Adjusting and optimizing method based on input and output characteristic parameters and device applying method

Abstract

The invention provides an adjustment and optimization method based on input and output characteristic parameters and a device applying the method, belongs to the field of intelligent parameter adjustment and optimization, and aims to solve the problem that process parameters are difficult to adjust timely and accurately in related technologies. By combining the expected output characteristic parameters and the pre-acquired parameter tuning model, the adjustment amount of the input parameters is intelligently analyzed, evaluated and determined, so that the input parameters can be timely, accurately and reasonably adjusted, and the process parameters can be accurately controlled, the output condition can be optimized, and the output effect can be improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent parameter tuning, and in particular to a tuning method based on input and output characteristic parameters and a device applying the method. Background Art

[0002] In various current engineering systems and data analysis and processing fields, accurate control of the relationship between input parameters and output parameters is crucial.

[0003] In many practical application scenarios, such as the control of industrial production processes, it is necessary to adjust production process parameters according to real-time production conditions (i.e., input parameters) to ensure that product quality (output parameters) meets the expected standards. However, existing methods often find it difficult to fully and accurately consider the complex nonlinear relationship between input and output parameters. Many traditional models simply adjust input parameters based on the feedback of a single output parameter, ignoring the correlation between multiple output feature parameters and the influence of different feature dimensions.

[0004] For example, in the field of intelligent data analysis, the processing of large amounts of data and the optimization of analysis results also rely on the reasonable adjustment of input data features (input parameters) to obtain more valuable analysis results (output parameters). However, when analyzing output feature parameters, traditional methods cannot deeply explore the potential information of each feature dimension, and it is difficult to effectively quantify the differences between different output feature parameters.

[0005] In addition, the existing parameter tuning model structure is relatively simple, lacks sufficient flexibility and adaptability, and cannot fully handle complex multi-parameter input and multi-feature output situations. When faced with dynamically changing input-output relationships, it is difficult to adjust parameters in a timely and accurate manner, resulting in unstable system performance and failure to meet the needs of efficient and accurate tuning in practical applications.

[0006] It is in this context that there is an urgent need for a model that can comprehensively consider multiple factors and fully and accurately tune the input parameters to solve the many problems existing in the existing technology. Summary of the invention

[0007] The present application provides a tuning method based on input and output characteristic parameters and a device applying the method, which can tune input parameters based on the input and output characteristic parameters to optimize output conditions.

[0008] In a first aspect, the present application provides a tuning method based on input and output feature parameters. The method comprises: Acquire a first input parameter and a second input parameter in real time, and acquire a plurality of output characteristic parameters in real time; Based on the pre-acquired expected output feature parameters, analyzing the evaluation value parameters of each feature dimension of each output feature parameter, and determining the difference in evaluation values ​​of each feature dimension of every two associated output feature parameters; Inputting the first input parameter, the second input parameter, the evaluation value parameter of each characteristic dimension of each output characteristic parameter, and the evaluation value difference of each characteristic dimension of each two adjacent sequences of output characteristic parameters into a pre-constructed parameter tuning model to obtain a first parameter adjustment amount, wherein the first parameter adjustment amount represents the amount by which the first input parameter needs to be adjusted; The parameter tuning model includes an input layer, multiple hidden layers and an output layer, the input layer includes input nodes, and the input nodes are used to input the first input parameter, the second input parameter and the evaluation value parameter of each output feature parameter; The output layer is used to output the first parameter adjustment amount; The parameter tuning model also includes an output comprehensive evaluation model, which is used to determine an output comprehensive evaluation parameter in combination with the evaluation value parameter of each feature dimension of each output feature parameter, and the comprehensive evaluation parameter carries a timestamp; The output comprehensive evaluation model is also used to determine the effect trend parameter according to the output comprehensive evaluation parameter; The hidden layer includes special neurons, and the special neurons are also used to input the output comprehensive evaluation parameters and effect trend parameters.

