A printer control method and system based on deep learning

Abstract

The present invention discloses a printer control method and system based on deep learning, the method comprising: collecting and preprocessing print task data, obtaining normalized file size data, file page number data and task submission time data, as well as unique hot coding data of file type, submission user ID and user priority; calculating file parameter feature vector according to the normalized file size data, file page number data and task submission time data, as well as unique hot coding data of file type, submission user ID and user priority; constructing a task merging model, merging tasks according to the file parameter feature vector, and outputting the merged file parameter feature vector; calculating the merged file priority according to the merged file parameter feature vector; executing the print task according to the merged file priority, and monitoring the printing process. The present invention can solve the problems of unreasonable priority sorting and low printing efficiency of traditional shared printers.

Description

Technical Field

[0001] The present invention relates to the field of printer technology, and in particular to a printer control method and system based on deep learning. Background Art

[0002] Shared printer task control involves scheduling, managing, and optimizing print tasks for all parties, ensuring an efficient and smooth printing process. Traditional printer task control methods primarily include simple task queuing and task prioritization. While these methods can maintain printing order to a certain extent, they still expose some shortcomings in practical applications.

[0003] First, traditional methods are relatively simple in file processing and usually do not conduct further in-depth analysis of the parameter types of the print files. Therefore, when multiple files with different parameter types need to be printed together, the printer needs to frequently switch between different files, and each switch will consume a certain amount of time, which is not conducive to further improving printing efficiency; secondly, traditional methods are not smart enough in sorting the priority of print tasks. They often only consider the submission time of the task, and ignore other important factors that may affect printing efficiency, such as user priority and task type, which can easily lead to an unreasonable execution order of print tasks, thereby making users wait longer and reducing printing efficiency. Summary of the Invention

[0004] In view of this, the present invention provides a printer task control method and system based on deep learning, aiming to solve the problems of unreasonable priority sorting and insufficient printing efficiency of traditional shared printers.

[0005] A printer task control method based on deep learning, comprising the following steps:

[0006] S1: Collect and preprocess print task data to obtain normalized file size data, file page number data, task submission time data, and one-hot encoded data of file type, submitting user ID, and user priority;

[0007] S2: Calculate the file parameter feature vector based on the normalized file size data, file page number data, task submission time data, and the one-hot encoded data of file type, submitting user ID, and user priority;

[0008] S3: Build a task merging model, merge tasks according to the file parameter feature vector, and output the merged file parameter feature vector;

[0009] S4: Calculate the priority of the merged file based on the merged file parameter feature vector;

[0010] S5: Execute the printing task according to the priority of the merged files and monitor the printing process.

[0011] Furthermore, the S1 step includes:

[0012] S11: Collecting print task data, wherein the print task data includes file type, file size, number of file pages, task submission time, submitting user ID, and user priority;

[0013] S12: For data with missing values ​​in the print task data, directly delete the entire data; for repeated print task data, sort them by time and keep the most recent print task data;

[0014] S13: normalizing the file size data, file page number data, and task submission time data using a Z-score method to obtain normalized file size data, file page number data, and task submission time data;

[0015] S14: Perform one-hot encoding on the file type data, the submitting user ID data, and the user priority data to obtain one-hot encoded data.

[0016] Furthermore, the S2 step includes:

[0017] First, we use a cubic polynomial transformation to enhance the normalized task submission time data. We then combine the normalized file size data and file page count data to calculate the numerical feature vector. We then calculate the encoded feature vector based on the one-hot encoded data.

[0018] Secondly, construct the encoding gate and the numerical gate, update the numerical feature vector and the encoding feature vector, and calculate the original parameter feature vector of the file;

[0019] Then, the task priority is calculated based on the one-hot encoded data of user priority, file page number, and file size;

[0020] Finally, according to the task priority, the original parameter feature vector of the file is updated and the file type feature is enhanced to obtain the file parameter feature vector.

