Artificial Intelligence-Based Production Material Requirement Prediction Method and System

Abstract

The present invention discloses a production material demand forecasting method and system based on artificial intelligence, belonging to the field of demand forecasting technology. The method includes: integrating production material data, optimizing production material data, building a production material demand forecasting model, and generating a forecast report. This solution clusters feature vectors and constructs feature subsets to obtain local classification confidence of the data. This is combined with a weight regularization term to construct a loss function, update the feature vector weights, analyze the feature importance to obtain a smooth threshold vector, perform feature vector screening, and reduce the complexity of the data. It also combines node attribute similarity and neighbor difference quantification to calculate bimodal dynamic weights and update node attributes. It introduces a second-order disturbance term and the sum of squared deviations to map node attributes, dynamically correct the noise distribution through gradients, sample potential variables, and introduce a cubic deviation term to reconstruct node attributes, thereby improving the accuracy and timeliness of production material demand forecasting.

Description

Technical Field

[0001] The present invention belongs to the technical field of demand forecasting, and specifically refers to a production material demand forecasting method and system based on artificial intelligence. Background Art

[0002] The production material demand forecasting method is an intelligent solution that uses artificial intelligence and big data analysis technology to integrate multi-dimensional data of historical production materials to predict future production material demand. It can achieve precise control of material inventory, reduce warehousing costs and out-of-stock risks, improve the reliability of production plans and supply chain collaboration efficiency, and help enterprises achieve lean production.

[0003] However, the existing production material demand forecasting methods have the problem that production material demand forecasting involves a large number of features, many of which are highly correlated or contain noise. The factors that really affect demand may be masked by a large number of redundant features, resulting in low accuracy and efficiency of demand forecasting; the existing production material demand forecasting methods have the problem that production material demand is affected by the interaction of many factors, which makes material demand exhibit highly nonlinear and dynamically changing properties. Traditional methods find it difficult to accurately capture these properties, resulting in large deviations in demand forecasting and low timeliness. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a production material demand forecasting method and system based on artificial intelligence. In view of the problem that the existing production material demand forecasting method involves a large number of features, many of which are highly correlated or contain noise, and the factors that really affect the demand may be masked by a large number of redundant features, resulting in low accuracy and efficiency of demand forecasting, this solution clusters the feature vectors and constructs feature subsets to obtain the local classification confidence of the data, combines them with the weight regularization term to construct a loss function, updates the feature vector weights, combines to obtain a normalized feature weight vector, analyzes the feature importance to obtain a smooth threshold vector, performs feature vector screening based on the two vectors, and completes the optimization, which can focus more on the factors that really affect the demand, reduce the influence of noise or redundant features, simplify the feature space, reduce the complexity of the data, and improve the accuracy and efficiency of production material demand forecasting. Efficiency: In view of the fact that the existing production material demand forecasting methods are affected by the interaction of many factors, the material demand presents highly nonlinear and dynamically changing properties. Traditional methods find it difficult to accurately capture these properties, resulting in large demand forecast deviations and low timeliness. This solution combines node attribute similarity and neighbor difference quantification, calculates bimodal dynamic weights and updates node attributes, introduces second-order disturbance terms and the sum of squared deviations, maps node attributes, dynamically corrects noise distribution through gradients, samples latent variables, introduces cubic deviation terms to reconstruct node attributes, and obtains the final node attributes through multiple iterations and makes predictions, thereby improving the accuracy of the material relationship description, so that the updated node attributes can better reflect the actual status of materials under complex relationships, and the latent variables can better reflect the intrinsic characteristics of material demand affected by multiple factors, thereby improving the accuracy and timeliness of production material demand forecasting.

[0005] The technical solution adopted by the present invention is as follows: The production material demand forecasting method based on artificial intelligence provided by the present invention comprises the following steps:

[0006] Step S1: Production material data integration;

[0007] Step S2: Optimize production material data;

[0008] Step S3: Build a production material demand forecasting model;

[0009] Step S4: Generate a forecast report.

[0010] Furthermore, in step S1, the production material data integration is to collect historical production material time series data and process the historical production material time series data;

[0011] The acquisition of historical production material time series data includes acquisition timestamps, material data, production data, order data and demand quantities;

[0012] The processing of the historical production material time series data is to perform data cleaning, data normalization, data coding and data set construction on the historical production material time series data to obtain a production material time series data set.

[0013] Furthermore, in step S2, the production material data optimization specifically includes the following steps:

[0014] Step S21: clustering of feature vectors: using the K-means clustering algorithm, all feature vectors in the production material time series dataset E are divided into K feature clusters, and the weight of each feature vector is initialized to 1 to generate a Q-dimensional initial feature weight vector;

[0015] Step S22: Construct feature subsets; create A empty sets, and for each feature cluster, randomly select The feature vectors are assigned to A sets in turn, and the assigned feature vectors are removed from the feature cluster; if the number of feature vectors in the feature cluster after removal is not zero, the remaining feature vectors in the feature cluster are uniformly and randomly assigned to any set, and the assigned feature vectors are removed from the feature cluster again until the number of feature vectors in each feature cluster is zero, and A feature subsets are obtained; where Z k is the number of eigenvectors in the kth feature cluster, k is the feature cluster index, is the floor operator;

