An electroplating power intelligent adjusting system and method based on dynamic load prediction

Abstract

This invention discloses an intelligent adjustment system and method for electroplating power supplies based on dynamic load prediction, relating to the field of dynamic load control technology. The system includes the following steps: For an electroplating power supply providing power to n electroplating production lines, historical load data, process parameters, and adjacency relationship data are input into an improved STGAN-Electro model, outputting the load prediction value for the electroplating production lines. Based on the load prediction value, a Stackelberg game model is constructed with the electroplating power supply as the leader and the n electroplating production lines as followers. The Stackelberg game model is solved to obtain the optimal resource allocation vector. Based on the optimal resource allocation vector, the output power of the electroplating power supply is adjusted, ultimately achieving a comprehensive technical effect of improving overall energy efficiency, ensuring process quality, and enhancing operational stability.

Description

Technical Field

[0001] This invention relates to the field of dynamic load control technology, and in particular to an intelligent adjustment system and method for electroplating power supplies based on dynamic load prediction. Background Technology

[0002] In recent years, with the advancement of intelligent transformation in manufacturing and refined energy management, the electroplating industry, as a typical high-energy-consuming production process, has seen dynamic regulation of its power load become crucial for improving energy efficiency, ensuring process quality, and achieving grid coordination. Traditional electroplating power supplies often employ regulation methods based on fixed rules or simple feedback, which are difficult to adapt to complex scenarios involving multiple parallel production lines, time-varying process parameters, and frequent load fluctuations, easily leading to energy waste, equipment overload, or unstable electroplating quality. Therefore, how to achieve coordinated optimization regulation based on accurate load forecasting has become a core challenge for energy conservation, consumption reduction, and stable operation in the electroplating process.

[0003] Existing dynamic load control methods mostly focus on short-term load forecasting or single-objective optimization, and still have the following significant shortcomings. Specifically, load forecasting models often rely solely on historical time-series data and fail to effectively integrate the spatiotemporal correlation between electroplating process parameters and production lines, resulting in insufficient forecast accuracy in scenarios involving process switching and production plan adjustments. Resource allocation between power supplies and production lines often adopts centralized optimization or simple game theory, failing to fully consider the quality constraints of electroplating processes and the differentiated needs of each production line, making it difficult to achieve synergistic optimization of energy efficiency and quality. Control strategies typically ignore the differences in operating efficiency of power supplies in different load ranges and system-level indicators such as peak-to-average load ratio and distribution balance, limiting the improvement of overall energy efficiency and operational stability. Summary of the Invention

[0004] The technical problem addressed by this invention is that existing dynamic load control methods mostly focus on short-term load forecasting or single-objective optimization, and still have the following significant shortcomings. Specifically, load forecasting models often rely solely on historical time-series data and fail to effectively integrate electroplating process parameters with the spatiotemporal correlation between production lines, resulting in insufficient forecast accuracy in scenarios involving process switching and production plan adjustments. Resource allocation between power supplies and production lines often employs centralized optimization or simple game theory, failing to fully consider electroplating process quality constraints and the differentiated needs of each production line, making it difficult to achieve optimal synergy between energy efficiency and quality. Control strategies typically ignore the differences in operating efficiency of power supplies in different load ranges and system-level indicators such as peak-to-average load ratio and distribution balance, limiting the improvement of overall energy efficiency and operational stability.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for intelligent adjustment of electroplating power supply based on dynamic load prediction includes the following steps: Step S1: For an electroplating power supply that powers n electroplating production lines, input historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model, and output the load prediction value of the electroplating production line. Step S2: Based on the load forecast, construct a Stackelberg game model with the electroplating power source as the leader and n electroplating production lines as followers. Step S3: Solve the Stackelberg game model to obtain the optimal resource allocation vector, and adjust the output power of the electroplating power supply based on the optimal resource allocation vector.

[0006] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, historical load data, process parameters, and adjacency relationship data are acquired. Historical load data includes the equipment operating power, number of parts processed, and electroplating process duration of the electroplating production line; Process parameters include current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters; Adjacency data includes electrical connection data, physical layout data, and process similarity data.

[0007] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, in step S1, for an electroplating power supply that supplies power to n electroplating production lines, historical load data, process parameters and adjacency relationship data are input into the improved STGAN-Electro model, and the load prediction value of the electroplating production line is output. Step S1 includes steps S101, S102, S103, S104, S105, S106 and S107. Step S101: Based on the improved STGAN-Electro model, the historical load data, process parameters and adjacency relationship data are standardized by using the Z-score standardization formula through the input fusion layer to obtain the corresponding standard historical load data, standard process data and standard adjacency relationship data. The standard historical load data and standard process data are merged to obtain merged data, and time information is added to the merged data to obtain the data to be analyzed. The time information includes hours, minutes, weekdays, and non-working days; Step S102: Project the data to be analyzed into a query matrix Q, a key matrix K, and a value matrix V using a multi-head temporal attention mechanism; Attention scores are calculated based on the query matrix Q, the key matrix K, and the value matrix V to obtain multi-head attention feature data. Multi-head attention feature data is processed through a local temporal channel to obtain local temporal feature data; Multi-head attention feature data is processed through a global temporal channel to obtain global temporal feature data; The local and global time feature data are merged to obtain dual-channel time feature data; The dual-channel temporal feature data is processed by a two-layer fully connected network, and the ReLU activation function is used to calculate the two-layer fully connected data. The time feature data is obtained by performing a residual join between the two fully connected layers of data and the dual-channel time feature data. Step S103: Use the standard adjacency relationship data as a local spatial mask; Based on temporal feature data, the similarity between electroplating production lines is calculated using a global spatial attention head to obtain a global spatial mask; Based on local spatial masks, local spatial feature data is obtained by processing temporal feature data through local spatial attention heads. Based on the global spatial mask, the temporal feature data is processed by the global spatial attention head to obtain the global spatial feature data; Local spatial feature data and global spatial feature data are concatenated to obtain dual spatial feature data; The dual-space feature data is mapped using a linear transformation matrix, and a bias term is added to obtain linearly mapped data. The linear mapping data is subjected to layer normalization to obtain spatial feature data.