[0009] By adopting the above technical solutions, the tuning model based on input and output feature parameters can break through traditional limitations, obtain dual input parameters and multiple output feature parameters in real time, and comprehensively consider the complex relationship between input and output. By deeply analyzing the evaluation value of each dimension of the output feature parameter, the difference in the evaluation value of the associated output feature parameter can be accurately determined, and potential information that is difficult to reach with traditional methods can be excavated. The unique parameter tuning model structure covers multiple hidden layers and special neurons, combined with the output comprehensive evaluation model. It can not only determine the input parameter adjustment amount based on multi-dimensional information, but also greatly enhance the model's flexibility and adaptability through comprehensive evaluation parameters and effect trend parameters. It can also quickly and accurately adjust parameters in a dynamically changing environment to ensure stable system performance and achieve efficient and accurate tuning. It effectively solves the problems of existing technologies in industrial production, intelligent data analysis and other fields, and provides strong support for the development of various industries.

[0010] Furthermore, each output feature parameter has evaluation value parameters of three feature dimensions; The output comprehensive evaluation model includes: , In the formula, E is the output comprehensive evaluation parameter, k is the number of output feature parameters, , , are the first evaluation value parameter, the second evaluation value parameter and the third evaluation value parameter of the i-th output feature parameter, respectively. , , are the first evaluation value difference, the second evaluation value difference and the third evaluation value difference of the i-th output feature parameter and the i+1-th output feature parameter respectively, is a pre-set weight coefficient; Comprehensive evaluation parameters of historical output with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the effect trend parameter.

[0011] Furthermore, the expected output feature parameters include a preset evaluation criterion for each feature dimension of each output feature parameter, and each preset evaluation criterion carries a dimension attention weight; The step of analyzing the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter to determine the difference in the evaluation value of each characteristic dimension of every two associated output characteristic parameters includes: Determine the evaluation value parameter of each feature dimension of the output feature parameter according to the preset evaluation benchmark of each feature dimension; The evaluation value parameters of each feature dimension of every two associated output feature parameters are compared to determine the corresponding evaluation value difference, and the associated relationship of the output feature parameters is pre-acquired.

[0012] In a second aspect, the present application provides a process parameter intelligent tuning device. The device includes a parameter acquisition module, a parameter processing module and a tuning determination module; The parameter acquisition module is used to acquire the first input parameter and the second input parameter in real time, and acquire multiple output characteristic parameters in real time; The parameter processing module is used to analyze the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter, and determine the difference in evaluation value of each characteristic dimension of every two associated output characteristic parameters; The tuning determination module is used to input the first input parameter, the second input parameter, the evaluation value parameter of each feature dimension of each output feature parameter, and the difference in the evaluation values ​​of each feature dimension of each two adjacent sequences of output feature parameters into a pre-constructed parameter tuning model to obtain a first parameter adjustment amount, wherein the first parameter adjustment amount represents the amount by which the first input parameter needs to be adjusted; The parameter tuning model includes an input layer, multiple hidden layers and an output layer, the input layer includes input nodes, and the input nodes are used to input the first input parameter, the second input parameter and the evaluation value parameter of each output feature parameter; The output layer is used to output the first parameter adjustment amount; The parameter tuning model also includes an output comprehensive evaluation model, which is used to determine an output comprehensive evaluation parameter in combination with the evaluation value parameter of each feature dimension of each output feature parameter, and the comprehensive evaluation parameter carries a timestamp; The output comprehensive evaluation model is also used to determine the effect trend parameter according to the output comprehensive evaluation parameter; The hidden layer includes special neurons, and the special neurons are also used to input the output comprehensive evaluation parameters and effect trend parameters.

[0013] Furthermore, the tuning determination module is further configured such that each output feature parameter has evaluation value parameters of three feature dimensions; The output comprehensive evaluation model includes: , In the formula, E is the output comprehensive evaluation parameter, k is the number of output feature parameters, , , are the first evaluation value parameter, the second evaluation value parameter and the third evaluation value parameter of the i-th output feature parameter, respectively. , , are the first evaluation value difference, the second evaluation value difference and the third evaluation value difference of the i-th output feature parameter and the i+1-th output feature parameter respectively, is a pre-set weight coefficient; Comprehensive evaluation parameters of historical output with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the effect trend parameter.

[0014] Furthermore, the parameter processing module is further configured such that the expected output feature parameter includes a preset evaluation benchmark for each feature dimension of each output feature parameter, and each preset evaluation benchmark carries a dimension attention weight; The step of analyzing the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter to determine the difference in the evaluation value of each characteristic dimension of every two associated output characteristic parameters includes: Determine the evaluation value parameter of each feature dimension of the output feature parameter according to the preset evaluation benchmark of each feature dimension; The evaluation value parameters of each feature dimension of every two associated output feature parameters are compared to determine the corresponding evaluation value difference, and the associated relationship of the output feature parameters is pre-acquired.

