Laser printer control task management method and system based on deep learning

Abstract

The application relates to the technical field of laser printers and discloses a laser printer control task management method and system based on deep learning, which comprises the following steps: collecting pretreatment data; constructing a deep learning model based on mixed attention; extracting local features in task, device and environment data by using a separable convolution structure of a convolutional neural network; capturing time sequence dependency in device running states and task execution processes by using a long short-term memory network; mining cross-dimension association weights between different feature dimensions based on a self-attention mechanism; optimizing hyperparameters of the mixed attention deep learning model by using an improved IWOA whale optimization algorithm to obtain a target mixed attention deep learning model; and inputting the pretreatment data into the target mixed attention deep learning model to output task scheduling priorities, optimal control parameters and fault risk prediction results. The method can optimize a task queue, improve printing efficiency, adjust control parameters and reduce device downtime.

Description

Technical Field

[0001] This invention relates to the field of laser printer technology, and in particular to a laser printer control task management method and system based on deep learning. Background Technology

[0002] With the rapid development of office automation and intelligent industrial production, laser printers, as core output devices, are widely used in various industries. Their task processing efficiency and equipment stability directly affect workflows and production efficiency. Currently, laser printer task management mostly adopts traditional manual scheduling and fixed parameter control modes, which have many shortcomings. Traditional task scheduling relies solely on user-preset priorities and does not dynamically adjust based on equipment operating status and environmental factors, easily leading to problems such as task congestion and equipment load imbalance. Equipment control parameters are mostly fixed values, unable to adaptively optimize according to printing task type and environmental changes, resulting in unstable print quality and high energy consumption. At the same time, existing fault prediction methods mostly rely on manual inspections or single sensor data, making it difficult to predict potential faults in advance and easily causing equipment downtime. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by designing a laser printer control task management method and system based on deep learning.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned deep learning-based laser printer control task management method: Data on pending print jobs, equipment operating status, and environmental impact from laser printers are collected. The collected data is then cleaned, outliers are removed, and features are standardized to obtain preprocessed data. We construct a hybrid attention deep learning model based on CNN-BiLSTM-Transformer. We use the separable convolutional structure of CNN convolutional neural network to extract local features from task, device and environment data. We use BiLSTM long short-term memory network to capture the temporal dependencies of device operation status and task execution process. We use Transformer self-attention mechanism to mine cross-dimensional correlation weights between different feature dimensions. The hyperparameters of the hybrid attention deep learning model are optimized using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model. The preprocessed data is input into the target hybrid attention deep learning model, which outputs task scheduling priority, optimal control parameters, and fault risk prediction results.

[0005] Furthermore, in the aforementioned deep learning-based laser printer control task management method, the process of collecting unprocessed printing task data, equipment operating status data, and environmental impact data from the laser printer, and then sequentially performing data cleaning, outlier removal, and feature standardization to obtain preprocessed data, includes: Collect data on pending print jobs, equipment operating status, and environmental impact from laser printers; The collected data is processed for missing and duplicate values. The mean and standard deviation of the data are calculated using the 3σ principle. Data that exceeds the threshold range are identified as outliers, and the cleaned data is obtained. The continuous features in the cleaned data are standardized using the Z-score standardization method to obtain preprocessed data.

[0006] Furthermore, in the aforementioned deep learning-based laser printer control task management method, the construction of a hybrid attention deep learning model based on CNN-BiLSTM-Transformer utilizes the separable convolutional structure of the CNN convolutional neural network to extract local features from task, device, and environmental data. It also uses a BiLSTM long short-term memory network to capture the temporal dependencies between device operating states and task execution processes, and mines cross-dimensional correlation weights between different feature dimensions based on the Transformer self-attention mechanism. This includes: The input layer of a CNN (Convolutional Neural Network) receives preprocessed data. It uses a single two-dimensional separable convolutional layer to split the input features into depth and spatial directions for convolution. The ReLU activation function is used to extract local features from the preprocessed data. The output of the convolutional layer is followed by a max-pooling layer to preserve key information about local features. Output local features.