[0008] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, in step S104, independent coding networks are established for the current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity and electroplating waveform parameters of the standard process data to obtain coding network data. Each coding network consists of two fully connected layers; The ReLU activation function is used to calculate the process feature vector by processing the encoded network data; The process feature vectors include current density feature vector, temperature feature vector, pH value feature vector, main salt concentration feature vector, additive concentration feature vector, stirring intensity feature vector, and electroplating waveform parameter feature vector; The current density feature vector, temperature feature vector, pH value feature vector, main salt concentration feature vector, additive concentration feature vector, stirring intensity feature vector, and electroplating waveform parameter feature vector are spliced ​​together to obtain process splicing feature data; The spliced ​​feature data is mapped using a linear transformation matrix, and a bias term is added to obtain the linear transformation process data; The linear transformation process data is subjected to layer normalization to obtain mapped process data. The process parameters are adjusted by the process adjustment weight matrix to obtain process feature data. And establish a mapping relationship between standard historical load data and process characteristic data; Step S105: The time feature data, spatial feature data and process feature data are spliced ​​together to obtain fused spliced ​​feature data; The fused and spliced ​​feature data is input into the gated weight generation network, and the time weight, spatial weight and process weight are obtained by calculation through a fully connected layer and a Softmax function. The weighted sum data is obtained by weighting and summing the time weight, spatial weight, and process weight with the corresponding time feature data, spatial feature data, and process feature data. The weighted summation data is residually concatenated with the time feature data, spatial feature data, and process feature data respectively to obtain time-enhanced feature data, spatial-enhanced feature data, and process-enhanced feature data; Temporal enhancement feature data, spatial enhancement feature data, and process enhancement feature data are concatenated and mapped to 64 dimensions through a linear transformation layer to obtain fused enhancement feature data; The weighted summation data is concatenated with the standard historical load data to obtain the first residual feature. At the same time, the weighted summation data is concatenated with the standard process parameters to obtain the second residual feature. The first residual feature and the second residual feature are concatenated to obtain fused feature data; Step S106: The fused feature data is processed by a one-dimensional convolutional layer, expanding the dimension of the fused feature data from 64 to 128, resulting in an intermediate feature with a dimension of 10×96×128. The 128-dimensional intermediate features are mapped to 64-dimensional intermediate features through a fully connected layer, and the ReLU activation function is used to calculate the 64-dimensional intermediate features to obtain adjusted feature data with dimensions of 10×96×64. Step S107: Extract the feature data of the last 16 time steps from the adjusted feature data along the time dimension, and map the 64-dimensional features of each time point to 3 dimensions through a fully connected layer to obtain feature data with dimensions of 10×16×3. The load prediction value of the electroplating production line is obtained by using a linear activation function on the feature data with dimensions of 10×16×3. The load forecast values ​​for the electroplating production line include the equipment operating power forecast, the number of parts processed forecast, and the electroplating process duration forecast.

[0009] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, step S2 involves constructing a Stackelberg game model with the electroplating power supply as the leader and n electroplating production lines as followers based on the load prediction value. Step S2 includes steps S201, S202, and S203; Step S201: Construct a quality parameter vector based on the electroplating process quality parameters. Specifically, this includes normalizing the current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters of the electroplating process quality parameters to obtain normalized current density, normalized temperature, normalized pH value, normalized main salt concentration, normalized additive concentration, normalized stirring intensity, and normalized electroplating waveform parameters. A quality parameter vector is constructed based on normalized current density, normalized temperature, normalized pH value, normalized main salt concentration, normalized additive concentration, normalized stirring intensity, and normalized electroplating waveform parameters.

[0010] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, in step S202, the expression corresponding to the utility function of the follower is: ; in, Indicates the first The effectiveness of an electroplating production line This represents a resource allocation vector, specifically including current, voltage, and power. Represents a mass parameter vector. The function representing quality satisfaction. This represents energy consumption cost, used to characterize the difference in actual operating efficiency of electroplating power supplies across different load ranges. Represents the cost of the resource allocation vector. , and Indicates the weighting coefficient; Step S203, the expression for the quality satisfaction function is: ; in, This indicates the quality satisfaction level of the electroplating production line. This represents the quality priority coefficient. This represents the tolerance variance for quality. Indicates time The mass parameter vector, Indicates time The target quality parameter vector; Energy consumption cost is calculated using a piecewise quadratic cost function, expressed as follows: ; in, The allowable operating power of the i-th electroplating production line is determined by the resource allocation vector, and k represents the load range within which the allowable operating power is located, determined by load forecast values. , and These represent the curvature coefficient, linear efficiency coefficient, and basic loss coefficient of the efficiency curve for this interval, respectively, obtained by curve fitting using historical operating efficiency data of the production line. This represents the electricity price coefficient.

[0011] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, in step S204, the expression corresponding to the leader's utility function is: ; in, R(A) represents the utility of the electroplating power source, R(A) represents the peak-to-average power ratio, and S(A) represents the balance of the resource allocation vector. , , and Indicates the weighting coefficient; The balance of the resource allocation vector The calculation is performed using a quantization formula based on information entropy, and the expression is: ; ; in, Indicates the first The proportion of resource allocation vectors for each electroplating production line; The expression for the peak-to-average load ratio is as follows. ; in, This indicates the peak-to-average load ratio. Let represent the modulus of the resource allocation vector of the i-th electroplating production line at time t, and represent its actual operating power.

[0012] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, in step S205, the solution of the Stackelberg game model satisfies the power balance constraint of the electroplating power supply. The expression for the power balance constraint of the electroplating power supply is: ; in, Indicates the upper limit of the grid access power. Indicates other auxiliary loads, This indicates the power buffer provided by an adjustable load.

[0013] As a preferred embodiment of the intelligent adjustment method for electroplating power supply based on dynamic load prediction described in this invention, step S3 involves solving the Stackelberg game model to obtain the optimal resource allocation vector, and adjusting the output power of the electroplating power supply based on the optimal resource allocation vector.

[0014] An intelligent regulation system for electroplating power supply based on dynamic load prediction includes a processing module, a construction module, and a regulation module; The processing module, for an electroplating power supply that powers n electroplating production lines, inputs historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model and outputs the load prediction value of the electroplating production lines. The module constructs a Stackelberg game model based on load forecasts, with the electroplating power source as the leader and n electroplating production lines as followers. The adjustment module solves the Stackelberg game model to obtain the optimal resource allocation vector, and adjusts the output power of the electroplating power supply based on the optimal resource allocation vector.