[0015] In summary, this application at least has the following beneficial effects: Provided are an intelligent continuous printing device and a control method thereof, which can determine the adjustment amount of input parameters according to multi-dimensional information, and are conducive to optimizing output conditions.

[0016] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present application, nor are they intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein: Figure 1 A schematic diagram of a tuning method based on input and output characteristic parameters in an embodiment of the present application is shown; Figure 2 A schematic diagram of a process parameter intelligent tuning device in an embodiment of the present application is shown; Figure 3 A schematic diagram of an intelligent continuous printing device in an embodiment of the present application is shown; Figure 4 A flow chart of a control method applied to an intelligent continuous printing device in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0020] The present application provides a tuning method based on input and output characteristic parameters and a device applying the method, which can tune input parameters based on input and output characteristic parameters to optimize output conditions.

[0021] In a first aspect, the present application discloses a tuning method based on input and output characteristic parameters. The model can be configured in a server, a controller, and the like.

[0022] Figure 1 A schematic diagram of a tuning method based on input and output characteristic parameters in an embodiment of the present application is shown.

[0023] Reference Figure 1 , the model specifically includes the following steps: S110: Acquire a first input parameter and a second input parameter in real time, and acquire a plurality of output characteristic parameters in real time; S120: Analyze the evaluation value parameter of each feature dimension of each output feature parameter based on the pre-acquired expected output feature parameter, and determine the difference in evaluation value of each feature dimension of every two associated output feature parameters; S130: Inputting the first input parameter, the second input parameter, the evaluation value parameter of each feature dimension of each output feature parameter, and the difference in evaluation values ​​of each feature dimension of each two adjacent sequences of output feature parameters into a pre-constructed parameter tuning model to obtain a first parameter adjustment amount, wherein the first parameter adjustment amount represents an amount by which the first input parameter needs to be adjusted; The parameter tuning model includes an input layer, multiple hidden layers and an output layer, the input layer includes an input node, and the input node is used to input the first input parameter, the second input parameter and the evaluation value parameter of each output feature parameter; The output layer is used to output the first parameter adjustment amount; The parameter tuning model also includes an output comprehensive evaluation model, which is used to determine an output comprehensive evaluation parameter in combination with the evaluation value parameter of each feature dimension of each output feature parameter, and the comprehensive evaluation parameter carries a timestamp; The output comprehensive evaluation model is also used to determine the effect trend parameter according to the output comprehensive evaluation parameter; The hidden layer includes special neurons, and the special neurons are also used to input the output comprehensive evaluation parameters and effect trend parameters.

[0024] Each output feature parameter has evaluation value parameters of three feature dimensions; The output comprehensive evaluation model includes: , In the formula, E is the output comprehensive evaluation parameter, k is the number of output feature parameters, , , are the first evaluation value parameter, the second evaluation value parameter and the third evaluation value parameter of the i-th output feature parameter, respectively. , , are the first evaluation value difference, the second evaluation value difference and the third evaluation value difference of the i-th output feature parameter and the i+1-th output feature parameter respectively, is a pre-set weight coefficient; Comprehensive evaluation parameters of historical output with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the effect trend parameter.

[0025] The expected output feature parameters include a preset evaluation benchmark for each feature dimension of each output feature parameter, and each preset evaluation benchmark carries a dimension attention weight; The step of analyzing the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter to determine the difference in the evaluation value of each characteristic dimension of every two associated output characteristic parameters includes: Determine the evaluation value parameter of each feature dimension of the output feature parameter according to the preset evaluation benchmark of each feature dimension; The evaluation value parameters of each feature dimension of every two associated output feature parameters are compared to determine the corresponding evaluation value difference, and the associated relationship of the output feature parameters is pre-acquired.

[0026] In a second aspect, an embodiment of the present application discloses a process parameter intelligent tuning device, which can be configured as a server or a controller, or included in a server or a controller.

[0027] Figure 2 A schematic diagram of a process parameter intelligent tuning device in an embodiment of the present application is shown.