[0007] Furthermore, in the aforementioned deep learning-based laser printer control task management method, the construction of a hybrid attention deep learning model based on CNN-BiLSTM-Transformer utilizes the separable convolutional structure of the CNN convolutional neural network to extract local features from task, device, and environmental data. It also uses a BiLSTM long short-term memory network to capture the temporal dependencies between device operating states and task execution processes, and mines cross-dimensional correlation weights between different feature dimensions based on the Transformer self-attention mechanism. This includes: The BiLSTM long short-term memory network receives local features from the output of the CNN convolutional neural network and sets up 2 BiLSTM layers; The forward LSTM is responsible for capturing the temporal dependencies from the past to the present, while the backward LSTM is responsible for capturing the temporal dependencies from the present to the future. The output features of the two are concatenated to obtain the temporal features.

[0008] Furthermore, in the aforementioned deep learning-based laser printer control task management method, the construction of a hybrid attention deep learning model based on CNN-BiLSTM-Transformer utilizes the separable convolutional structure of the CNN convolutional neural network to extract local features from task, device, and environmental data. It also uses a BiLSTM long short-term memory network to capture the temporal dependencies between device operating states and task execution processes, and mines cross-dimensional correlation weights between different feature dimensions based on the Transformer self-attention mechanism. This includes: The Transformer self-attention mechanism adopts a multi-head self-attention mechanism with 8 heads and 32 feature dimensions for each head. It quantifies the correlation features between different feature dimensions by calculating the attention weights between feature vectors. The encoder is set to a single Transformer encoder, which includes a multi-head self-attention layer and an FFN feedforward neural network. The hidden layer dimension of the feedforward neural network is set to 512, and the GELU activation function is used to perform non-linear transformation on the features output by the self-attention layer. The associated features, local features, and temporal features are weighted and fused to obtain a comprehensive feature vector.

[0009] Furthermore, in the aforementioned deep learning-based laser printer control task management method, the step of optimizing the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model includes: The fixed weights in the traditional whale optimization algorithm are replaced with adaptive weights that are dynamically adjusted with the number of iterations. After the whale's position is updated, a Logistic chaotic mapping is introduced to perturb the position, resulting in the improved IWOA whale optimization algorithm.

[0010] Furthermore, in the aforementioned deep learning-based laser printer control task management method, the step of optimizing the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model includes: The whale population size is set to 30, each whale individual corresponds to a set of hyperparameter combinations, and an initial population that meets the search range of each hyperparameter is randomly generated. At the same time, the maximum number of iterations T=300 and the fitness function threshold=0.001 are set. The fitness function is calculated using the test set error of the hybrid attention deep learning model as the fitness function. Based on the improved IWOA algorithm's search, encirclement, and attack mechanisms, combined with adaptive weights and chaotic perturbations, the position of each individual whale is updated, and a new fitness value is calculated after each update, retaining the individual with the best fitness value. If the current iteration count reaches the maximum iteration count, stop the optimization and output the current optimal hyperparameter combination; substitute the optimal hyperparameter combination obtained by optimization into the hybrid attention deep learning model to obtain the target hybrid attention deep learning model.

[0011] Furthermore, in the deep learning-based laser printer control task management system, the laser printer control task management system includes the following modules: The print data acquisition module is used to collect print task data, equipment operating status data and environmental impact data of the laser printer. The collected data is then cleaned, outlier removed and feature standardized to obtain preprocessed data. The hybrid model building module is used to build a hybrid attention deep learning model based on CNN-BiLSTM-Transformer. It uses the separable convolutional structure of CNN convolutional neural network to extract local features from task, device and environment data, captures the temporal dependencies of device operation status and task execution process through BiLSTM long short-term memory network, and mines cross-dimensional correlation weights between different feature dimensions based on Transformer self-attention mechanism. The hybrid model optimization module is used to optimize the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model. The task control and management module is used to input the preprocessed data into the target hybrid attention deep learning model and output the task scheduling priority, optimal control parameters and fault risk prediction results.

[0012] Furthermore, in the deep learning-based laser printer control task management system, the hybrid model optimization module includes the following sub-modules: An improved submodule is used to replace the fixed weights in the traditional whale optimization algorithm with adaptive weights that are dynamically adjusted with the number of iterations. After the whale's position is updated, a Logistic chaotic mapping is introduced to perturb the position, resulting in the improved IWOA whale optimization algorithm.