[0016] Step S23: updating the feature vector weight; comprising the following steps:

[0017] Step S231: Calculate the local classification confidence; based on the feature vector weight, calculate the weighted distance between each two data in the production material time series data set under each feature subset, then introduce the temperature coefficient to control the distribution smoothness to obtain the similarity between each two data, and perform normalization to obtain the local classification confidence of each data under each feature subset;

[0018] Step S232: defining a loss function; combining the local classification confidence with the weight regularization term to construct a loss function;

[0019] Step S233: Update; pre-set the first convergence threshold P1 and the maximum update number C1, for each feature subset, use the gradient ascent method to iteratively update the feature vector weight until Or when the maximum number of iterations is reached, the update is stopped and the feature weight vector at this time is output; where ω a is the feature weight vector corresponding to the a-th feature subset, Yes a The corresponding loss function, J maxis the maximum loss function value obtained in the current update process;

[0020] Step S234: combining; recombining the updated feature weight vectors of the A feature subsets according to the index of the initial feature weight vector to obtain a Q-dimensional feature weight vector and performing linear normalization to obtain a Q-dimensional normalized feature weight vector h;

[0021] Step S24: Feature importance analysis: Calculate the Pearson correlation coefficient between every two eigenvectors in the production material time series dataset to obtain the redundancy score of each eigenvector with the remaining eigenvectors, combine the normalized feature weight vector and the redundancy score to generate a deterministic threshold vector, and perform a sliding average filter on the deterministic threshold vector to obtain a smoothed threshold vector.

[0022] Step S25: Feature vector screening; if the value of the vth column of the normalized feature weight vector h is greater than the value of the vth column of the smoothing threshold vector, the corresponding vth column feature vector in the production material time series dataset is used as a valid feature vector; otherwise, the vth column feature vector is used as a redundant feature vector; a demand forecast dataset is constructed based on all valid feature vectors to complete the optimization.

[0023] Furthermore, in step S3, the construction of the production material demand forecasting model specifically includes the following steps:

[0024] Step S31: Construct an adjacency matrix; use the demand forecast dataset as input data, treat each data in the demand forecast dataset as a node, the node attribute is the corresponding production material time series data, and pre-set a similarity threshold L. If the cosine similarity between two node attributes is less than the similarity threshold, then there is an edge between the corresponding two nodes, and the edge weight is 1; otherwise, there is no edge between the corresponding two nodes, and the edge weight is 0; obtain the adjacency matrix, and perform symmetric normalization on the adjacency matrix;

[0025] Step S32: Update node attributes; combine node attribute similarity and neighbor difference quantification to calculate the bimodal dynamic weight between each node and its neighboring nodes, and complete the node attribute update based on the bimodal dynamic weight; the formula used is as follows:

[0026] ;

[0027] ;

[0028] Among them, p1, p2 and p3 are node indexes, is the updated node attribute of node p1, s p1 、s p2 and s p3They are the original node attributes of node p1, node p2 and node p3 respectively. is the set of adjacent nodes of node p1, is the bimodal dynamic weight between node p1 and node p2, W and q are the learnable weight matrix and weight vector respectively, λ is the control factor, and They are node attribute similarity and neighbor difference quantification, is the activation function, is the connection operation, T is the transposition operation, δ is the adjustment factor, is the square of the L2 norm;

[0029] Step S33: Latent variable sampling: By introducing the second-order perturbation term and the sum of squared deviations, the node attributes are mapped to the mean and variance of the latent space, and the noise distribution is dynamically corrected through the gradient of the KL divergence loss. The latent variables are obtained by sampling based on the corrected distribution. The formula used is as follows:

[0030] ;

[0031] ;

[0032] ;

[0033] Where x p1 is the latent variable corresponding to node p1, μ p1 and are the mean and variance of the attribute mapping of node p1 to the latent space, and are two independent feedforward neural networks, N att is the total dimension of the node attribute, d is the dimension index, z d is the global attribute mean in the d-th dimension, , is the value of the node attribute of node p1 in the dth dimension, is a small perturbation of the node attribute of node p1 in the dth dimension, and are the second-order disturbance term and the sum of squared deviations, is the element-wise multiplication operator, is a random noise vector following a standard normal distribution, , N node is the number of nodes, e is the node index, is a standard normal distribution with mean 0 and covariance matrix I, is the KL divergence loss, is the e-th component of the noise vector, yes right The partial derivative of It's a network μ right The partial derivative of

[0034] Step S34: Reconstruct node attributes; introduce cubic deviation terms and use latent variables to reconstruct node attributes; the formula used is as follows:

[0035] ;

[0036] in, is the node attribute of node p1 after reconstruction, is a decoder that reconstructs node attributes using latent variables, is the latent variable mean in the w-th dimension, , is the value of the latent variable corresponding to node p1 in the wth dimension, N var is the total dimension of the latent variable, w is the dimension index, is the cubic deviation term;

[0037] Step S35: Prediction; pre-set the second convergence threshold P2 and the maximum number of iterations C2, and repeat steps S32-S34 until the mean square error between the reconstructed node attributes and the original node attributes is less than or equal to P2 or the maximum number of iterations is reached. Stop the iteration and input the node attributes U at this time into the spatiotemporal convolution layer to capture the spatiotemporal dependency, and then perform mean pooling to obtain the final node attributes U final And predict the production material demand through the fully connected layer and output the predicted data label.