[0028] Reference Figure 2 , the device specifically includes a parameter acquisition module, a parameter processing module and a tuning determination module; The parameter acquisition module is used to acquire the first input parameter and the second input parameter in real time, and acquire multiple output characteristic parameters in real time; The parameter processing module is used to analyze the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter, and determine the difference in evaluation value of each characteristic dimension of every two associated output characteristic parameters; The tuning determination module is used to input the first input parameter, the second input parameter, the evaluation value parameter of each feature dimension of each output feature parameter, and the difference in the evaluation values ​​of each feature dimension of each two adjacent sequences of output feature parameters into a pre-constructed parameter tuning model to obtain a first parameter adjustment amount, wherein the first parameter adjustment amount represents the amount by which the first input parameter needs to be adjusted; The output layer is used to output the first parameter adjustment amount; The parameter tuning model also includes an output comprehensive evaluation model, which is used to determine an output comprehensive evaluation parameter in combination with the evaluation value parameter of each feature dimension of each output feature parameter, and the comprehensive evaluation parameter carries a timestamp; The output comprehensive evaluation model is also used to determine the effect trend parameter according to the output comprehensive evaluation parameter; The hidden layer includes special neurons, and the special neurons are also used to input the output comprehensive evaluation parameters and effect trend parameters.

[0029] The tuning determination module is further configured such that each output feature parameter has evaluation value parameters of three feature dimensions; The output comprehensive evaluation model includes: , In the formula, E is the output comprehensive evaluation parameter, k is the number of output feature parameters, , , are the first evaluation value parameter, the second evaluation value parameter and the third evaluation value parameter of the i-th output feature parameter, respectively. , , are the first evaluation value difference, the second evaluation value difference and the third evaluation value difference of the i-th output feature parameter and the i+1-th output feature parameter respectively, is a pre-set weight coefficient; Comprehensive evaluation parameters of historical output with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the effect trend parameter.

[0030] The expected output feature parameters include a preset evaluation benchmark for each feature dimension of each output feature parameter, and each preset evaluation benchmark carries a dimension attention weight; The step of analyzing the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter to determine the difference in the evaluation value of each characteristic dimension of every two associated output characteristic parameters includes: Determine the evaluation value parameter of each feature dimension of the output feature parameter according to the preset evaluation benchmark of each feature dimension; The evaluation value parameters of each feature dimension of every two associated output feature parameters are compared to determine the corresponding evaluation value difference, and the associated relationship of the output feature parameters is pre-acquired.

[0031] The above is a disclosure of the tuning method and the device applying the tuning method of the embodiment of the present application. In order to ensure that the tuning method is better understood, the principles of the tuning method and the device are further disclosed below based on a specific intelligent continuous printing scenario.

[0032] In the scenario of intelligent continuous printing, the hardware environment in which the tuning method is applied is an intelligent continuous printing device.

[0033] Figure 3 A schematic diagram of an intelligent continuous printing device in an embodiment of the present application is shown.

[0034] Reference Figure 3 The device includes a feeding module, multiple printing modules (only one is shown in the figure) and a receiving module which are arranged in sequence, and also includes a humidification module and a humidity detection module configured at the feeding module and an image acquisition module configured at the receiving module.

[0035] Among them, the loading module is used to roll up printed paper and can input the printed paper into multiple subsequent printing modules; multiple printing modules are arranged in sequence and continuously and work together. Each printing module can complete a partial printing pattern on the printed product, and the partial printing patterns printed by all printing modules are combined to form an overall printing pattern on the printed product; the receiving module is used to roll up the printed product (that is, the printing paper continuously printed with the overall printing pattern).

[0036] The device also includes a humidification module and a humidity detection module configured at the feeding module and an image acquisition module configured at the receiving module.

[0037] Among them, the humidification module is specifically selected as a humidifier, which is configured at the feeding module and is used to humidify the printed paper. The humidification power data of the humidification module is controllable; the humidity detection module is also arranged at the feeding module, which can also be called the initial printing node, and is used to detect the humidity data of the printed paper to be input into the subsequent printing module; the image acquisition module is arranged at the receiving module, and can be specifically selected as a high-speed camera, which is used to collect the printing image data of the printed materials that have been fully dried and can be rolled up. Each printing image data contains an overall printing pattern of a printed product.

[0038] The device also includes a control module, which is connected to the humidification module, the humidity detection module and the image acquisition module to obtain the humidity data, the printing image data and the humidification power data of the humidification module, and is used to control the humidification power of the humidification module. Of course, the control module also serves as the control center of the device, connecting the feeding module, the printing module and the receiving module to achieve coordinated control of the feeding, continuous printing and receiving processes.