[0013] Furthermore, in the deep learning-based laser printer control task management system, the hybrid model optimization module includes the following sub-modules: The configuration submodule is used to set the whale population size to 30, with each individual whale corresponding to a set of hyperparameter combinations, randomly generating an initial population that meets the search range of each hyperparameter, and setting the maximum number of iterations T=300 and the fitness function threshold=0.001. The computation submodule is used to calculate the fitness function using the test set error of the hybrid attention deep learning model as the fitness function. The update submodule is used to update the position of each whale individual based on the search, encirclement, and attack mechanism of the improved IWOA algorithm, combined with adaptive weights and chaotic perturbations. After each update, a new fitness value is calculated, and the individual with the best fitness value is retained. The submodule is obtained to stop the optimization if the current iteration count reaches the maximum iteration count, and outputs the current optimal hyperparameter combination; the optimal hyperparameter combination obtained by optimization is substituted into the hybrid attention deep learning model to obtain the target hybrid attention deep learning model.

[0014] Its beneficial effects lie in the fact that by collecting task, equipment, and environmental data from multiple dimensions and performing refined preprocessing, invalid data is effectively eliminated and dimensional differences are removed, providing high-quality input for the model and improving the accuracy of subsequent analysis. The constructed CNN-BiLSTM-Transformer hybrid model integrates local feature extraction, temporal dependency capture, and cross-dimensional correlation mining capabilities. Compared with a single model, it can more comprehensively capture the inherent correlation of data, improving the accuracy of task scheduling, parameter control, and fault prediction. The improved IWOA algorithm is used to optimize hyperparameters. Through adaptive weights and chaotic perturbation strategies, the problems of slow convergence and local optima in traditional algorithms are solved, ensuring that the model reaches optimal performance. Referring to the application experience of relevant optimization algorithms, the model's optimization accuracy and efficiency can be significantly improved. It can optimize the task queue, improve printing efficiency, adaptively adjust control parameters, reduce energy consumption, and predict fault risks in advance, reducing equipment downtime and improving the overall intelligent management level and operational stability of laser printers. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of the laser printer control task management method based on deep learning in this invention. Figure 2 This is a schematic diagram of a second embodiment of the laser printer control task management method based on deep learning in this invention. Figure 3 This is a schematic diagram of the first embodiment of the laser printer control task management system based on deep learning in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a deep learning-based laser printer control task management method includes the following steps: Step 101: Collect the print job data, equipment operation status data and environmental impact data of the laser printer, and perform cleaning, outlier removal and feature standardization on the collected data in sequence to obtain preprocessed data. Specifically, in this embodiment, the data of the laser printer's print jobs to be processed, the equipment's operating status, and the environmental impact are collected; the collected data is processed for missing and duplicate values, and the mean and standard deviation of the data are calculated using the 3σ principle. Data exceeding the threshold range are identified as outliers, thus obtaining cleaned data; the continuous features in the cleaned data are standardized using the Z-score standardization method to obtain preprocessed data.

[0020] Data for pending print jobs is collected through the laser printer's task queue interface. Core data includes task ID, task type, printing parameters, print resolution, paper size, number of copies, color mode (black and white / color), print speed requirement, task submission time, task priority, user presets, task data volume, file size, etc. The collection frequency is synchronized with the task submission frequency. The corresponding data is collected immediately upon receiving a new task to ensure that no task is missed.

[0021] Equipment operating status data is collected through the printer's built-in sensors, including temperature, humidity, speed, toner, and drum sensors, as well as the equipment control system interface. Core data includes drum temperature, laser head power, drum wear level, remaining toner, remaining paper, actual printing speed, motor speed, equipment runtime, cumulative number of pages printed, voltage / current parameters of each component, and fault history records. If any, the collection frequency is set to once per second to ensure real-time capture of changes in equipment operating status.

[0022] Environmental impact data is collected through environmental sensors deployed around the printer. Core data includes ambient temperature, ambient humidity, ambient dust concentration, power supply voltage stability and fluctuation range, ambient light intensity, etc. The collection frequency is set to once every 5 seconds to avoid interference from environmental factors and reduce redundant data.

[0023] For missing value handling, non-critical missing values ​​in task data, such as task notes, are filled with no notes; for critical missing values, such as print resolution and toner remaining, the mean of the same type of data is used. For continuous data, such as temperature, speed, or mode, and for discrete data, such as task type, the missing value is filled. If the missing value rate exceeds 30%, the data is removed to avoid affecting the overall data quality.