[0021] Furthermore, the step S2 includes:

[0022] S21: Use cubic polynomial transformation to enhance the normalized task submission time data, and combine it with the normalized file size data and file page number data to calculate the numerical feature vector. The calculation formula is as follows:

[0023] ;

[0024] ;

[0025] Where, is the cubic polynomial transformation sequence of the normalized task submission time data, Submit time data for normalized tasks, is the square of the normalized task submission time data, The cube of the normalized task submission time data; is a numerical eigenvector, For multi-layer perceptron operation, For splicing operations, is the normalized file size data, is the normalized file page number data;

[0026] According to the one-hot encoded data, the encoded feature vector is calculated using the following formula:

[0027] ;

[0028] Where, is the encoded feature vector, One-hot encoding data for file types, To submit the one-hot encoded data of the user ID, One-hot encoding data of user priority;

[0029] S22: Construct the encoding gate and the numerical gate, update the numerical feature vector and the encoding feature vector, and calculate the original parameter feature vector of the file. The calculation formula is as follows:

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] Where, is the numerical gate, is the sigmoid function, is the numerical weight, is the numerical bias vector, For the coded door, is the encoding weight, is the encoding bias vector, is the updated numerical feature vector, is an element-wise multiplication operation, is the updated encoded feature vector, is the original parameter feature vector of the file;

[0036] S23: Calculate the task priority based on the one-hot encoded data of user priority, file page number, and file size. The calculation formula is as follows:

[0037] ;

[0038] Where, For task priority;

[0039] S24: Based on the task priority, the original file parameter feature vector is updated and the file type feature is enhanced to obtain the file parameter feature vector. The calculation formula is as follows:

[0040] ;

[0041] ;

[0042] Where, is the original parameter feature vector of the updated file, is the file parameter feature vector;

[0043] Furthermore, the S3 step includes:

[0044] Firstly, a task merging model is constructed using a neural network-based encoder to calculate the similarity between files based on their parameter feature vectors.

[0045] Secondly, the similarity between files is adjusted according to the task priority, and whether to merge tasks is determined based on the similarity between the adjusted files;

[0046] Finally, for tasks that need to be merged, the merged file parameter feature vector is calculated and output.

[0047] Furthermore, the S3 step includes:

[0048] S31: A task merging model is constructed using a neural network-based encoder. The similarity between files is calculated based on the file parameter feature vectors. The calculation formula is as follows:

[0049] ;

[0050] ;

[0051] Where, is the file parameter feature vector of the i-th file, i is the first file index, is the representation vector of the file parameter feature vector of the i-th file in the low-dimensional embedding space. Similarly, is the representation vector of the file parameter feature vector of the j-th file in the low-dimensional embedding space, j is the second file index, For encoder operation, are encoder parameters, and the encoder operation adopts a multilayer perceptron or Transformer block as the neural network model, is the similarity between the i-th file and the j-th file, is the vector norm;

[0052] S32: Adjust the similarity between files based on the task priority. The adjustment formula is as follows:

[0053] ;

[0054] Where, is the adjusted similarity between the i-th file and the j-th file, is the task priority influence coefficient, is the task priority of the i-th file, is the task priority of the jth file;

[0055] Determine whether to merge tasks based on the similarity between the adjusted files:

[0056] when When , no task merging operation is performed;

[0057] when When , merge the tasks of the i-th and j-th files;

[0058] S33: For tasks that need to be merged, calculate and output the merged file parameter feature vector. The calculation formula is as follows:

[0059] ;

[0060] Where, is the parameter feature vector of the merged file.

[0061] Furthermore, the S4 step includes:

[0062] S41: Separate the features of the number of file pages and the task submission time based on the merged file parameter feature vector. The separation formula is as follows:

[0063] ;

[0064] Where, is the separated document page number feature, Submit time characteristics for the separated tasks, For DenseNet network operation;

[0065] S42: Calculate the priority of the merged file based on the separated file page count feature and task submission time feature, combined with the merged file parameter feature vector. The calculation formula is as follows:

[0066] ;

[0067] Where, The priority of the merged file.

[0068] Furthermore, the step S5 includes:

[0069] S51: sorting the printing tasks according to the priority of the merged files;

[0070] S52: Allocate the print tasks to the shared printer in order according to the print task sorting result, and execute the print tasks;

[0071] S53: After each printing task is completed, the printing duration information is recorded.