[0039] The principle of the control module's control over the humidification power is the specific application of the aforementioned tuning method, wherein the first input parameter is characterized by the humidity data of the initial printing node, the second input parameter is characterized by the humidification power used to humidify the initial printing node, the output characteristic parameter is characterized by the specific situation of the printing image data of the printed product, the evaluation value parameter is characterized by the color density uniformity, pattern clarity and printing standard of partial printed patterns from different printing steps, and the evaluation value difference is characterized by the color density difference value, pattern clarity difference value and printing standard difference value of adjacent printed patterns.

[0040] Figure 2 A flow chart of a control method applied to an intelligent continuous printing device in an embodiment of the present application is shown.

[0041] Reference Figure 2 , the tuning method in the control module is specifically configured to include the following steps: S210: Acquire humidity data of an initial printing node, humidification power data of a humidification module for humidifying the initial printing node, and printing image data of a printed product.

[0042] Humidity data and humidification power data are acquired in real time. The humidity data and humidification power data mentioned here are real-time data. Printing image data are acquired one by one according to complete printed products. The printing image data mentioned here refers to the overall printing pattern of the latest complete printed product. Here, the data acquisition process is essentially the data access process, which will not be elaborated.

[0043] S220: Analyze the printing image data of the printed product based on the pre-acquired standard printing data, determine multiple partial printing patterns in the printing image data that originate from different printing steps, determine the color density uniformity, pattern clarity and printing standard of each partial printing pattern, and determine the color density difference value, pattern clarity difference value and printing standard difference value of adjacent partial printing patterns.

[0044] In the method of this step, the standard printing data includes a standard printing image and a standard partial pattern contained in the standard printing image, and each standard partial pattern carries a partial attention weight.

[0045] The method of this step specifically includes: identifying partial printed patterns in the printed image data based on standard printing data, and determining the quantified color density uniformity and pattern clarity of each partial printed pattern; comparing each partial printed pattern with the corresponding standard partial pattern, and determining the quantified printing standard of each partial printed pattern; comparing the color density uniformity, pattern clarity and printing standard of partial printed patterns with adjacent parts, and determining the quantified color density difference value, pattern clarity difference value and printing standard difference value.

[0046] Specifically, the standard printing data not only reflects the ideal effect of the overall printing pattern, but also reflects the shape, position, color, density, etc. of each ideal partial printing pattern, as well as the printing module corresponding to each partial printing pattern. Based on the standard printing data and image analysis technology, each actual partial printing pattern in the actual printing image data can be determined.

[0047] After determining each part of the printed pattern in the printed image data, the color density uniformity, pattern clarity and printing standard of each part of the printed pattern can be analyzed. These analyses are achieved based on differential small areas (pixel points).

[0048] About color density uniformity For the calculation, please refer to the following contents: Assume that for some printed patterns , divide it into In each small area The color density is measured as .

[0049] First calculate the average value of the color density .

[0050] Then calculate the color density uniformity , the variance formula can be used to measure uniformity, that is, A smaller variance value indicates that the color density is more evenly distributed in each area of ​​the pattern. In order to make the uniformity value in a more intuitive range (for example, between 0 and 1), it can be normalized, such as , when the variance is 0, indicates complete uniformity; the larger the variance, The closer it is to 0.

[0051] About pattern clarity For the calculation, please refer to the following contents: Edge detection algorithms (such as Sobel operator, Canny operator, etc.) can be used to detect the edges of patterns. Taking the Sobel operator as an example, for pattern images ,Convolution operations are performed in the horizontal and vertical directions respectively.

[0052] set up is the gradient amplitude in the horizontal direction, is the gradient amplitude in the vertical direction, calculated by the following formula: , , In this formula, * represents the convolution operation.

[0053] Then calculate the gradient magnitude of each pixel .

[0054] Pattern clarity It can be measured by calculating the average value of the gradient amplitude of all edge pixels, that is, ,in is the number of edge pixels. It can also be normalized to a suitable range, for example, by dividing it by a reference value (such as the sum of the gradient amplitudes corresponding to the clearest pattern in theory) to obtain is the sum of the reference gradient amplitudes.