[0038] Furthermore, in step S4, the generation of the forecast report is to collect and process real-time production material time series data, and input it into the production material demand forecast model for forecasting, obtain the demand for the corresponding production material according to the output data label, and generate the production material demand forecast report;

[0039] The collecting of real-time production material time series data includes collecting timestamps, material data, production data and order data;

[0040] The processing of the real-time production material time series data is to perform data cleaning, data normalization and data coding processing on the real-time production material time series data.

[0041] The artificial intelligence-based production material demand forecasting system provided by the present invention includes a production material data integration module, a production material data optimization module, a production material demand forecasting model construction module, and a forecast report generation module;

[0042] The production material data integration module collects and processes historical production material time series data and sends the data to the production material data optimization module;

[0043] The production material data optimization module clusters the feature vectors and constructs feature subsets to obtain the local classification confidence of the data, combines it with the weight regularization term to construct a loss function, updates the feature vector weights, combines to obtain a normalized feature weight vector, analyzes the feature importance to obtain a smoothing threshold vector, performs feature vector screening based on the two vectors, completes the optimization, and sends the data to the production material demand forecasting model construction module;

[0044] The production material demand forecasting model module combines node attribute similarity and neighbor difference quantification, calculates bimodal dynamic weights and updates node attributes, introduces second-order disturbance terms and sum of squared deviations, maps node attributes, dynamically corrects noise distribution through gradients, samples latent variables, introduces cubic deviation terms to reconstruct node attributes, iterates multiple times to obtain the final node attributes and perform predictions, and sends the data to the forecast report generation module;

[0045] The forecast report generation module obtains the demand quantity of the corresponding production material according to the data tags output by the production material demand forecast model, and generates a production material demand forecast report.

[0046] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0047] (1) In view of the problem in existing production material demand forecasting methods that production material demand forecasting involves a large number of features, many of which are highly correlated or contain noise, and the factors that really affect demand may be masked by a large number of redundant features, resulting in low accuracy and efficiency of demand forecasting, this scheme clusters the feature vectors and constructs feature subsets to obtain the local classification confidence of the data. It combines this with the weight regularization term to construct a loss function, updates the feature vector weights, and combines them to obtain a normalized feature weight vector. The feature importance is analyzed to obtain a smooth threshold vector. Feature vectors are screened based on the two vectors to complete the optimization. This can focus more on the factors that really affect demand, reduce the impact of noise or redundant features, simplify the feature space, reduce the complexity of the data, and improve the accuracy and efficiency of production material demand forecasting.

[0048] (2) In view of the fact that the existing production material demand forecasting methods are affected by the interaction of many factors, the material demand presents highly nonlinear and dynamically changing properties. Traditional methods are difficult to accurately capture these properties, resulting in large deviations in demand forecasting and low timeliness. This scheme combines the node attribute similarity and neighbor difference quantification, calculates the bimodal dynamic weights and updates the node attributes, introduces the second-order disturbance term and the sum of squared deviations, maps the node attributes, dynamically corrects the noise distribution through the gradient, samples the latent variables, introduces the cubic deviation term to reconstruct the node attributes, obtains the final node attributes through multiple iterations and makes predictions, improves the accuracy of the material relationship description, enables the updated node attributes to better reflect the actual status of the material under complex relationships, more accurately maps the node attributes to the latent space, and the latent variables can better reflect the intrinsic characteristics of material demand affected by multiple factors. It uses the material relationship information more accurately, thereby improving the accuracy and timeliness of production material demand forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of the process of the production material demand forecasting method based on artificial intelligence provided by the present invention;

[0050] Figure 2 A schematic diagram of the production material demand forecasting system based on artificial intelligence provided by the present invention;

[0051] Figure 3 Schematic diagram of the process of step S2;

[0052] Figure 4 Schematic diagram of the process of step S3.

[0053] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0055] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0056] Example 1, see Figure 1 The present invention provides an artificial intelligence-based production material demand forecasting method, which includes the following steps:

[0057] Step S1: Production material data integration; collecting and processing historical production material time series data;

[0058] Step S2: Production material data optimization: cluster the feature vectors and construct feature subsets to obtain the local classification confidence of the data. Combined with the weight regularization term, the loss function is constructed, the feature vector weights are updated, and the normalized feature weight vector is obtained by combining. The feature importance is analyzed to obtain a smooth threshold vector. Feature vectors are screened based on the two vectors to complete the optimization.

[0059] Step S3: Construct a production material demand forecasting model; combine node attribute similarity and neighbor difference quantification, calculate bimodal dynamic weights and update node attributes, introduce second-order disturbance terms and the sum of squared deviations, map node attributes, dynamically correct noise distribution through gradients, sample latent variables, introduce cubic deviation terms to reconstruct node attributes, and iterate multiple times to obtain the final node attributes and perform predictions;

[0060] Step S4: Generate a forecast report; obtain the demand for corresponding production materials based on the data tags output by the production material demand forecast model, and generate a production material demand forecast report.