[0072] The present invention also discloses a printer control system based on deep learning, comprising:

[0073] Data preprocessing unit: collects and preprocesses print task data to obtain normalized file size data, file page number data, task submission time data, and one-hot encoded data of file type, submitting user ID, and user priority;

[0074] File parameter feature vector calculation unit: calculates the file parameter feature vector based on the normalized file size data, file page number data, task submission time data, and the one-hot encoded data of the file type, submitting user ID, and user priority;

[0075] Task merging unit: builds a task merging model, merges tasks according to the file parameter feature vector, and outputs the merged file parameter feature vector;

[0076] Priority sorting unit: calculates the priority of the merged file based on the parameter feature vector of the merged file;

[0077] File printing unit: executes printing tasks according to the priority of the merged files and monitors the printing process;

[0078] The above-mentioned printer control method based on deep learning is implemented through this system.

[0079] Compared with the prior art, the advantages of the present invention are:

[0080] (1) The present invention comprehensively considers multiple dimensions such as user priority, number of file pages, and file size, and uses a neural network model to calculate task priority, which helps the system prioritize important and urgent printing tasks. At the same time, the present invention constructs a task merging model to intelligently determine which files should be merged based on file parameter feature vectors and task priorities, thereby reducing file switching time and improving printing efficiency.

[0081] (2) The present invention generates a coding gate and a numerical gate through two nonlinear transformations. The coding gate contains numerical features, and the numerical gate contains coding features. After fusing each other's feature information, the original numerical feature vector and coding feature vector are updated, thereby generating a high-quality file original parameter feature vector. In subsequent steps, with the help of the file original parameter feature vector, the task priority can be calculated more reasonably, and it can be determined which files should be merged, thereby improving printing efficiency.

[0082] (3) The present invention proposes intelligent task merging based on deep learning. It combines file type and task priority in the form of autoencoder clustering, adaptively calculates the similarity of each file, merges tasks for files with higher similarity, and calculates the parameter feature vector of the merged file. Files with lower similarity are not merged, and the original file parameter feature vector is retained, which reduces the switching time of printing tasks and improves the efficiency of shared printing. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A flowchart of a printer control method based on deep learning provided by the present invention;

[0084] Figure 2 Schematic diagram of the process of obtaining the original parameter feature vector of the file in the method of the present invention;

[0085] Figure 3 Schematic diagram of the process of task merging in the method of the present invention. DETAILED DESCRIPTION

[0086] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention fall within the scope of protection of the present invention.

[0087] Example 1:

[0088] See also Figure 1 , a printer control method based on deep learning, comprising the following steps:

[0089] S1: Collect and pre-process print task data to obtain normalized file size data, file page number data, task submission time data, and unique hot encoding data of file type, submitting user ID, and user priority; specifically,

[0090] S11: Collecting print task data, wherein the print task data includes file type, file size, number of file pages, task submission time, submitting user ID, and user priority;

[0091] S12: For data with missing values ​​in the print task data, directly delete the entire data; for duplicate print task data, sort them by time, retain the most recent print task data and delete other duplicate print task data;

[0092] S13: normalizing the file size data, file page number data, and task submission time data using a Z-score method to obtain normalized file size data, file page number data, and task submission time data;

[0093] S14: Perform one-hot encoding on the file type data, the submitting user ID data, and the user priority data to obtain one-hot encoded data, where the one-hot encoded data includes the one-hot encoded data of the file type, the one-hot encoded data of the submitting user ID, and the one-hot encoded data of the user priority.