[0015] The beneficial effects of this invention are as follows: Firstly, addressing the problem that existing load forecasting models neglect the spatiotemporal correlation between process parameters and production lines, the STGAN-Electro model is improved by integrating historical load, process parameter, and adjacency relationship data, and introducing a dual-channel spatiotemporal attention mechanism, significantly improving prediction accuracy under complex operating conditions. Secondly, addressing the difficulty of traditional centralized optimization in balancing quality and energy efficiency, a Stackelberg game model with the power supply as the leader and the production line as the follower is constructed. Through constraints of the quality satisfaction function and benchmark mapping, process quality requirements are embedded in the resource allocation process, achieving synergistic optimization of quality and energy consumption. Thirdly, in terms of target design, by introducing a piecewise quadratic energy consumption cost function and peak-to-average power ratio and balance indexes, the power supply load efficiency characteristics are finely characterized and load stability is promoted, ultimately achieving a comprehensive technical effect of improving overall energy efficiency, ensuring process quality, and enhancing operational stability. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of an intelligent adjustment method for electroplating power supplies based on dynamic load prediction, provided as an embodiment of the present invention.

[0017] Figure 2 This is a basic flowchart of an intelligent adjustment system for electroplating power supply based on dynamic load prediction, provided as an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for intelligent adjustment of electroplating power supply based on dynamic load prediction is provided, comprising the following steps: Step S1: For an electroplating power supply that powers n electroplating production lines, input historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model, and output the load prediction value of the electroplating production line. Step S2: Based on the load forecast, construct a Stackelberg game model with the electroplating power source as the leader and n electroplating production lines as followers. Step S3: Solve the Stackelberg game model to obtain the optimal resource allocation vector, and adjust the output power of the electroplating power supply based on the optimal resource allocation vector.

[0020] In one embodiment, firstly, multi-source data, including historical load data, process parameters, and adjacency relationship data, is input into an improved STGAN-Electro model for processing to obtain load prediction values ​​for the electroplating production lines, achieving high-precision prediction of the future load of n electroplating production lines. Based on these load prediction values, a Stackelberg game model is constructed with the electroplating power supply as the leader and each production line as the follower. The utility functions of both the leader and followers embed multiple constraints and objectives related to process quality satisfaction, segmented energy consumption costs, and equilibrium. Solving this game model yields the optimal resource allocation vector under the constraints of process quality and power balance, and the electroplating power supply is adjusted in real time accordingly. The system significantly improves the accuracy of dynamic load prediction under complex electroplating conditions, providing a reliable basis for optimized control. Through game theory equilibrium, it achieves optimal resource allocation with limited power capacity, while ensuring the electroplating process quality and overall energy efficiency of each production line. It effectively smooths the total load curve, reduces the peak-to-average power ratio, and enhances operational stability and economy. Addressing the current problems of single prediction dimensions, fragmented optimization objectives, and insufficient coordination in electroplating workshop load control, the system combines multi-source data fusion prediction with multi-layer game theory collaborative optimization, linking accurate prediction with intelligent decision control. This solves the industry pain point of difficulty in coordinating the optimization of energy efficiency, quality, and load stability in electroplating production.

[0021] Acquire historical load data, process parameters, and adjacency relationship data; Historical load data includes the equipment operating power, number of parts processed, and electroplating process duration of the electroplating production line; Process parameters include current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters; Adjacency data includes electrical connection data, physical layout data, and process similarity data.

[0022] Step S1: For an electroplating power supply that powers n electroplating production lines, input historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model, and output the load prediction value of the electroplating production line. Step S1 includes steps S101, S102, S103, S104, S105, S106 and S107. Step S101: Based on the improved STGAN-Electro model, the historical load data, process parameters and adjacency relationship data are standardized by using the Z-score standardization formula through the input fusion layer to obtain the corresponding standard historical load data, standard process data and standard adjacency relationship data. The standard historical load data and standard process data are merged to obtain merged data, and time information is added to the merged data to obtain the data to be analyzed. Time information includes hours, minutes, weekdays, and non-working days; Step S102: Project the data to be analyzed into a query matrix Q, a key matrix K, and a value matrix V using a multi-head temporal attention mechanism; Attention scores are calculated based on the query matrix Q, the key matrix K, and the value matrix V to obtain multi-head attention feature data. Multi-head attention feature data is processed through a local temporal channel to obtain local temporal feature data; Multi-head attention feature data is processed through a global temporal channel to obtain global temporal feature data; The local and global time feature data are merged to obtain dual-channel time feature data; The dual-channel temporal feature data is processed by a two-layer fully connected network, and the ReLU activation function is used to calculate the two-layer fully connected data. The time feature data is obtained by performing a residual join between the two fully connected layers of data and the dual-channel time feature data. Step S103: Use the standard adjacency relationship data as a local spatial mask; Based on temporal feature data, the similarity between electroplating production lines is calculated using a global spatial attention head to obtain a global spatial mask; Based on local spatial masks, local spatial feature data is obtained by processing temporal feature data through local spatial attention heads. Based on the global spatial mask, the temporal feature data is processed by the global spatial attention head to obtain the global spatial feature data; Local spatial feature data and global spatial feature data are concatenated to obtain dual spatial feature data; The dual-space feature data is mapped using a linear transformation matrix, and a bias term is added to obtain linearly mapped data. By performing layer normalization on the linear mapping data, spatial feature data is obtained.