[0061] Example 2, see Figure 1 ,This embodiment is based on the above embodiment.,In step S1, the production material data integration is to collect and process the historical production material time series data;

[0062] The acquisition of historical production material time series data includes acquisition timestamps, material data, production data, order data and demand quantities;

[0063] The material data includes material type, inventory quantity, inventory turnover rate and daily consumption;

[0064] The production data includes daily production volume, production batch plan, production cycle, production schedule and production equipment status;

[0065] The order data includes order quantity and order product type;

[0066] Processing the historical production material time series data is to perform data cleaning, data normalization, data encoding and data set construction on the historical production material time series data to obtain a production material time series data set;

[0067] The data cleaning is to remove the error values, missing values and abnormal values in the data;

[0068] The data normalization is to unify the numerical data into the same range using the maximum-minimum scaling method;

[0069] The data encoding is to convert categorical data into numerical data using One-Hot encoding;

[0070] The data set is constructed by taking the material demand as the data label, and then constructing the production material time series data set based on the historical production material time series data after data cleaning, data normalization and data encoding processing.

[0071] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the production material data optimization specifically includes the following contents:

[0072] Step S21: clustering feature vectors. In production material demand forecasting, the original data is usually high-dimensional and redundant. Direct use will lead to high computational complexity and model overfitting. K-means clustering is used to group similar features into the same cluster, reducing the complexity of the feature space. The K-means clustering algorithm is used to divide all feature vectors in the production material time series dataset E into K feature clusters. The elbow method is used to determine the K value, and the weight of each feature vector is initialized to 1 to generate a Q-dimensional initial feature weight vector.

[0073] Step S22: Construct feature subsets; a single feature subset may miss key features or introduce noise, and the computational cost of the full feature set is high. By constructing feature subsets, each subset covers different feature combinations to avoid loss of key information; create A empty sets, , for each feature cluster, randomly select The feature vectors are assigned to A sets in turn, and the assigned feature vectors are removed from the feature cluster; if the number of feature vectors in the feature cluster after removal is not zero, the remaining feature vectors in the feature cluster are uniformly and randomly assigned to any set, and the assigned feature vectors are removed from the feature cluster again until the number of feature vectors in each feature cluster is zero, and A feature subsets are obtained; where Z k is the number of eigenvectors in the kth feature cluster, k is the feature cluster index, is the floor operator;

[0074] Step S23: updating the feature vector weight; comprising the following steps:

[0075] Step S231: Calculate the local classification confidence. The demand pattern of similar materials in the production data may be misjudged due to noise. The local classification confidence is used to highlight the stable demand pattern and suppress the interference of outliers. Based on the feature vector weight, the weighted distance between each two data in the production material time series data set under each feature subset is calculated. The temperature coefficient is then introduced to control the distribution smoothness to obtain the similarity between each two data. After normalization, the local classification confidence of each data under each feature subset is obtained. The formula used is as follows:

[0076] ;

[0077] ;

[0078] ;

[0079] Where, Under the ath feature subset, data b in the production material time series data set i The local classification confidence of and They are respectively the data b in the production material time series data set under the ath feature subset i and data b j The weighted distance and similarity between them; Under the ath feature subset, data b in the production material time series data set n and data b j The similarity between i 、b j and b n are the i-th, j-th, and n-th data in the production material time series dataset, i, j, and n are data indexes, and a is the feature subset index. and The data b found based on weighted distance are i and data b n The nearest neighbor set, the number of data in the nearest neighbor set is equal to , H is the number of data in the production material time series dataset, m is the feature vector index, M a is the number of eigenvectors in the ath feature subset, is the square value of the weight of the mth eigenvector in the ath feature subset, and Data b in the ath feature subseti and data b j The value on the mth eigenvector, is the temperature coefficient, ; is the indicator function, when y i =y j hour, is 1, otherwise, is 0; y i and y j Data b i and data b j Corresponding data labels;

[0080] Step S232: Define a loss function; feature weights may overly rely on a few high-variance features and ignore long-term stable features. The defined loss function prevents the weights from being overly concentrated on a single feature and prevents overfitting; the local classification confidence is combined with the weight regularization term to construct a loss function; the formula used is as follows:

[0081] ;

[0082] Where, ω a is the feature weight vector corresponding to the ath feature subset, H is the number of data in the production material time series dataset, Yes a The corresponding loss function is, is the weight regularization term, γ is the regularization adjustment factor, γ=0.01;

[0083] Step S233: Update; dynamically adjust the weights by gradient ascent method to ensure efficiency and accuracy; pre-set the first convergence threshold P1 and the maximum number of updates C1, and for each feature subset, iteratively update the feature vector weights by gradient ascent method until Or when the maximum number of iterations is reached, the update is stopped and the feature weight vector at this time is output; the formula used is as follows:

[0084] ;

[0085] Where, and are the feature weight vectors corresponding to the ath feature subset at the cth update, η is the learning rate, η=0.01, c is the update number index, yes The corresponding loss function is, yes right The gradient of J max is the maximum loss function value obtained in the current update process, P1=10 -4 , C1=100;