[0024] Duplicate data handling involves double verification using task ID and acquisition time to eliminate duplicate data. For example, duplicate reports due to sensor malfunctions retain the valid data from the first acquisition. Format standardization involves converting time data in different formats to YYYY-MM-DD HH:MM:SS format; encoding discrete data, such as task type (text printing = 0, image printing = 1, batch printing = 2); and converting data with inconsistent units, such as voltage and current, to standard units.

[0025] The 3σ principle is employed, applicable to continuous data such as roller temperature, toner remaining quantity, and box plot analysis, and also suitable for discrete data such as a combination of print count and failure count. For the 3σ principle, the mean μ and standard deviation σ of the data are calculated, and data exceeding the range [μ-3σ, μ+3σ] are identified as outliers. For box plot analysis, data exceeding 1.5 times the interquartile range are identified as outliers. Outlier handling involves directly removing invalid data caused by sensor malfunctions; for outliers caused by random fluctuations, linear interpolation of adjacent data is used to replace them, ensuring the continuity of the data sequence. Z-score standardization is used to standardize all preprocessed continuous features.

[0026] Step 102: Construct a hybrid attention deep learning model based on CNN-BiLSTM-Transformer. Utilize the separable convolutional structure of the CNN convolutional neural network to extract local features from task, device, and environmental data. Use the BiLSTM long short-term memory network to capture the temporal dependencies in device operating status and task execution. Based on the Transformer self-attention mechanism, mine cross-dimensional correlation weights between different feature dimensions. Specifically, in this embodiment, the input layer of the CNN convolutional neural network receives preprocessed data, and through a two-dimensional separable convolutional layer, the input features are split into depth and spatial directions for convolution. The ReLU activation function is used to extract local features from the preprocessed data. After the output of the convolutional layer, a max pooling layer is connected to retain the key information of the local features and output the local features.

[0027] The BiLSTM (Long Short-Term Memory) network receives local features from the output of the CNN (Convolutional Neural Network) and sets up two BiLSTM layers: the forward LSTM is responsible for capturing the temporal dependencies from the past to the present, and the backward LSTM is responsible for capturing the temporal dependencies from the present to the future. The output features of the two are concatenated to obtain the temporal features.

[0028] The Transformer self-attention mechanism employs a multi-head self-attention mechanism with 8 heads, each with a feature dimension of 32. It quantifies the correlation features between different feature dimensions by calculating the attention weights between feature vectors. The encoder consists of a single Transformer encoder layer containing a multi-head self-attention layer and an FFN feedforward neural network. The hidden layer dimension of the feedforward neural network is set to 512, and the GELU activation function is used to perform a non-linear transformation on the features output by the self-attention layer. The correlated features, local features, and temporal features are weighted and fused to obtain a comprehensive feature vector.

[0029] The CNN module employs a separable convolutional structure to reduce model parameters, improve computational efficiency, and accurately extract key local features from the data. Specifically, the convolutional layer is set up as follows: the input layer receives a preprocessed data matrix with dimensions [number of samples, number of features, time step]. First, a single 2D separable convolutional layer splits the input features into depth and spatial directions for separate convolutions. The kernel size is set to 3×3, the stride to 1, and the padding method is set to "same" to ensure the feature map size after convolution matches the input. The ReLU activation function is used to extract local features from the data, such as the local correlation between task type and printing parameters, or the local variation patterns of operating parameters of a device component. The pooling layer is set up as follows: after the convolutional layer output, a max-pooling layer is connected. The pooling kernel size is 2×2, and the stride is 2, used to reduce feature dimensionality, reduce redundant information, and retain key information of local features. After the pooling layer output, a batch normalization (BN) layer is applied to accelerate model training convergence and avoid the vanishing gradient problem. The CNN module ultimately outputs a local feature matrix with dimensions [number of samples, number of local features, time step], which serves as the input to the BiLSTM module.

[0030] The BiLSTM module utilizes the bidirectional memory property of BiLSTM to capture the temporal dependencies between device operating states and task execution. It receives the local feature matrix output from the CNN module and uses two BiLSTM layers, each with 64 hidden units and a dropout rate of 0.2 to avoid overfitting. It employs a sequence-return mode, meaning each layer's output includes feature information from all time steps, rather than just the last time step. The forward LSTM captures temporal dependencies from the past to the present, such as the impact of historical device operating states on the current state. The backward LSTM captures temporal dependencies from the present to the future, such as the impact of the current task execution on subsequent device states. The output features of both are concatenated to obtain a temporal feature matrix with dimensions [number of samples, time step, 128], consisting of two layers and 64 units. After outputting from the BiLSTM module, the temporal feature matrix is ​​converted into a fixed-dimensional feature vector with dimensions [number of samples, 256] through a fully connected layer, serving as input to the Transformer module.