[0023] Step S104: Establish independent coding networks for the current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity and electroplating waveform parameters of the standard process data to obtain coding network data. Each coding network consists of two fully connected layers; The ReLU activation function is used to calculate the process feature vector by processing the encoded network data; The process feature vectors include current density feature vector, temperature feature vector, pH value feature vector, main salt concentration feature vector, additive concentration feature vector, stirring intensity feature vector, and electroplating waveform parameter feature vector; The current density feature vector, temperature feature vector, pH value feature vector, main salt concentration feature vector, additive concentration feature vector, stirring intensity feature vector, and electroplating waveform parameter feature vector are spliced ​​together to obtain process splicing feature data; The spliced ​​feature data is mapped using a linear transformation matrix, and a bias term is added to obtain the linear transformation process data; The linear transformation process data is subjected to layer normalization to obtain mapped process data. The process parameters are adjusted by the process adjustment weight matrix to obtain process feature data. And establish a mapping relationship between standard historical load data and process characteristic data; Step S105: The time feature data, spatial feature data and process feature data are spliced ​​together to obtain fused spliced ​​feature data; The fused and spliced ​​feature data is input into the gated weight generation network, and the time weight, spatial weight and process weight are obtained by calculation through a fully connected layer and a Softmax function. The weighted sum data is obtained by weighting and summing the time weight, spatial weight, and process weight with the corresponding time feature data, spatial feature data, and process feature data. The weighted summation data is residually concatenated with the time feature data, spatial feature data, and process feature data respectively to obtain time-enhanced feature data, spatial-enhanced feature data, and process-enhanced feature data; Temporal enhancement feature data, spatial enhancement feature data, and process enhancement feature data are concatenated and mapped to 64 dimensions through a linear transformation layer to obtain fused enhancement feature data; The weighted summation data is concatenated with the standard historical load data to obtain the first residual feature. At the same time, the weighted summation data is concatenated with the standard process parameters to obtain the second residual feature. The first residual feature and the second residual feature are concatenated to obtain the fused feature data; Step S106: The fused feature data is processed by a one-dimensional convolutional layer, expanding the dimension of the fused feature data from 64 to 128, resulting in an intermediate feature with a dimension of 10×96×128. The 128-dimensional intermediate features are mapped to 64-dimensional intermediate features through a fully connected layer, and the ReLU activation function is used to calculate the 64-dimensional intermediate features to obtain adjusted feature data with dimensions of 10×96×64. Step S107: Extract the feature data of the last 16 time steps from the adjusted feature data along the time dimension, and map the 64-dimensional features of each time point to 3 dimensions through a fully connected layer to obtain feature data with dimensions of 10×16×3. The load prediction value of the electroplating production line is obtained by using a linear activation function on the feature data with dimensions of 10×16×3. The load forecast for the electroplating production line includes the forecast of equipment operating power, the forecast of the number of parts processed, and the forecast of electroplating process duration.

[0024] In one embodiment, high-precision prediction of the electroplating production line load is achieved by fusing multi-source data, providing reliable input for subsequent optimization and control. First, historical load data (including equipment operating power, number of processed parts, and electroplating process duration), process parameters (including current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters), and adjacency relationship data (including electrical connection relationships, physical layout relationships, and process similarity relationships) are acquired. The historical load data, process parameters, and adjacency relationship data are standardized using the Z-score standardization formula to eliminate dimensional differences, resulting in standard historical load data, standard process data, and standard adjacency relationship data. After standardization, the standard historical load data and standard process data are merged, and time information, such as hours, minutes, working days, and non-working days, is added to form the data to be analyzed, which is used to introduce time period characteristics. Subsequently, the data to be analyzed is projected into a query matrix Q, a key matrix K, and a value matrix V using a multi-head time attention mechanism. Attention scores are calculated to obtain multi-head attention feature data. This multi-head attention feature data is processed through local and global time channels and then merged into dual-channel time feature data. The local time channel is used to capture short-term fluctuations, such as minute-level changes, while the global time channel is used to capture long-term trends, such as daily patterns. The dual-channel time feature data is then processed through a two-layer fully connected network, and the ReLU activation function is used to calculate the two-layer fully connected data. The two-layer fully connected data is then residually connected to the dual-channel time feature data to obtain the time feature data. This design aims to extract dynamic patterns in the time dimension from the steps themselves, avoiding the shortcomings of traditional time series models in modeling local and global correlations. From a core design perspective, this provides an accurate time series basis for prediction, ensuring the correlation between load changes and time factors. In spatial feature extraction, standard adjacency relationship data is used as a local spatial mask. For example, electrical connections directly affect load transfer. Simultaneously, based on temporal feature data, the similarity between production lines is calculated using a global spatial attention head. For instance, process similarity leads to load coordination, resulting in a global spatial mask. Using these masks, temporal feature data is processed through local and global spatial attention heads to generate local and global spatial feature data. These are then concatenated to obtain dual spatial feature data, which is further transformed linearly and normalized to obtain the final spatial feature data. This aims to model the spatial relationships between production lines, addressing the problem of traditional forecasting neglecting production line interactions. From a design perspective, it enhances the systematic nature of forecasting, reflecting the load distribution under multi-production line coordination. For process features, an independent encoding network is established for each process parameter. Each network consists of two fully connected layers, using the ReLU activation function to obtain the corresponding process feature vector, such as the current density feature vector. These vectors are concatenated, mapped through a linear transformation matrix, and a bias term is added. After layer normalization and adjustment by the process adjustment weight matrix, the process feature data is obtained. The process adjustment weight matrix is ​​a learnable parameter matrix during model training, used to adaptively calibrate the contribution of different process parameters to the final load. At the same time, a mapping relationship between standard historical load data and process feature data is established. For example, current density is directly related to operating power through Ohm's law. This encoding process ensures that the physical impact of process parameters is accurately quantified. It is designed to achieve efficient embedding of process parameters from the steps themselves, solving the prediction bias caused by traditional models that only use historical loads. From the core design, it improves the physical interpretability of the prediction and provides a connection for the quality constraints in the subsequent game model. In the feature fusion stage, temporal, spatial, and process feature data are concatenated and input into a gated weight generation network. This network consists of a fully connected layer and a Softmax function. The fully connected layer and Softmax function calculate time, spatial, and process weights. These weights are then weighted and summed with their corresponding temporal, spatial, and process feature data to obtain weighted sum data. This weighted sum data is then residually concatenated with each feature data to obtain enhanced feature data. This enhanced feature data is then concatenated and mapped through a linear transformation layer to form 64-dimensional fused enhanced feature data. Simultaneously, the weighted sum data is concatenated with standard historical load data and standard process parameter residuals to obtain first and second residual features. These are then concatenated to form the fused feature data. This mechanism adaptively balances the importance of different feature sources; for example, when process changes drastically, process weights are automatically increased, enhancing feature robustness from the step itself. From a design core perspective, it achieves dynamic integration of multi-source data, improving the model's adaptability to complex operating conditions. Next, a one-dimensional convolutional layer is used to perform deep feature extraction on the fused feature data, expanding its dimension from 64 to 128, resulting in an intermediate feature with dimensions of 10×96×128. The first dimension, 10, corresponds to the number of electroplating production lines, n; in this embodiment, n=10 is used for explanation. The second dimension, 96, is the historical time step, set to the shortest historical window required to fully cover a production batch with daily periodicity. The time interval of historical data, such as 15 minutes, is consistent with subsequent adjustment cycles. This time scale is mainly based on the common minimum time unit of electroplating production scheduling and power grid response, for example, 24-hour data with 15-minute intervals. The third dimension, 128, is the extended feature dimension, reflecting the convolutional layer's ability to enhance the model's nonlinear expression. The specific values ​​of these dimensions can be determined based on the actual number of production lines, data acquisition frequency, and model complexity. The complexity requirements were adjusted, with n=10, historical time step size 96, and feature dimension 128 to demonstrate the model's ability to process multi-production line, long-cycle time-series data. Subsequently, the intermediate feature was mapped to 64 dimensions through a fully connected layer and nonlinearly transformed using the ReLU activation function to obtain adjusted feature data with dimensions of 10×96×64. Finally, to generate load forecasts for a certain future period, feature data for the last 16 time steps were extracted along the time dimension. This setting corresponds to the forecast range for the next 4 hours, matching the common cycle of electroplating production planning and scheduling. The 64-dimensional feature at each time point was mapped to 3 dimensions through a fully connected layer to obtain feature data with dimensions of 10×16×3. The final load forecast value was output using a linear activation function, including equipment operating power forecast, number of processed parts forecast, and electroplating process duration forecast. The improved STGAN-Electro model is trained using historical data. Its training objective is to minimize the mean square error between the predicted load value and the actual load value, and an adaptive moment estimator optimizer can be used for parameter updates. This step significantly improves load forecasting accuracy through standardization, spatiotemporal attention, process coding, and dynamic fusion. For example, in process switching scenarios, the error is reduced by approximately 20% compared to traditional time-series models, and it can simultaneously output multi-dimensional load indicators. From the perspective of the top-level design of the solution, high-precision forecasting provides key input for the subsequent Stackelberg game model, ensuring that resource allocation is based on reliable forward-looking information, thereby supporting the synergistic goals of energy efficiency optimization, quality assurance, and load stability.