[0094] For example, if the system supports three file types: PDF, DOCX, and PPTX, the one-hot encoding of the file type can be expressed as:

[0095] PDF: [1, 0, 0]

[0096] DOCX: [0, 1, 0]

[0097] PPTX: [0, 0, 1]

[0098] There are three users in the system: User001, User002, and User003. The unique hot encoding of the submitted user ID can be expressed as:

[0099] User001: [1, 0, 0]

[0100] User002: [0, 1, 0]

[0101] User003: [0, 0, 1]

[0102] The user priorities in the system are divided into three levels: high, medium, and low. The one-hot encoding of the user priority can be expressed as:

[0103] High: [1, 0, 0]

[0104] Medium: [0, 1, 0]

[0105] Low: [0, 0, 1]

[0106] S2: See also Figure 2 ,Calculate the file parameter feature vector based on the normalized file size data, file page number data, task submission time data, and the one-hot encoding data of file type, submitting user ID and user priority;

[0107] S21: Use cubic polynomial transformation to enhance the normalized task submission time data, and combine it with the normalized file size data and file page number data to calculate the numerical feature vector. The calculation formula is as follows:

[0108] ;

[0109] ;

[0110] Where, is the cubic polynomial transformation sequence of the normalized task submission time data, Submit time data for normalized tasks, is the square of the normalized task submission time data, The cube of the normalized task submission time data; is a numerical eigenvector, For multi-layer perceptron operation, For splicing operations, is the normalized file size data, is the normalized file page number data;

[0111] The present invention performs a cubic polynomial transformation on the normalized task submission time, enhancing the task submission time data in a nonlinear manner. The enhanced cubic polynomial transformation sequence is then concatenated and fused with the normalized file size data and file page number data to obtain a numerical feature vector. Compared with the traditional method of direct concatenation and fusion, the idea of ​​the present invention can improve the model's attention to task submission time, making the numerical feature vector more expressive.

[0112] According to the one-hot encoded data, the encoded feature vector is calculated using the following formula:

[0113] ;

[0114] Where, is the encoded feature vector, One-hot encoding data for file types, To submit the one-hot encoded data of the user ID, One-hot encoding data of user priority;

[0115] S22: Construct the encoding gate and the numerical gate, update the numerical feature vector and the encoding feature vector, and calculate the original parameter feature vector of the file. The calculation formula is as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] Where, is the numerical gate, is the sigmoid function, is the numerical weight, is the numerical bias vector, For the coded door, is the encoding weight, is the encoding bias vector, is the updated numerical feature vector, is an element-wise multiplication operation, is the updated encoded feature vector, is the original parameter feature vector of the file;

[0122] In a printer task control system, the processing of file features is crucial for subsequent task merging and priority sorting. Traditional methods find it difficult to fully capture the complex relationships and dynamic changes between file features, and lack the ability to dynamically process the complex relationships between file features. Each feature is relatively independent and information is not interoperable, which in turn affects the representation effect of the overall feature vector. Therefore, the present invention introduces a gating mechanism to dynamically adjust the importance of numerical features and coding features. The gating mechanism generates a coding gate and a numerical gate through two nonlinear transformations. The coding gate contains numerical features, and the numerical gate contains coding features. After fusing each other's feature information, the original numerical feature vector and the coding feature vector are updated. Finally, the two updated feature vectors are spliced ​​to obtain the file original parameter feature vector, realizing dynamic adjustment and optimization of the numerical features and coding features. The file original parameter feature vector constructed in this way, due to better information interaction, helps the system understand the overall characteristics of the file. In subsequent steps, it can also more reasonably calculate the task priority and determine which files should be merged, thereby improving printing efficiency.

[0123] S23: Calculate the task priority based on the one-hot encoded data of user priority, file page number, and file size. The calculation formula is as follows:

[0124] ;

[0125] Where, For task priority;

[0126] Since most traditional methods determine task priority based on a single feature (such as user priority, task submission time, etc.), ignoring the impact of file characteristics (such as file page number and file size) on priority, this invention comprehensively considers user priority, file page number and file size to calculate task priority more comprehensively;

[0127] S24: Based on the task priority, the original file parameter feature vector is updated and the file type feature is enhanced to obtain the file parameter feature vector. The calculation formula is as follows:

[0128] ;

[0129] ;

[0130] Where, is the original parameter feature vector of the updated file, is the file parameter feature vector;

[0131] The present invention performs feature enhancement through task priority and file type to improve the representation ability of file parameter feature vectors; considering that the file type feature is very important in the subsequent task merging step and can reduce the switching time, the present invention enhances the file type in the data dimension and feature dimension through direct splicing and multi-layer perceptron fusion, which is conducive to the model to better pay attention to the file type and pave the way for subsequent task merging.