[0031] The Transformer module, through cross-dimensional correlation mining, introduces a self-attention mechanism to uncover cross-dimensional correlation weights between three main categories of features: task, device, and environment. Examples include the correlation between ambient humidity and toner consumption, and the correlation between print task type and laser head power. This enhances the model's ability to capture key correlated features. Specifically, a multi-head self-attention mechanism is employed, with 8 heads, each with a feature dimension of 32, for a total dimension of 256 = 8 × 32. Attention weights between feature vectors quantify the degree of correlation between different feature dimensions, focusing on feature correlations that significantly impact task scheduling, device control, and fault prediction, such as the correlation between toner remaining and the number of print jobs, and the correlation between ambient temperature and roller malfunction. A single Transformer encoder layer is set, containing a multi-head self-attention layer and a feedforward neural network (FFN). The hidden layer dimension of the feedforward neural network is set to 512, and the activation function is GELU, used to perform non-linear transformations on the features output by the self-attention layer, enhancing feature representation capabilities. After encoder output, layer normalization is applied to ensure feature stability.

[0032] The feature vector output by the Transformer module is weighted and fused with the local features output by the CNN module and the temporal features output by the BiLSTM module. The weights are adaptively allocated by the self-attention mechanism to obtain the final comprehensive feature vector with dimensions [number of samples, 256], which is used for input and output layers.

[0033] Based on task requirements, a multi-output layer is designed to output three core results. Specifically, the task scheduling priority output layer uses a fully connected layer and a Softmax activation function to output three priority categories: high, medium, and low probability distributions. The final task scheduling priority is determined based on the maximum probability. The optimal control parameter output layer uses a fully connected layer and a Linear activation function to output the specific values ​​of four core control parameters: laser head power, drum speed, toner supply, and paper feed speed, ensuring that the parameters meet equipment operating specifications and task requirements. The fault risk prediction output layer uses a fully connected layer and a Sigmoid activation function to output the probability (0-1) of equipment failure within the next hour, along with the predicted fault type, such as insufficient toner, drum overheating, or drum damage, achieved through multi-class Softmax.

[0034] Step 103: Optimize the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model; Specifically, in this embodiment, the fixed weights in the traditional whale optimization algorithm are replaced with adaptive weights that are dynamically adjusted with the number of iterations. After the whale position is updated, a Logistic chaotic mapping is introduced to perturb the position, resulting in the improved IWOA whale optimization algorithm.

[0035] The whale population size is set to 30, with each individual whale corresponding to a set of hyperparameter combinations. An initial population that meets the search range of each hyperparameter is randomly generated. The maximum number of iterations T=300 and the fitness function threshold=0.001 are set. The fitness function is calculated using the test set error of the hybrid attention deep learning model. Based on the search, encirclement, and attack mechanism of the improved IWOA algorithm, combined with adaptive weights and chaotic perturbations, the position of each individual whale is updated. After each update, a new fitness value is calculated, and the individual with the best fitness value is retained. If the current iteration count reaches the maximum number of iterations, the optimization stops, and the current optimal hyperparameter combination is output. The optimal hyperparameter combination obtained by optimization is substituted into the hybrid attention deep learning model to obtain the target hybrid attention deep learning model.