[0025] Step S2: Based on the load forecast, construct a Stackelberg game model with the electroplating power source as the leader and n electroplating production lines as followers. Step S2 includes steps S201, S202, and S203; Step S201: Construct a quality parameter vector based on the electroplating process quality parameters. Specifically, this includes normalizing the current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters of the electroplating process quality parameters to obtain normalized current density, normalized temperature, normalized pH value, normalized main salt concentration, normalized additive concentration, normalized stirring intensity, and normalized electroplating waveform parameters. A quality parameter vector is constructed based on normalized current density, normalized temperature, normalized pH value, normalized main salt concentration, normalized additive concentration, normalized stirring intensity, and normalized electroplating waveform parameters.

[0026] Step S202, the expression corresponding to the follower's utility function is, ; in, Indicates the first The effectiveness of an electroplating production line This represents a resource allocation vector, specifically including current, voltage, and power. Represents a mass parameter vector. The function representing quality satisfaction. This represents energy consumption cost, used to characterize the difference in actual operating efficiency of electroplating power supplies across different load ranges. Represents the cost of the resource allocation vector. , and Indicates the weighting coefficient; Step S203, the expression for the quality satisfaction function is, ; in, This indicates the quality satisfaction level of the electroplating production line. This represents the quality priority coefficient. This represents the tolerance variance for quality. Indicates time The mass parameter vector, Indicates time Target quality parameter vector; Energy consumption cost is calculated using a piecewise quadratic cost function, expressed as follows: ; in, The allowable operating power of the i-th electroplating production line is determined by the resource allocation vector, and k represents the load range within which the allowable operating power is located, determined by load forecast values. , and These represent the curvature coefficient, linear efficiency coefficient, and basic loss coefficient of the efficiency curve for this interval, respectively, obtained by curve fitting using historical operating efficiency data of the production line. This represents the electricity price coefficient.

[0027] Step S204, the expression corresponding to the leader's utility function is, ; in, R(A) represents the utility of the electroplating power source, R(A) represents the peak-to-average power ratio, and S(A) represents the balance of the resource allocation vector. , , and Indicates the weighting coefficient; Balance of resource allocation vector The calculation is performed using a quantization formula based on information entropy, and the expression is: ; ; in, Indicates the first The proportion of resource allocation vectors for each electroplating production line; The expression for the peak-to-average load ratio is: ; in, This indicates the peak-to-average load ratio. Let represent the modulus of the resource allocation vector of the i-th electroplating production line at time t, and represent its actual operating power.

[0028] Step S205: The solution of the Stackelberg game model satisfies the power balance constraint of the electroplating power source. The expression for the power balance constraint of the electroplating power supply is: ; in, Indicates the upper limit of the grid access power. Indicates other auxiliary loads, This indicates the power buffer provided by an adjustable load.