[0086] Step S234: combining; recombining the updated feature weight vectors of the A feature subsets according to the index of the initial feature weight vector to obtain a Q-dimensional feature weight vector and performing linear normalization to obtain a Q-dimensional normalized feature weight vector h;

[0087] Step S24: Feature importance analysis; Multicollinearity often exists in production material time series data. Simply relying on weights may retain redundant features, resulting in poor model interpretability. Feature importance analysis is used to eliminate weight mutations, improve robustness, ensure that key features are retained first, and avoid interference from repeated features. To further eliminate redundant features, the Pearson correlation coefficient between every two feature vectors in the production material time series data set is calculated to obtain the redundancy score of each feature vector with the remaining feature vectors. The normalized feature weight vector and the redundancy score are combined to generate a deterministic threshold vector. The deterministic threshold vector is subjected to sliding average filtering to obtain a smoothed threshold vector. The formula used is as follows:

[0088] ;

[0089] ;

[0090] ;

[0091] Where, is the Pearson correlation coefficient between the u-th and v-th eigenvectors, R v is the redundancy score corresponding to the vth eigenvector, u and v are eigenvector indices, max(R) is the maximum redundancy score, α is the weight adjustment factor, α=0.7, h v is the weight value corresponding to the vth eigenvector in the normalized feature weight vector h, f v and are the deterministic threshold and smoothing threshold corresponding to the vth eigenvector, o is the temporary traversal index, is the deterministic threshold corresponding to the v+oth eigenvector;

[0092] Step S25: Feature vector screening; improve the quality of training data through feature vector screening, so that the model can focus more on key factors; if the value of the vth column of the normalized feature weight vector h is greater than the value of the vth column of the smoothing threshold vector, then the corresponding vth column feature vector in the production material time series dataset is used as a valid feature vector; otherwise, the vth column feature vector is used as a redundant feature vector; construct a demand forecast dataset based on all valid feature vectors to complete the optimization.

[0093] By performing the above operations, in order to address the problem in existing production material demand forecasting methods that production material demand forecasting involves a large number of features, many of which are highly correlated or contain noise, and the factors that really affect demand may be masked by a large number of redundant features, resulting in low accuracy and efficiency of demand forecasting, this solution clusters the feature vectors and constructs feature subsets to obtain the local classification confidence of the data, combines it with the weight regularization term to construct a loss function, updates the feature vector weights, and combines them to obtain a normalized feature weight vector. The feature importance is analyzed to obtain a smoothing threshold vector, and feature vector screening is performed based on the two vectors to complete the optimization. This can focus more on the factors that really affect demand, reduce the impact of noise or redundant features, simplify the feature space, reduce the complexity of the data, and improve the accuracy and efficiency of production material demand forecasting.

[0094] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, building a production material demand forecasting model specifically includes the following steps:

[0095] Step S31: Construct an adjacency matrix; in the production material demand forecast, the time series data of different materials may have implicit correlations, but traditional methods are difficult to quantify this relationship. The nonlinear relationship of the material demand data is converted into a graph structure to avoid the interference of noise edges. Symmetric normalization can balance the node degree differences and prevent high-demand materials from excessively dominating model learning; the demand forecast data set is used as input data, and each data in the demand forecast data set is used as a node. The node attribute is the corresponding production material time series data. A similarity threshold L is pre-set. If the cosine similarity between the attributes of two nodes is less than the similarity threshold, there is an edge between the corresponding two nodes, and the edge weight is 1; otherwise, there is no edge between the corresponding two nodes, and the edge weight is 0; obtain the adjacency matrix, and perform symmetric normalization on the adjacency matrix; the formula used is as follows:

[0096] ;

[0097] Where, F and are the original adjacency matrix and the symmetric normalized adjacency matrix, G is the degree matrix, I is the identity matrix, and L=0.8;

[0098] Step S32: Update node attributes. Material demand is affected by multiple factors. A single similarity metric cannot take into account both semantic associations and local structural differences. The bimodal dynamic weight simultaneously integrates node attribute similarity and neighbor difference quantification, which can both identify collaborative demand patterns and perceive local anomalies. By combining node attribute similarity and neighbor difference quantification, the bimodal dynamic weight between each node and its adjacent nodes is calculated, and the node attributes are updated based on the bimodal dynamic weight. The formula used is as follows:

[0099] ;

[0100] ;

[0101] Among them, p1, p2 and p3 are node indexes, is the updated node attribute of node p1, s p1 、s p2 and s p3 They are the original node attributes of node p1, node p2 and node p3 respectively. is the set of adjacent nodes of node p1, is the bimodal dynamic weight between node p1 and node p2, W and q are the learnable weight matrix and weight vector respectively, λ is the control factor, λ=0.1, and They are node attribute similarity and neighbor difference quantification, is the activation function, is the connection operation, T is the transposition operation, δ is the adjustment factor, δ=0.3, is the square of the L2 norm;