[0036] Hyperparameters include the number of separable convolutional kernels, pooling kernel size, number of hidden layer units, dropout rate, number of layers, number of multi-head attention heads, and the dimension of hidden layers in the feedforward neural network. Building upon the traditional IWOA (Whale Optimization) algorithm, this paper introduces an adaptive weight adjustment mechanism and a chaotic perturbation strategy to address the slow convergence speed and susceptibility to local optima inherent in traditional algorithms. Specifically, the adaptive weight adjustment replaces the fixed weights of the traditional algorithm with dynamically adjusted weights that change with the number of iterations. The chaotic perturbation strategy introduces a Logistic chaotic mapping to perturb the position after the whale's position is updated, preventing the algorithm from getting trapped in local optima. The whale population size is set to 30, with each individual whale corresponding to a set of hyperparameter combinations. An initial population is randomly generated that conforms to the search range of each hyperparameter. The maximum number of iterations T = 300, and the fitness function threshold = 0.001. The mean squared error (MSE) of the test set of the hybrid attention deep learning model is used as the fitness function; a smaller fitness value indicates better performance of the hyperparameter combination. Based on the improved IWOA algorithm's search, encirclement, and attack mechanisms, combined with adaptive weights and chaotic perturbations, the position of each individual whale, i.e., the hyperparameter combination, is updated. After each update, a new fitness value is calculated, and the individual with the best fitness value is retained, representing the optimal hyperparameter combination. If the current iteration count reaches the maximum iteration count T, or the optimal fitness value is less than the set threshold, the optimization stops, and the current optimal hyperparameter combination is output; otherwise, iteration continues. The optimal hyperparameter combination obtained through optimization is substituted into the original hybrid attention deep learning model, and the model is retrained to obtain the target hybrid attention deep learning model with the best performance.

[0037] Step 104: Input the preprocessed data into the target hybrid attention deep learning model, and output the task scheduling priority, optimal control parameters and fault risk prediction results.

[0038] Specifically, in this embodiment, the input data is fed into the target hybrid attention deep learning model. The model, through the collaborative work of CNN, BiLSTM, and Transformer modules, completes feature extraction, temporal analysis, and association mining, ultimately outputting three core results. For details... Task scheduling priority: Outputs the priority of all pending print tasks, categorized as high, medium, or low. Priority is determined by a comprehensive analysis of task type, printing parameters, current device load, and environmental factors. For example, when printing batch images or when the device load is low, the priority is set to high. The priority ranking results are also output to provide a basis for printer task queue scheduling.

[0039] The optimal control parameters are output to match the current task and equipment status, including laser head power, roller speed, toner supply, and paper feed speed. After the control parameters are output, they are directly synchronized to the printer control system to achieve adaptive control of the equipment.

[0040] The fault risk prediction result outputs the fault risk probability of the equipment in the next hour, ranging from 0 to 1. When the probability is ≥0.7, it is judged as high risk, and the possible fault type and fault cause are also output. When the probability is between 0.3 and 0.7, it is judged as medium risk, and the equipment inspection is prompted. When the probability is <0.3, it is judged as low risk, and the equipment is operating normally.

[0041] Its beneficial effects lie in the fact that by collecting task, equipment, and environmental data from multiple dimensions and performing refined preprocessing, invalid data is effectively eliminated and dimensional differences are removed, providing high-quality input for the model and improving the accuracy of subsequent analysis. The constructed CNN-BiLSTM-Transformer hybrid model integrates local feature extraction, temporal dependency capture, and cross-dimensional correlation mining capabilities. Compared with a single model, it can more comprehensively capture the inherent correlation of data, improving the accuracy of task scheduling, parameter control, and fault prediction. The improved IWOA algorithm is used to optimize hyperparameters. Through adaptive weights and chaotic perturbation strategies, the problems of slow convergence and local optima in traditional algorithms are solved, ensuring that the model reaches optimal performance. Referring to the application experience of relevant optimization algorithms, the model's optimization accuracy and efficiency can be significantly improved. It can optimize the task queue, improve printing efficiency, adaptively adjust control parameters, reduce energy consumption, and predict fault risks in advance, reducing equipment downtime and improving the overall intelligent management level and operational stability of laser printers.

[0042] Please see Figure 2 In the deep learning-based laser printer control task management method, the hyperparameters of the hybrid attention deep learning model are optimized using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model, which includes the following steps: Step 201: Set the whale population size to 30, each whale individual corresponds to a set of hyperparameter combinations, randomly generate an initial population that meets the search range of each hyperparameter, and set the maximum number of iterations T=300 and the fitness function threshold=0.001. Step 202: Calculate the fitness function using the test set error of the hybrid attention deep learning model as the fitness function; Step 203: Based on the search, encirclement, and attack mechanism of the improved IWOA algorithm, combined with adaptive weights and chaotic perturbations, update the position of each individual whale. After each update, calculate the new fitness value and retain the individual with the best fitness value. Step 204: If the current iteration count reaches the maximum iteration count, stop the optimization and output the current optimal hyperparameter combination; substitute the optimal hyperparameter combination obtained by optimization into the hybrid attention deep learning model to obtain the target hybrid attention deep learning model.

