composition and process parameters of an aluminum product for implementing the method for performance prediction based on elemental composition and process parameters of an aluminum product according to any one of claims 1 to 8, characterized in that, It comprises: a data processing module and a function construction module. The data processing module is configured to obtain a target product historical element matching ratio sequence, a process parameter sequence, a process production control parameter sequence, and a corresponding performance evaluation index parameter sequence, and obtain a target product process matching node network combined with a graph algorithm. The function construction module is configured to obtain a matching ratio-performance forward prediction function and a performance-matching ratio reverse reasoning function based on the element matching ratio sequence, the process parameter sequence, the process production control parameter sequence, and the performance evaluation index parameter sequence combined with a multiple regression algorithm and a factor analysis algorithm.

10. The system for performance prediction based on aluminum product element ratio and process parameter of claim 9, wherein, The system further comprises a matching prediction module and a reinforcement adjustment module. The matching prediction module is configured to obtain a target product real-time performance requirement, obtain a matched initial element matching ratio sequence and process parameter sequence by a parameter disassembly matching model with the performance-matching ratio reverse reasoning function combined with the target product process matching node network, and obtain a corresponding predicted performance and a corresponding reverse reasoning secondary element matching ratio sequence and process parameter sequence by a performance prediction model with the matching ratio-performance forward prediction function and the performance-matching ratio reverse reasoning function. The reinforcement adjustment module is configured to perform forward and reverse cyclic approximation adjustment based on the parameter deviation of the initial element matching ratio sequence and the process parameter sequence and the secondary element matching ratio sequence and the process parameter sequence, and the parameter deviation of the real-time performance requirement and the predicted performance, until the parameter deviation and the performance prediction deviation satisfy the corresponding preset threshold at the same time, and output the element matching ratio and process parameter sequence satisfying the threshold condition.

Images