[0132] S3: See also Figure 3 , build a task merging model, merge tasks according to the file parameter feature vector, and output the merged file parameter feature vector;

[0133] S31: A task merging model is constructed using a neural network-based encoder. The similarity between files is calculated based on the file parameter feature vectors. The calculation formula is as follows:

[0134] ;

[0135] ;

[0136] Where, is the file parameter feature vector of the i-th file, i is the first file index, is the representation vector of the file parameter feature vector of the i-th file in the low-dimensional embedding space. Similarly, is the representation vector of the file parameter feature vector of the j-th file in the low-dimensional embedding space, j is the second file index, For encoder operation, are encoder parameters, and the encoder operation adopts a multilayer perceptron or Transformer block as the neural network model, is the similarity between the i-th file and the j-th file, is the vector norm;

[0137] ;

[0138] Where, is the adjusted similarity between the i-th file and the j-th file, is the task priority influence coefficient, They are the task priority of the i-th file and the task priority of the j-th file respectively;

[0139] Determine whether to merge tasks based on the similarity between the adjusted files:

[0140] when When , no task merging operation is performed;

[0141] when When , merge the tasks of the i-th and j-th files;

[0142] In addition, for two files with a small number of pages, when it is detected that they are highly similar but have a large difference in task priorities, correcting the task priorities may result in , which results in tasks not being merged and prolonged file switching time;

[0143] Therefore, when When similarity adjustment is not performed directly based on task priority, the threshold of file page count is calculated first. The calculation formula is as follows:

[0144] ;

[0145] ;

[0146] ;

[0147] Where, is the average number of pages of all files to be printed within the preset time. The number of all files to be printed within the preset time. is the counting index, is the number of pages of the k-th file, It is the standard deviation of the number of pages of all documents to be printed within the preset time. The specific duration of the preset time can be set according to actual work needs, such as 0.5 hours, 1 hour, etc. is the file page number threshold, is the file page number threshold adjustment factor;

[0148] When the number of pages of both files is less than When , merge the tasks directly;

[0149] When at least one of the two files has a page count greater than or equal to When , no task merging is performed;

[0150] S33: For tasks that need to be merged, calculate and output the merged file parameter feature vector. The calculation formula is as follows:

[0151] ;

[0152] Where, is the parameter feature vector of the merged file;

[0153] Since most traditional methods only merge tasks based on a single rule (such as file size, number of pages, etc.), ignoring the more complex similarities between files and the influence of task priority, the present invention first constructs a task merging model and calculates the similarity between files based on file parameter feature vectors. It then adjusts the similarity by introducing the influence of task priority. Finally, it determines whether to merge tasks based on the adjusted similarity and calculates the merged file parameter feature vector. This allows for more accurate judgment of which files should be merged, reduces unnecessary printing switching time, and optimizes the printing process.

[0154] S4: Calculate the priority of the merged file based on the merged file parameter feature vector;

[0155] S41: Based on the merged file parameter feature vector, the feature separation of the number of file pages and the task submission time is performed. The calculation method is:

[0156] ;

[0157] in, They are the number of separated file pages and task submission time characteristics, For DenseNet network operation;

[0158] DenseNet, a deep convolutional neural network, significantly enhances feature reuse and gradient flow by introducing dense connections. This allows each layer in the network to directly receive the output of all previous layers as input, helping to capture richer feature information. Feature separation can clearly distinguish different aspects of the merged file parameter feature vector (such as the number of file pages and task submission time), allowing the model to more flexibly handle different input data. Even when faced with new or unseen file types or task situations, the model can accurately prioritize based on the separated features, enhancing the model's generalization and adaptability.

[0159] S42: Calculate the priority of the merged file based on the separated file page number feature and the task submission time feature, combined with the merged file parameter feature vector. The calculation method is:

[0160] ;

[0161] Where, The priority of the merged file.