[0043] The above describes embodiments of the deep learning-based laser printer control task management method of the present invention. Please refer to [link / reference]. Figure 3 In the deep learning-based laser printer control task management system, the laser printer control task management system includes the following modules: The print data acquisition module is used to collect print task data, equipment operating status data and environmental impact data of the laser printer. The collected data is then cleaned, outlier removed and feature standardized to obtain preprocessed data. The hybrid model building module is used to build a hybrid attention deep learning model based on CNN-BiLSTM-Transformer. It uses the separable convolutional structure of CNN convolutional neural network to extract local features from task, device and environment data, captures the temporal dependencies of device operation status and task execution process through BiLSTM long short-term memory network, and mines cross-dimensional correlation weights between different feature dimensions based on Transformer self-attention mechanism. The hybrid model optimization module is used to optimize the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model. The task control and management module is used to input preprocessed data into the target hybrid attention deep learning model and output task scheduling priority, optimal control parameters and fault risk prediction results.

[0044] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.

Claims

1. A laser printer control task management method based on deep learning, characterized in that, The laser printer control task management method includes the following steps: Data on pending print jobs, equipment operating status, and environmental impact from laser printers are collected. The collected data is then cleaned, outliers are removed, and features are standardized to obtain preprocessed data. We construct a hybrid attention deep learning model based on CNN-BiLSTM-Transformer. We use the separable convolutional structure of CNN convolutional neural network to extract local features from task, device and environment data. We use BiLSTM long short-term memory network to capture the temporal dependencies of device operation status and task execution process. We use Transformer self-attention mechanism to mine cross-dimensional correlation weights between different feature dimensions. The hyperparameters of the hybrid attention deep learning model are optimized using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model. The preprocessed data is input into the target hybrid attention deep learning model, which outputs task scheduling priority, optimal control parameters, and fault risk prediction results.

2. The laser printer control task management method based on deep learning as described in claim 1, characterized in that, The process involves collecting data on print jobs to be processed from the laser printer, equipment operating status data, and environmental impact data. The collected data is then sequentially cleaned, outlier removed, and feature standardized to obtain preprocessed data, including: Collect data on pending print jobs, equipment operating status, and environmental impact from laser printers; The collected data is processed for missing and duplicate values. The mean and standard deviation of the data are calculated using the 3σ principle. Data that exceeds the threshold range are identified as outliers, and the cleaned data is obtained. The continuous features in the cleaned data are standardized using the Z-score standardization method to obtain preprocessed data.

3. The laser printer control task management method based on deep learning as described in claim 1, characterized in that, The construction of a hybrid attention deep learning model based on CNN-BiLSTM-Transformer utilizes the separable convolutional structure of CNN convolutional neural networks to extract local features from task, device, and environmental data. It captures temporal dependencies in device operating states and task execution through BiLSTM long short-term memory networks, and mines cross-dimensional correlation weights between different feature dimensions based on the Transformer self-attention mechanism. This includes: The input layer of a CNN (Convolutional Neural Network) receives preprocessed data. It uses a single two-dimensional separable convolutional layer to split the input features into depth and spatial directions for convolution. The ReLU activation function is used to extract local features from the preprocessed data. The output of the convolutional layer is followed by a max-pooling layer to preserve key information about local features. Output local features.

4. The laser printer control task management method based on deep learning as described in claim 1, characterized in that, The construction of a hybrid attention deep learning model based on CNN-BiLSTM-Transformer utilizes the separable convolutional structure of CNN convolutional neural networks to extract local features from task, device, and environmental data. It captures temporal dependencies in device operating states and task execution through BiLSTM long short-term memory networks, and mines cross-dimensional correlation weights between different feature dimensions based on the Transformer self-attention mechanism. This includes: The BiLSTM long short-term memory network receives local features from the output of the CNN convolutional neural network and sets up 2 BiLSTM layers; The forward LSTM is responsible for capturing the temporal dependencies from the past to the present, while the backward LSTM is responsible for capturing the temporal dependencies from the present to the future. The output features of the two are concatenated to obtain the temporal features.