[0029] In one embodiment, this step follows the high-precision load prediction results obtained in step S1, aiming to construct a collaborative optimization decision-making framework that transforms prediction information into an executable optimal resource allocation strategy. Its core is to establish a Stackelberg game model with the electroplating power source as the leader and n electroplating production lines as followers. By simulating the master-slave interaction and dynamic game between the leader (global optimization) and followers (individual optimization), the optimal control scheme is solved under multiple objectives of ensuring process quality, improving energy efficiency, and smoothing load. First, the core decision variables in the model and their data sources and processing methods are clarified. The quality parameter vector is one of the core decision variables of the followers, directly representing their process state, derived from the real-time or planned process parameters of the production line, including current density, temperature, and p. H-value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters are scaled to the [0,1] interval through linear transformation to eliminate dimensional differences between parameters, i.e., normalized, and arranged in order to form a quality parameter vector. The resource allocation vector is the core decision output of the leader and the key input of the follower, corresponding to the controllable physical output of the power supply, including current, voltage, and operating power. Its candidate range is determined based on the power supply capacity and the load prediction value of step S1. Based on the defined decision variables, an optimization objective model for both sides of the game is constructed. For the follower (the i-th production line), its objective is to maximize its own utility function, which is expressed by the quality satisfaction function (using a Gaussian kernel-based form, with key parameters such as quality tolerance variance σ² set to 0.01 and quality priority coefficient). (Set according to production line requirements) Quantify the degree of process quality compliance, and use a segmented secondary energy consumption cost function (its coefficient) , and The efficiency curve was obtained by fitting historical power-efficiency curves segmented by load range. The load range k was divided according to the typical efficiency characteristics of the electroplating power supply, specifically into three ranges: light load, medium load, and heavy load. The corresponding power ranges are (0%, 30%), (30%, 70%), and (70%, 100%) of the power supply's rated power, respectively. This division is based on statistical analysis of historical operating data of the power supply. Significant inflection points on the efficiency curve are typically observed around 30% and 70% of the rated power. The aim is to ensure that the efficiency-power relationship within each range is... It can be more accurately fitted by a quadratic curve; when solving the model, the value of k is determined according to the power range to which the predicted power of the equipment operation obtained in step S1 belongs) to finely characterize the efficiency differences and electricity costs under different loads; for the leader (electroplating power supply), its goal is to maximize the global utility function under the power balance constraint, which coordinates the total mass, total energy consumption cost and two key indicators: load peak-to-average ratio and resource allocation balance. The load peak-to-average ratio is used to smooth the load curve, and the resource allocation balance is based on the information entropy quantification formula to promote fair resource allocation; To achieve the above optimizations and ensure process feasibility, a system was established. and The constraints include, specifically, datum mapping constraints, adjustment range constraints, and absolute boundary constraints; The expression for the datum mapping constraint is: ; This represents the baseline value of the quality parameter vector for the i-th electroplating production line under ideal or standard operating conditions. This represents a function determined by the physicochemical laws of the electroplating process, such as Ohm's law, establishing a system based on the allocation of given resources. Mapping relationship to standard quality parameters; The expression for adjusting the range constraint is: ; This represents the adjustment vector of the quality parameter relative to the reference value. Represent a scalar function, determined by resource allocation. The decision defines the maximum allowable adjustment range. This represents the adjustment range function; The expression for absolute boundary constraints is, ; The process feasible region represents the vector of all quality parameters for the i-th electroplating production line. This is a set defined by process knowledge, specifying that each quality parameter must fall within the physical, chemical, and safety ranges specified by the current resource allocation. Irrelevant; The benchmark mapping function, adjustment amplitude function, and feasible region are determined based on the fundamental physicochemical laws of the electroplating process and Ohm's law. The specific determination process includes using a resource allocation vector... Distribution current in and mass parameter vector Current density in For example, according to Ohm's law and the definition of current density Establish a baseline mapping function The expression is, ,in, Given the known cathode plating area; Based on process experience, the allowable current density relative adjustment margin is set as follows: , where 0 < <1, and combined with the absolute safe range of current density Determine the adjustment range function for, feasible region Directly defined as the set of absolutely safe ranges for current density, specifically... ; Furthermore, the model solution must satisfy power balance constraints, ensuring that the total power demand of all production lines does not exceed the total power supply capacity determined by the grid access limit, other auxiliary loads, and adjustable load buffers. The purpose of this step is to decouple the complex centralized problem into a two-layer optimization structure through a master-slave game framework, reducing the solution complexity and precisely quantifying the factors of process quality, nonlinear efficiency, and stability into optimizable objectives. Simultaneously, by introducing… The physical process constraints between the two processes ensure that the optimization model closely matches the actual laws of the electroplating process. From the core design perspective, this model is the central hub for achieving a closed loop of accurate prediction and intelligent decision-making. It uses predictive information to collaboratively optimize multiple objectives such as quality, energy efficiency, load smoothing, and fair allocation, generating direct execution instructions for step S3. Its technical effect is that the resulting resource allocation scheme can dynamically respond to electricity prices and proactively avoid quality risks and load peaks, and is expected to achieve multiple benefits such as improving overall energy efficiency, stabilizing electroplating quality, and reducing the peak-to-average load ratio.

[0030] Step S3: Solve the Stackelberg game model to obtain the optimal resource allocation vector, and adjust the output power of the electroplating power supply based on the optimal resource allocation vector.

[0031] In one embodiment, this step aims to apply in real time the solution and optimization results of the Stackelberg game model constructed in step S2, thereby completing the intelligent adjustment closed loop from prediction-decision to execution. First, the game model is solved using backward induction. Given the initial resource allocation scheme of the leader (electroplating power source), the local optimal quality parameter vector of each follower (electroplating production line) is solved sequentially. This process involves, given a resource allocation vector Next, based on the Ai established in step S2 and Inter-constraints, specifically datum mapping constraints, adjustment range constraints, and absolute boundary constraints, call the quality satisfaction function in step S203. The solution is then performed; subsequently, the optimal responses of each follower are calculated. Feedback is given to the leader's utility function, and the solution is found to satisfy the power balance constraint, thereby maximizing the global utility function. Maximize the globally optimal resource allocation vector =[ , ,..., The model solution is an iterative calculation process. A convergence threshold of 0.01 is set, based on the real-time requirement of a 15-minute adjustment cycle, determined through prior simulation testing. At this threshold, the iterative solution of the game model can converge within dozens of iterations, requiring significantly less computation time than the adjustment cycle. Simultaneously, this threshold means that the calculation terminates when the global utility improvement from optimization is less than 1%. This precision meets the decision-making requirements for electroplating process control. When the change in the global utility function value obtained from two consecutive iterations is less than this threshold, convergence is determined, and the optimal solution is output. Key input data relied upon for this solution process include predicted equipment operating power values ​​used to determine the load range k, and the coefficients of the piecewise quadratic cost function. , and and in power balance constraints Apart from the predicted values, other parameters such as the cost function coefficients and the grid power limit are preset parameters, all directly derived from the prediction results of step S1 and the preset parameters. The preset system parameters include, but are not limited to, the rated parameters of the power source, the grid contract limits, the efficiency curve coefficients of the historical fitting, and the process quality target parameters, thus obtaining the optimal resource allocation vector. Subsequently, the central controller decodes the current, voltage, and power setting instructions corresponding to each production line and sends them to the execution unit of the electroplating power supply through the communication network, thereby adjusting the actual output power of each production line in real time. This step uses a basic adjustment cycle of 15 minutes, which is consistent with the time resolution of the prediction data. The setting of this cycle continues the aforementioned basis, ensuring that the closed loop from data acquisition, prediction, optimization to control execution runs on a unified time base, which can match the production cycle and meet the time budget of rolling calculation, thereby realizing dynamic rolling optimization and execution. The purpose of this step is to ensure the effective acquisition of the game equilibrium solution within an acceptable time by employing backward induction and common convergence criteria in engineering, thus making the theoretical model computationally feasible for online application. At its core, this step is the final and crucial link in achieving a fully automated closed-loop prediction-optimization-execution process. It directly and accurately transforms the predicted information and optimization decisions generated in previous steps into control actions for physical power supply equipment. Based on forward-looking information, it proactively and collaboratively adjusts power allocation. The technical effect is that through closed-loop execution of the optimal resource allocation vector, it responds to electricity price signals in real time and maintains the load stably within the high-efficiency operating range (such as the medium-load zone), while strictly ensuring the quality of the electroplating process. It is expected to achieve the overall energy efficiency improvement and the collaborative optimization goal of reducing the peak-to-average load ratio.