[0102] Step S33: Sampling latent variables; Material demand data is uncertain, and traditional deterministic coding is difficult to characterize its probability distribution. The propagation error of the input noise to the mean is quantified by the second-order disturbance term to improve the robustness to data fluctuations. The squared deviation constrains the discreteness of the latent space to prevent the latent variables from deviating excessively from the global mean, ensuring the generation of reasonable demand scenarios. The KL divergence gradient corrects the noise to dynamically adjust the sampling direction to alleviate the posterior collapse problem under sparse demand data. By introducing the second-order disturbance term and the squared deviation, the node attributes are mapped to the mean and variance of the latent space, and the noise distribution is dynamically corrected by the gradient of the KL divergence loss. The latent variables are obtained based on the corrected distribution sampling. The formula used is as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] Where x p1 is the latent variable corresponding to node p1, μ p1 and are the mean and variance of the attribute mapping of node p1 to the latent space, and are two independent feedforward neural networks, N att is the total dimension of the node attribute, d is the dimension index, zd is the global attribute mean in the d-th dimension, , is the value of the node attribute of node p1 in the dth dimension, is a small perturbation of the node attribute of node p1 in the dth dimension, and are the second-order disturbance term and the sum of squared deviations, is the element-wise multiplication operator, is a random noise vector following a standard normal distribution, , N node is the number of nodes, e is the node index, is a standard normal distribution with mean 0 and covariance matrix I, is the KL divergence loss, is the e-th component of the noise vector, yes right The partial derivative of It's a network μ right The partial derivative of

[0107] Step S34: Reconstruct node attributes. Material demand reconstruction requires balancing overall trend fitting with local mutation recovery. The cubic bias term amplifies the penalty for latent variables deviating from the mean, forcing the model to pay more attention to abnormal demand patterns. This is superior to the traditional L2 loss, which is overly sensitive to small fluctuations. The decoder and the cubic term work together to ensure the accuracy of mainstream demand reconstruction while enhancing the ability to generate outliers. The cubic bias term is introduced to reconstruct node attributes using latent variables. The formula used is as follows:

[0108] ;

[0109] in, is the node attribute of node p1 after reconstruction, is a decoder that reconstructs node attributes using latent variables, is the latent variable mean in the w-th dimension, , is the value of the latent variable corresponding to node p1 in the wth dimension, N var is the total dimension of the latent variable, w is the dimension index, is the cubic deviation term;

[0110] Step S35: Forecast; Material demand has spatiotemporal dependency, the spatiotemporal convolution layer jointly processes the graph structure and time series, captures cross-material collaboration and time series lag effects, aggregates multi-node information through mean pooling, suppresses single material noise, and outputs a stable forecast; pre-set the second convergence threshold P2 and the maximum number of iterations C2, repeat steps S32-S34 until the mean square error between the reconstructed node attributes and the original node attributes is less than or equal to P2 or the maximum number of iterations is reached, and the node attributes U at this time are input into the spatiotemporal convolution layer to capture spatiotemporal dependency, and then perform mean pooling to obtain the final node attributes U final And predict the production material demand through the fully connected layer and output the predicted data label; the formula used is as follows:

[0111] ;

[0112] in, and They are mean pooling function and spatiotemporal convolution layer, P2=10 -3 , C2=150.

[0113] By performing the above operations, the existing production material demand forecasting methods have the problem that production material demand is affected by the interaction of many factors, which makes the material demand present highly nonlinear and dynamically changing properties. Traditional methods find it difficult to accurately capture these properties, resulting in large demand forecast deviations and low timeliness. This scheme combines node attribute similarity and neighbor difference quantification, calculates bimodal dynamic weights and updates node attributes, introduces second-order disturbance terms and the sum of squared deviations, maps node attributes, dynamically corrects noise distribution through gradients, samples latent variables, introduces cubic deviation terms to reconstruct node attributes, obtains the final node attributes through multiple iterations and makes predictions, thereby improving the accuracy of material relationship descriptions, enabling the updated node attributes to better reflect the actual status of materials under complex relationships, more accurately map node attributes to latent space, and latent variables to better reflect the intrinsic characteristics of material demand affected by multiple factors. Material relationship information is more accurately utilized, thereby improving the accuracy and timeliness of production material demand forecasts.

[0114] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, generating a forecast report involves collecting and processing real-time production material time series data, inputting the data into a production material demand forecasting model for forecasting, obtaining the demand for the corresponding production material based on the output data labels, and generating a production material demand forecasting report.

[0115] The collecting of real-time production material time series data includes collecting timestamps, material data, production data and order data;

[0116] The processing of the real-time production material time series data is to perform data cleaning, data normalization and data coding processing on the real-time production material time series data.

[0117] Example 6, see Figure 2 This embodiment is based on the above embodiment. The artificial intelligence-based production material demand forecasting system provided by the present invention includes a production material data integration module, a production material data optimization module, a production material demand forecasting model building module, and a forecast report generation module;

[0118] The production material data integration module collects and processes historical production material time series data and sends the data to the production material data optimization module;

[0119] The production material data optimization module clusters the feature vectors and constructs feature subsets to obtain the local classification confidence of the data, combines it with the weight regularization term to construct a loss function, updates the feature vector weights, combines to obtain a normalized feature weight vector, analyzes the feature importance to obtain a smoothing threshold vector, performs feature vector screening based on the two vectors, completes the optimization, and sends the data to the production material demand forecasting model construction module;

[0120] The production material demand forecasting model module combines node attribute similarity and neighbor difference quantification, calculates bimodal dynamic weights and updates node attributes, introduces second-order disturbance terms and sum of squared deviations, maps node attributes, dynamically corrects noise distribution through gradients, samples latent variables, introduces cubic deviation terms to reconstruct node attributes, iterates multiple times to obtain the final node attributes and perform predictions, and sends the data to the forecast report generation module;

[0121] The forecast report generation module obtains the demand quantity of the corresponding production material according to the data tags output by the production material demand forecast model, and generates a production material demand forecast report.