5. The laser printer control task management method based on deep learning as described in claim 1, characterized in that, The construction of a hybrid attention deep learning model based on CNN-BiLSTM-Transformer utilizes the separable convolutional structure of CNN convolutional neural networks to extract local features from task, device, and environmental data. It captures temporal dependencies in device operating states and task execution through BiLSTM long short-term memory networks, and mines cross-dimensional correlation weights between different feature dimensions based on the Transformer self-attention mechanism. This includes: The Transformer self-attention mechanism adopts a multi-head self-attention mechanism with 8 heads and 32 feature dimensions for each head. It quantifies the correlation features between different feature dimensions by calculating the attention weights between feature vectors. The encoder is set to a single Transformer encoder, which includes a multi-head self-attention layer and an FFN feedforward neural network. The hidden layer dimension of the feedforward neural network is set to 512, and the GELU activation function is used to perform non-linear transformation on the features output by the self-attention layer. The associated features, local features, and temporal features are weighted and fused to obtain a comprehensive feature vector.

6. The laser printer control task management method based on deep learning as described in claim 1, characterized in that, The process of optimizing the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model includes: The fixed weights in the traditional whale optimization algorithm are replaced with adaptive weights that are dynamically adjusted with the number of iterations. After the whale's position is updated, a Logistic chaotic mapping is introduced to perturb the position, resulting in the improved IWOA whale optimization algorithm.

7. The laser printer control task management method based on deep learning as described in claim 1, characterized in that, The process of optimizing the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model includes: The whale population size is set to 30, each whale individual corresponds to a set of hyperparameter combinations, and an initial population that meets the search range of each hyperparameter is randomly generated. At the same time, the maximum number of iterations T=300 and the fitness function threshold=0.001 are set. The fitness function is calculated using the test set error of the hybrid attention deep learning model as the fitness function. Based on the improved IWOA algorithm's search, encirclement, and attack mechanisms, combined with adaptive weights and chaotic perturbations, the position of each individual whale is updated, and a new fitness value is calculated after each update, retaining the individual with the best fitness value. If the current iteration count reaches the maximum iteration count, stop the optimization and output the current optimal hyperparameter combination; substitute the optimal hyperparameter combination obtained by optimization into the hybrid attention deep learning model to obtain the target hybrid attention deep learning model.

8. A laser printer control task management system based on deep learning, characterized in that, The laser printer control task management system includes the following modules: The print data acquisition module is used to collect print task data, equipment operating status data and environmental impact data of the laser printer. The collected data is then cleaned, outlier removed and feature standardized to obtain preprocessed data. The hybrid model building module is used to build a hybrid attention deep learning model based on CNN-BiLSTM-Transformer. It uses the separable convolutional structure of CNN convolutional neural network to extract local features from task, device and environment data, captures the temporal dependencies of device operation status and task execution process through BiLSTM long short-term memory network, and mines cross-dimensional correlation weights between different feature dimensions based on Transformer self-attention mechanism. The hybrid model optimization module is used to optimize the hyperparameters of the hybrid attention deep learning model using the improved IWOA whale optimization algorithm to obtain the target hybrid attention deep learning model. The task control and management module is used to input the preprocessed data into the target hybrid attention deep learning model and output the task scheduling priority, optimal control parameters and fault risk prediction results.

9. The deep learning-based laser printer control task management system as described in claim 8, characterized in that, The hybrid model optimization module includes the following sub-modules: An improved submodule is used to replace the fixed weights in the traditional whale optimization algorithm with adaptive weights that are dynamically adjusted with the number of iterations. After the whale's position is updated, a Logistic chaotic mapping is introduced to perturb the position, resulting in the improved IWOA whale optimization algorithm.

10. The deep learning-based laser printer control task management system as described in claim 8, characterized in that, The hybrid model optimization module includes the following sub-modules: The configuration submodule is used to set the whale population size to 30, with each individual whale corresponding to a set of hyperparameter combinations, randomly generating an initial population that meets the search range of each hyperparameter, and setting the maximum number of iterations T=300 and the fitness function threshold=0.

001. The computation submodule is used to calculate the fitness function using the test set error of the hybrid attention deep learning model as the fitness function. The update submodule is used to update the position of each whale individual based on the search, encirclement, and attack mechanism of the improved IWOA algorithm, combined with adaptive weights and chaotic perturbations. After each update, a new fitness value is calculated, and the individual with the best fitness value is retained. The submodule is obtained to stop the optimization if the current iteration count reaches the maximum iteration count, and outputs the current optimal hyperparameter combination; the optimal hyperparameter combination obtained by optimization is substituted into the hybrid attention deep learning model to obtain the target hybrid attention deep learning model.

Images