[0162] S5: Execute the printing task according to the priority of the merged files and monitor the printing process;

[0163] S51: sorting the printing tasks according to the priority of the merged files;

[0164] S52: Allocate the print tasks to the shared printer in order according to the print task sorting result, and execute the print tasks;

[0165] S53: After each printing task is completed, the printing duration information is recorded.

[0166] Example 2:

[0167] A printer control system based on deep learning, comprising:

[0168] Data preprocessing unit: collects and preprocesses print task data to obtain normalized file size data, file page number data, task submission time data, and one-hot encoded data of file type, submitting user ID, and user priority;

[0169] File parameter feature vector calculation unit: calculates the file parameter feature vector based on the normalized file size data, file page number data, task submission time data, and the one-hot encoded data of the file type, submitting user ID, and user priority;

[0170] Task merging unit: builds a task merging model, merges tasks according to the file parameter feature vector, and outputs the merged file parameter feature vector;

[0171] Priority sorting unit: calculates the priority of the merged file based on the parameter feature vector of the merged file;

[0172] File printing unit: executes printing tasks according to the priority of the merged files and monitors the printing process;

[0173] The method in Example 1 is implemented by the deep learning-based printer control system in this embodiment.

[0174] It should be noted that the serial numbers of the above-mentioned embodiments of the present invention are for descriptive purposes only and do not represent the advantages or disadvantages of the embodiments. In addition, the terms "including", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method comprising the element.

[0175] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0176] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A printer control method based on deep learning, characterized in that: The following steps are involved: S1: Collect and preprocess the printing task data to obtain normalized file size data, file page number data, task submission time data, and unique hot encoding data of file type, submitting user ID, and user priority; S2: Calculate the file parameter feature vector based on the normalized file size data, file page number data, task submission time data, and the one-hot encoding data of file type, submitting user ID, and user priority; S3: Build a task merging model, merge tasks according to the file parameter feature vector, and output the merged file parameter feature vector; S4: Calculate the priority of the merged file according to the merged file parameter feature vector; S5: Execute the printing task according to the priority of the merged files and monitor the printing process.

2. The printer control method based on deep learning according to claim 1, characterized in that: The S1 step includes: S11: Collecting print task data, wherein the print task data includes file type, file size, number of file pages, task submission time, submitting user ID and user priority; S12: For data with missing values ​​in the print task data, directly delete the entire data; for repeated print task data, sort them by time and keep the most recent print task data; S13: normalizing the file size data, the file page number data, and the task submission time data using a Z-score method to obtain normalized file size data, file page number data, and task submission time data; S14: Perform one-hot encoding processing on the file type data, the submitting user ID data and the user priority data to obtain one-hot encoded data.

3. The printer control method based on deep learning according to claim 1, characterized in that: The S2 step includes: First, the normalized task submission time data is enhanced using a cubic polynomial transformation, and the numerical feature vector is calculated based on the normalized file size data and file page number data. The encoding feature vector is calculated based on the one-hot encoding data. Secondly, construct the encoding gate and the numerical gate, update the numerical feature vector and the encoding feature vector, and calculate the original parameter feature vector of the file; Then, the task priority is calculated based on the one-hot encoded data of user priority, file page number, and file size; Finally, according to the task priority, the original parameter feature vector of the file is updated, and the file type feature is enhanced to obtain the file parameter feature vector.

4. The printer control method based on deep learning according to claim 3, characterized in that: The S2 step includes: S21: Use cubic polynomial transformation to enhance the normalized task submission time data, and combine the normalized file size data and file page number data to calculate the numerical feature vector. The calculation formula is as follows: ； ； In the formula, is the cubic polynomial transformation sequence of the normalized task submission time data, Submit time data for normalized tasks, is the square of the normalized task submission time data, The cube of the normalized task submission time data; is a numerical feature vector, For multi-layer perceptron operation, For splicing operation, is the normalized file size data, is the normalized file page number data; According to the one-hot encoded data, the encoded feature vector is calculated using the following formula: ； In the formula, is the encoded feature vector, is the one-hot encoded data of the file type, To submit the one-hot encoded data of the user ID, One-hot encoding data of user priority; S22: Construct the encoding gate and the numerical gate, update the numerical feature vector and the encoding feature vector, and calculate the original parameter feature vector of the file. The calculation formula is as follows: ； ； ； ； ； In the formula, is the numerical gate, is the sigmoid function, is the numerical weight, is the numerical bias vector, For the coded gate, is the encoding weight, is the encoding bias vector, is the updated numerical feature vector, is an element-by-element multiplication operation, is the updated encoded feature vector, is the original parameter feature vector of the file; S23: Calculate the task priority based on the one-hot encoded data of user priority, file page number and file size. The calculation formula is as follows: ； In the formula, Prioritize the tasks; S24: According to the task priority, the original parameter feature vector of the file is updated, and the file type feature is enhanced to obtain the file parameter feature vector. The calculation formula is as follows: ； ； In the formula, is the original parameter feature vector of the updated file, is the file parameter feature vector.