[0122] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0123] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0124] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. The production material demand forecasting method based on artificial intelligence is characterized by: The method comprises the following steps: Step S1: Production material data integration; collecting and processing historical production material time series data; Step S2: Production material data optimization: cluster the feature vectors and construct feature subsets to obtain the local classification confidence of the data. Combined with the weight regularization term, the loss function is constructed, the feature vector weights are updated, and the normalized feature weight vector is obtained by combining. The feature importance is analyzed to obtain a smooth threshold vector. Feature vectors are screened based on the two vectors to complete the optimization. Step S3: Construct a production material demand forecasting model; combine node attribute similarity and neighbor difference quantification, calculate bimodal dynamic weights and update node attributes, introduce second-order disturbance terms and the sum of squared deviations, map node attributes, dynamically correct noise distribution through gradients, sample latent variables, introduce cubic deviation terms to reconstruct node attributes, and iterate multiple times to obtain the final node attributes and perform predictions; Step S4: Generate a forecast report; obtain the demand for corresponding production materials based on the data tags output by the production material demand forecast model, and generate a production material demand forecast report; Step S3 includes step S32: updating node attributes; combining node attribute similarity and neighbor difference quantification, calculating the bimodal dynamic weight between each node and its neighboring nodes, and completing the node attribute update based on the bimodal dynamic weight; the formula used is as follows: ； ； Among them, p1, p2 and p3 are node indexes, is the updated node attribute of node p1, s p1 、s p2 and s p3 They are the original node attributes of node p1, node p2 and node p3 respectively. is the set of adjacent nodes of node p1, is the bimodal dynamic weight between node p1 and node p2, W and q are the learnable weight matrix and weight vector respectively, λ is the control factor, and They are node attribute similarity and neighbor difference quantification, is the activation function, is the connection operation, T is the transposition operation, δ is the adjustment factor, is the square of the L2 norm; Step S3 includes step S34: reconstructing node attributes; introducing a cubic deviation term and using latent variables to reconstruct node attributes; the formula used is as follows: ； in, is the node attribute of node p1 after reconstruction, is a decoder that reconstructs node attributes using latent variables, is the latent variable mean in the w-th dimension, is the value of the latent variable corresponding to node p1 in the wth dimension, N var is the total dimension of the latent variable, w is the dimension index, is the cubic deviation term, p1 is the node index, x p1 is the latent variable corresponding to node p1, and λ is the control factor.

2. The artificial intelligence-based production material demand forecasting method according to claim 1, characterized in that: In step S3, the construction of the production material demand forecasting model specifically includes the following steps: Step S31: Construct an adjacency matrix; use the demand forecast dataset as input data, treat each data in the demand forecast dataset as a node, the node attribute is the corresponding production material time series data, and pre-set a similarity threshold L. If the cosine similarity between two node attributes is less than the similarity threshold, then there is an edge between the corresponding two nodes, and the edge weight is 1; otherwise, there is no edge between the corresponding two nodes, and the edge weight is 0; obtain the adjacency matrix, and perform symmetric normalization on the adjacency matrix; Step S32: Update node attributes; Step S33: latent variable sampling; Step S34: Reconstructing node attributes; Step S35: Prediction; pre-set the second convergence threshold P2 and the maximum number of iterations C2, and repeat steps S32-S34 until the mean square error between the reconstructed node attributes and the original node attributes is less than or equal to P2 or the maximum number of iterations is reached. Stop the iteration and input the node attributes U at this time into the spatiotemporal convolution layer to capture the spatiotemporal dependency, and then perform mean pooling to obtain the final node attributes U final And predict the production material demand through the fully connected layer and output the predicted data label.

3. The method for predicting production material demand based on artificial intelligence according to claim 2, characterized in that: In step S33, the latent variables are sampled; by introducing a second-order perturbation term and the sum of squared deviations, the node attributes are mapped to the mean and variance of the latent space, and the noise distribution is dynamically corrected by the gradient of the KL divergence loss, and the latent variables are obtained based on the corrected distribution sampling; the formula used is as follows: ； ； ； Where x p1 is the latent variable corresponding to node p1, μ p1 and are the mean and variance of the attribute mapping of node p1 to the latent space, and are two independent feedforward neural networks, N att is the total dimension of the node attribute, d is the dimension index, z d is the global attribute mean in the d-th dimension, is the value of the node attribute of node p1 in the dth dimension, is a small perturbation of the node attribute of node p1 in the dth dimension, and are the second-order disturbance term and the sum of squared deviations, is the element-wise multiplication operator, is a random noise vector following a standard normal distribution, , N node is the number of nodes, e is the node index, is a standard normal distribution with mean 0 and covariance matrix I, is the KL divergence loss, is the e-th component of the noise vector, yes right The partial derivative of It's a network μ right The partial derivative of .