[0032] Example 2, refer to Figure 2 In another embodiment of the present invention, which differs from the first embodiment, an intelligent adjustment system for electroplating power supply based on dynamic load prediction is provided, including a processing module, a construction module and an adjustment module; The processing module, for an electroplating power supply that powers n electroplating production lines, inputs historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model and outputs the load prediction value of the electroplating production lines. The module constructs a Stackelberg game model based on load forecasts, with the electroplating power source as the leader and n electroplating production lines as followers. The adjustment module solves the Stackelberg game model to obtain the optimal resource allocation vector, and adjusts the output power of the electroplating power supply based on the optimal resource allocation vector.

[0033] In one embodiment, firstly, multi-source data, including historical load data, process parameters, and adjacency relationship data, is input into an improved STGAN-Electro model for processing to obtain load prediction values ​​for the electroplating production lines, achieving high-precision prediction of the future load of n electroplating production lines. Based on these load prediction values, a Stackelberg game model is constructed with the electroplating power supply as the leader and each production line as the follower. The utility functions of both the leader and followers embed multiple constraints and objectives related to process quality satisfaction, segmented energy consumption costs, and equilibrium. Solving this game model yields the optimal resource allocation vector under the constraints of process quality and power balance, and the electroplating power supply is adjusted in real time accordingly. The system significantly improves the accuracy of dynamic load prediction under complex electroplating conditions, providing a reliable basis for optimized control. Through game theory equilibrium, it achieves optimal resource allocation with limited power capacity, while ensuring the electroplating process quality and overall energy efficiency of each production line. It effectively smooths the total load curve, reduces the peak-to-average power ratio, and enhances operational stability and economy. Addressing the current problems of single prediction dimensions, fragmented optimization objectives, and insufficient coordination in electroplating workshop load control, the system combines multi-source data fusion prediction with multi-layer game theory collaborative optimization, linking accurate prediction with intelligent decision control. This solves the industry pain point of difficulty in coordinating the optimization of energy efficiency, quality, and load stability in electroplating production.

[0034] This invention addresses the problem of existing load forecasting models neglecting the spatiotemporal correlation between process parameters and production lines by improving the STGAN-Electro model. By integrating historical load, process parameter, and adjacency relationship data and introducing a dual-channel spatiotemporal attention mechanism, it significantly improves forecasting accuracy under complex operating conditions. Secondly, to address the difficulty of balancing quality and energy efficiency in traditional centralized optimization, a Stackelberg game model with the power supply as the leader and the production line as the follower is constructed. Through constraints of a quality satisfaction function and a benchmark mapping, process quality requirements are embedded in the resource allocation process, achieving synergistic optimization of quality and energy consumption. In terms of target design, by introducing a piecewise quadratic energy consumption cost function and peak-to-average power ratio and balance indexes, the power supply load efficiency characteristics are finely characterized, promoting load stability. Ultimately, this achieves a comprehensive technical effect of improving overall energy efficiency, ensuring process quality, and enhancing operational stability.

[0035] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0036] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for intelligent adjustment of electroplating power supply based on dynamic load prediction, characterized in that, Includes the following steps, Step S1: For an electroplating power supply that powers n electroplating production lines, input historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model, and output the load prediction value of the electroplating production line. Step S2: Based on the load forecast, construct a Stackelberg game model with the electroplating power source as the leader and n electroplating production lines as followers. Step S3: Solve the Stackelberg game model to obtain the optimal resource allocation vector, and adjust the output power of the electroplating power supply based on the optimal resource allocation vector.

2. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 1, characterized in that, Acquire historical load data, process parameters, and adjacency relationship data; Historical load data includes the equipment operating power, number of parts processed, and electroplating process duration of the electroplating production line; Process parameters include current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters; Adjacency data includes electrical connection data, physical layout data, and process similarity data.

3. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 1, characterized in that, Step S1: For an electroplating power supply that powers n electroplating production lines, input historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model, and output the load prediction value of the electroplating production line. Step S1 includes steps S101, S102, S103, S104, S105, S106 and S107. Step S101: Based on the improved STGAN-Electro model, the historical load data, process parameters and adjacency relationship data are standardized by using the Z-score standardization formula through the input fusion layer to obtain the corresponding standard historical load data, standard process data and standard adjacency relationship data. The standard historical load data and standard process data are merged to obtain merged data, and time information is added to the merged data to obtain the data to be analyzed. The time information includes hours, minutes, weekdays, and non-working days; Step S102: Project the data to be analyzed into a query matrix Q, a key matrix K, and a value matrix V using a multi-head temporal attention mechanism; Attention scores are calculated based on the query matrix Q, the key matrix K, and the value matrix V to obtain multi-head attention feature data. Multi-head attention feature data is processed through a local temporal channel to obtain local temporal feature data; Multi-head attention feature data is processed through a global temporal channel to obtain global temporal feature data; The local and global time feature data are merged to obtain dual-channel time feature data; The dual-channel temporal feature data is processed by a two-layer fully connected network, and the ReLU activation function is used to calculate the two-layer fully connected data. The time feature data is obtained by performing a residual join between the two fully connected layers of data and the dual-channel time feature data. Step S103: Use the standard adjacency relationship data as a local spatial mask; Based on temporal feature data, the similarity between electroplating production lines is calculated using a global spatial attention head to obtain a global spatial mask; Based on local spatial masks, local spatial feature data is obtained by processing temporal feature data through local spatial attention heads. Based on the global spatial mask, the temporal feature data is processed by the global spatial attention head to obtain the global spatial feature data; Local spatial feature data and global spatial feature data are concatenated to obtain dual spatial feature data; The dual-space feature data is mapped using a linear transformation matrix, and a bias term is added to obtain linearly mapped data. The linear mapping data is subjected to layer normalization to obtain spatial feature data.

4. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 3, characterized in that, Step S104: Establish independent coding networks for the current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity and electroplating waveform parameters of the standard process data to obtain coding network data. Each coding network consists of two fully connected layers; The ReLU activation function is used to calculate the process feature vector by processing the encoded network data; The process feature vectors include current density feature vector, temperature feature vector, pH value feature vector, main salt concentration feature vector, additive concentration feature vector, stirring intensity feature vector, and electroplating waveform parameter feature vector; The current density feature vector, temperature feature vector, pH value feature vector, main salt concentration feature vector, additive concentration feature vector, stirring intensity feature vector, and electroplating waveform parameter feature vector are spliced ​​together to obtain process splicing feature data; The spliced ​​feature data is mapped using a linear transformation matrix, and a bias term is added to obtain the linear transformation process data; The linear transformation process data is subjected to layer normalization to obtain mapped process data. The process parameters are adjusted by the process adjustment weight matrix to obtain process feature data. And establish a mapping relationship between standard historical load data and process characteristic data; Step S105: The time feature data, spatial feature data and process feature data are spliced ​​together to obtain fused spliced ​​feature data; The fused and spliced ​​feature data is input into the gated weight generation network, and the time weight, spatial weight and process weight are obtained by calculation through a fully connected layer and a Softmax function. The weighted sum data is obtained by weighting and summing the time weight, spatial weight, and process weight with the corresponding time feature data, spatial feature data, and process feature data. The weighted summation data is residually concatenated with the time feature data, spatial feature data, and process feature data respectively to obtain time-enhanced feature data, spatial-enhanced feature data, and process-enhanced feature data; Temporal enhancement feature data, spatial enhancement feature data, and process enhancement feature data are concatenated and mapped to 64 dimensions through a linear transformation layer to obtain fused enhancement feature data; The weighted summation data is concatenated with the standard historical load data to obtain the first residual feature. At the same time, the weighted summation data is concatenated with the standard process parameters to obtain the second residual feature. The first residual feature and the second residual feature are concatenated to obtain fused feature data; Step S106: The fused feature data is processed by a one-dimensional convolutional layer, expanding the dimension of the fused feature data from 64 to 128, resulting in an intermediate feature with a dimension of 10×96×128. The 128-dimensional intermediate features are mapped to 64-dimensional intermediate features through a fully connected layer, and the ReLU activation function is used to calculate the 64-dimensional intermediate features to obtain adjusted feature data with dimensions of 10×96×64. Step S107: Extract the feature data of the last 16 time steps from the adjusted feature data along the time dimension, and map the 64-dimensional features of each time point to 3 dimensions through a fully connected layer to obtain feature data with dimensions of 10×16×3. The load prediction value of the electroplating production line is obtained by using a linear activation function to calculate the feature data with dimensions of 10×16×3. The load forecast values ​​for the electroplating production line include the equipment operating power forecast, the number of parts processed forecast, and the electroplating process duration forecast.

5. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 4, characterized in that, Step S2: Based on the load forecast, construct a Stackelberg game model with the electroplating power source as the leader and n electroplating production lines as followers. Step S2 includes steps S201, S202, and S203; Step S201: Construct a quality parameter vector based on the electroplating process quality parameters. Specifically, this includes normalizing the current density, temperature, pH value, main salt concentration, additive concentration, stirring intensity, and electroplating waveform parameters of the electroplating process quality parameters to obtain normalized current density, normalized temperature, normalized pH value, normalized main salt concentration, normalized additive concentration, normalized stirring intensity, and normalized electroplating waveform parameters. A quality parameter vector is constructed based on normalized current density, normalized temperature, normalized pH value, normalized main salt concentration, normalized additive concentration, normalized stirring intensity, and normalized electroplating waveform parameters.

6. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 5, characterized in that, Step S202, the expression corresponding to the utility function of the follower is, ; in, Indicates the first The effectiveness of an electroplating production line This represents a resource allocation vector, specifically including current, voltage, and power. Represents a mass parameter vector. The function representing quality satisfaction. This represents energy consumption cost, used to characterize the difference in actual operating efficiency of electroplating power supplies across different load ranges. Represents the cost of the resource allocation vector. , and Indicates the weighting coefficient; Step S203, the expression for the quality satisfaction function is: ; in, This indicates the quality satisfaction level of the electroplating production line. This represents the quality priority coefficient. This represents the tolerance variance for quality. Indicates time The mass parameter vector, Indicates time Target quality parameter vector; Energy consumption cost is calculated using a piecewise quadratic cost function, expressed as follows: ; in, The allowable operating power of the i-th electroplating production line is determined by the resource allocation vector, and k represents the load range within which the allowable operating power is located, determined by load forecast values. , and These represent the curvature coefficient, linear efficiency coefficient, and basic loss coefficient of the efficiency curve for this interval, respectively, obtained by curve fitting using historical operating efficiency data of the production line. This represents the electricity price coefficient.

7. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 6, characterized in that, Step S204, the expression corresponding to the leader's utility function is, ; in, R(A) represents the utility of the electroplating power source, R(A) represents the peak-to-average power ratio, and S(A) represents the balance of the resource allocation vector. , , and Indicates the weighting coefficient; The balance of the resource allocation vector The calculation is performed using a quantization formula based on information entropy, and the expression is: ； ； in, Indicates the first The proportion of resource allocation vectors for each electroplating production line; The expression for the peak-to-average load ratio is as follows. ; in, This indicates the peak-to-average load ratio. Let represent the modulus of the resource allocation vector of the i-th electroplating production line at time t, and represent its actual operating power.

8. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 7, characterized in that, Step S205: The solution of the Stackelberg game model satisfies the power balance constraint of the electroplating power source. The expression for the power balance constraint of the electroplating power supply is: ; in, Indicates the upper limit of the grid access power. Indicates other auxiliary loads, This indicates the power buffer provided by an adjustable load.

9. The intelligent adjustment method for electroplating power supply based on dynamic load prediction as described in claim 8, characterized in that, Step S3: Solve the Stackelberg game model to obtain the optimal resource allocation vector, and adjust the output power of the electroplating power supply based on the optimal resource allocation vector.

10. An intelligent regulation system for electroplating power supply based on dynamic load prediction, applied to the intelligent regulation method for electroplating power supply based on dynamic load prediction as described in any one of claims 1-9, characterized in that, It includes a processing module, a building module, and a regulation module; The processing module, for an electroplating power supply that powers n electroplating production lines, inputs historical load data, process parameters and adjacency relationship data into the improved STGAN-Electro model and outputs the load prediction value of the electroplating production lines. The module constructs a Stackelberg game model based on load forecasts, with the electroplating power source as the leader and n electroplating production lines as followers. The adjustment module solves the Stackelberg game model to obtain the optimal resource allocation vector, and adjusts the output power of the electroplating power supply based on the optimal resource allocation vector.

Images