5. The printer control method based on deep learning according to claim 4, characterized in that: The S3 step includes: Firstly, a neural network-based encoder is used to build a task merging model, and the similarity between files is calculated according to the file parameter feature vectors; Secondly, the similarity between files is adjusted according to the task priority, and whether to merge tasks is determined based on the similarity between the adjusted files; Finally, for tasks that need to be merged, the merged file parameter feature vector is calculated and output.

6. The printer control method based on deep learning according to claim 5, characterized in that: The S3 step includes: S31: A task merging model is constructed using a neural network-based encoder. The similarity between files is calculated based on the file parameter feature vector. The calculation formula is as follows: ； ； In the formula, is the file parameter feature vector of the i-th file, i is the first file index, is the representation vector of the file parameter feature vector of the i-th file in the low-dimensional embedding space. Similarly, is the representation vector of the file parameter feature vector of the jth file in the low-dimensional embedding space, j is the second file index, For encoder operation, are encoder parameters, and the encoder operation adopts a multilayer perceptron or a Transformer block as a neural network model, is the similarity between the i-th file and the j-th file, is the vector norm; S32: Adjust the similarity between files according to the task priority. The adjustment formula is as follows: ； In the formula, is the adjusted similarity between the i-th file and the j-th file, is the task priority influence coefficient, is the task priority of the i-th file, is the task priority of the jth file; Decide whether to merge tasks based on the similarity between the adjusted files: when When , no task merging operation is performed; when When , merge the tasks of the i-th and j-th files; S33: For the tasks that need to be merged, calculate and output the merged file parameter feature vector, and the calculation formula is as follows: ； In the formula, is the parameter feature vector of the merged file.

7. The printer control method based on deep learning according to claim 6, characterized in that: The S4 step includes: S41: Separate the features of the number of file pages and the task submission time according to the merged file parameter feature vector. The separation formula is as follows: ； In the formula, is the separated file page number feature, Submit time characteristics for the separated tasks, For DenseNet network operation; S42: Calculate the priority of the merged file based on the separated file page number feature and task submission time feature, combined with the merged file parameter feature vector. The calculation formula is as follows: ； In the formula, The priority of the merged file.

8. The printer control method based on deep learning according to any one of claims 1 to 7, characterized in that: The S5 step includes: S51: sorting the printing tasks according to the priority of the merged files; S52: Allocate the print tasks to the shared printer in order according to the print task sorting result, and execute the print tasks; S53: After each printing task is completed, the printing duration information is recorded.

9. A printer control system based on deep learning, characterized in that: include: Data preprocessing unit: collects and preprocesses printing task data to obtain normalized file size data, file page number data, task submission time data, and unique hot encoding data of file type, submitting user ID, and user priority; File parameter feature vector calculation unit: calculates the file parameter feature vector according to the normalized file size data, file page number data, task submission time data, and the one-hot encoding data of the file type, submitting user ID, and user priority; Task merging unit: builds a task merging model, merges tasks according to the file parameter feature vector, and outputs the merged file parameter feature vector; Priority sorting unit: calculates the priority of the merged file according to the parameter feature vector of the merged file; File printing unit: executes printing tasks according to the priority of the merged files and monitors the printing process; To implement the printer control method based on deep learning as described in any one of claims 1-8.

Images