4. The method for forecasting production material demand based on artificial intelligence according to claim 1, characterized in that: In step S2, the production material data optimization specifically includes the following steps: Step S21: clustering of feature vectors: using the K-means clustering algorithm, all feature vectors in the production material time series dataset E are divided into K feature clusters, and the weight of each feature vector is initialized to 1 to generate a Q-dimensional initial feature weight vector; Step S22: Construct feature subsets; create A empty sets, and for each feature cluster, randomly select The feature vectors are assigned to A sets in turn, and the assigned feature vectors are removed from the feature cluster; if the number of feature vectors in the feature cluster after removal is not zero, the remaining feature vectors in the feature cluster are uniformly and randomly assigned to any set, and the assigned feature vectors are removed from the feature cluster again until the number of feature vectors in each feature cluster is zero, and A feature subsets are obtained; where Z k is the number of eigenvectors in the kth feature cluster, k is the feature cluster index, is the floor operator; Step S23: Update feature vector weights; Step S24: Feature importance analysis: Calculate the Pearson correlation coefficient between every two eigenvectors in the production material time series dataset to obtain the redundancy score of each eigenvector with the remaining eigenvectors, combine the normalized feature weight vector and the redundancy score to generate a deterministic threshold vector, and perform a sliding average filter on the deterministic threshold vector to obtain a smoothed threshold vector. Step S25: Feature vector screening; if the value of the vth column of the normalized feature weight vector h is greater than the value of the vth column of the smoothing threshold vector, the corresponding vth column feature vector in the production material time series dataset is used as a valid feature vector; otherwise, the vth column feature vector is used as a redundant feature vector; a demand forecast dataset is constructed based on all valid feature vectors to complete the optimization.

5. The method for forecasting production material demand based on artificial intelligence according to claim 4, characterized in that: In step S23, the updating of the feature vector weights specifically includes the following steps: Step S231: Calculate the local classification confidence; based on the feature vector weight, calculate the weighted distance between each two data in the production material time series data set under each feature subset, then introduce the temperature coefficient to control the distribution smoothness to obtain the similarity between each two data, and perform normalization to obtain the local classification confidence of each data under each feature subset; Step S232: defining a loss function; combining the local classification confidence with the weight regularization term to construct a loss function; Step S233: Update; pre-set the first convergence threshold P1 and the maximum update number C1, for each feature subset, use the gradient ascent method to iteratively update the feature vector weight until Or when the maximum number of iterations is reached, the update is stopped and the feature weight vector at this time is output; where ω a is the feature weight vector corresponding to the a-th feature subset, Yes a The corresponding loss function, J max is the maximum loss function value obtained in the current update process; Step S234: Combination; the updated feature weight vectors of the A feature subsets are recombined according to the index of the initial feature weight vector to obtain a Q-dimensional feature weight vector and linearly normalized to obtain a Q-dimensional normalized feature weight vector h.

6. The method for forecasting production material demand based on artificial intelligence according to claim 1, characterized in that: In step S1, the production material data integration is to collect historical production material time series data and process the historical production material time series data; The acquisition of historical production material time series data includes acquisition timestamps, material data, production data, order data and demand quantities; The processing of the historical production material time series data is to perform data cleaning, data normalization, data coding and data set construction on the historical production material time series data to obtain a production material time series data set.

7. The method for forecasting production material demand based on artificial intelligence according to claim 1, characterized in that: In step S4, the generation of the forecast report is to collect and process the real-time production material time series data, and input it into the production material demand forecast model for forecasting, obtain the demand for the corresponding production material according to the output data label, and generate the production material demand forecast report; The collecting of real-time production material time series data includes collecting timestamps, material data, production data and order data; The processing of the real-time production material time series data is to perform data cleaning, data normalization and data coding processing on the real-time production material time series data.

8. An artificial intelligence-based production material demand forecasting system, configured to implement the artificial intelligence-based production material demand forecasting method according to any one of claims 1 to 7, characterized in that: It includes production material data integration module, production material data optimization module, production material demand forecasting model building module and forecast report generation module; The production material data integration module collects and processes historical production material time series data and sends the data to the production material data optimization module; The production material data optimization module clusters the feature vectors and constructs feature subsets to obtain the local classification confidence of the data, combines it with the weight regularization term to construct a loss function, updates the feature vector weights, combines to obtain a normalized feature weight vector, analyzes the feature importance to obtain a smoothing threshold vector, performs feature vector screening based on the two vectors, completes the optimization, and sends the data to the production material demand forecasting model construction module; The production material demand forecasting model module combines node attribute similarity and neighbor difference quantification, calculates bimodal dynamic weights and updates node attributes, introduces second-order disturbance terms and sum of squared deviations, maps node attributes, dynamically corrects noise distribution through gradients, samples latent variables, introduces cubic deviation terms to reconstruct node attributes, iterates multiple times to obtain the final node attributes and perform predictions, and sends the data to the forecast report generation module; The forecast report generation module obtains the demand quantity of the corresponding production material according to the data tags output by the production material demand forecast model, and generates a production material demand forecast report.

Images