[0055] About printing standards For the calculation of , please refer to the following contents: Printing pattern Key indicators are used to measure the difference from the standard printing effect, such as color deviation , shape deviation , Dimensional deviation For each key indicator , with corresponding actual deviation value .

[0056] First, normalize each deviation value. For example, if the color deviation range is , then the normalized color deviation Similarly, the deviations of other key indicators are normalized.

[0057] Printing Standard It can be obtained by calculating the weighted sum of these normalized deviations, that is, ,in is the weight of each key indicator deviation in the standard measurement, and . The closer the value is to 1, the closer the printed pattern is to the standard printing effect.

[0058] Based on the above content, the quantitative color density uniformity, pattern clarity and printing standard of each partial printed pattern can be determined. On this basis, assuming that there are adjacent parts between every two partial printed patterns on the overall printed pattern, and assuming that there are k partial printed patterns, there are k-1 color density difference values, pattern clarity difference values ​​and printing standard difference values, and the color density difference value is the absolute value of the difference in color density uniformity between two adjacent partial printed patterns, the pattern clarity difference value is the absolute value of the difference in pattern clarity between two adjacent partial printed patterns, and the printing standard difference value is the absolute value of the difference in printing standards between two adjacent partial printed patterns.

[0059] S230: Input the humidity data, humidification power data, color density uniformity, pattern clarity and printing standard of each partial printed pattern, and color density difference value, pattern clarity difference value and printing standard difference value of each adjacent partial printed pattern into a pre-trained humidity control model to determine the power adjustment amount of the humidification power, and the power adjustment amount is used to control the change of the humidification power data of the humidification module.

[0060] The humidity control model includes an input layer, multiple hidden layers and an output layer. The input layer includes input nodes, and the input nodes are used to input the humidity data, humidification power data, and the color density uniformity, pattern clarity and printing standard of each partial printed pattern, as well as the color density difference value, pattern clarity difference value and printing standard difference value of each adjacent partial printed pattern; the output layer is used to output the power adjustment amount of the humidification module.

[0061] The humidity control model is specifically a neural network model, which has one input layer, three hidden layers and one output layer.

[0062] Assuming that there are k partial printed patterns on the printed product, there are k color density uniformity, pattern clarity and printing standard, as well as k-1 color density difference values, pattern clarity difference values ​​and printing standard difference values, plus the humidity data and humidification power data used for input, the input layer has a total of input nodes, each input node is used to input one of the aforementioned quantized data, and the quantized data input into the neural network model can be normalized data.

[0063] In the input layer, , , are the color density uniformity, pattern clarity and printing standard of the i-th part of the printed pattern, , , are the color density difference value, pattern clarity difference value and printing standard difference value between the i-th part of the printed pattern and the i+1-th part of the printed pattern, h is the humidity data, and P is the humidification power data.

[0064] In a specific example, regarding three hidden layers, the first hidden layer has neurons, using sigmoid activation function, for the first hidden layer Neurons , the input weighted sum is: , in The input layer neurons to the first hidden layer The connection weights of neurons, The first hidden layer The bias of a neuron.

[0065] The output after the Sigmoid activation function is: , The second hidden layer has neurons, using the ReLU activation function.

[0066] For the mth neuron in the second layer , the input weighted sum is: , in The first hidden layer neurons to the second hidden layer The connection weights of neurons, The second hidden layer The bias of a neuron.

[0067] The output after the ReLU activation function is: , Assume that the third hidden layer has The Tanh activation function is used.

[0068] For the third hidden layer Neurons , the input weighted sum is: , in The second hidden layer neurons to the third hidden layer The connection weights of neurons, The third hidden layer The bias of a neuron.

[0069] The output after the Tanh activation function is: , The output layer has only one output , that is, the adjustment amount of humidification power.

[0070] The output is calculated using linear regression: , in The third hidden layer The connection weights from neurons to the output layer, is the bias of the output layer.

[0071] Regarding the loss function and training method of the neural network model, please refer to the following content: The mean square error (MSE) loss function is used to measure the deviation between the predicted humidification power adjustment and the actual expected humidification power adjustment, as well as the deviation between the comprehensive printing effect value and the ideal printing effect value (which can be set according to experience or standards). Assume that the number of training samples is , No. The actual humidification power adjustment of the sample is , the humidification power adjustment predicted by the model is , No. The ideal comprehensive printing effect of samples is , the calculated comprehensive printing effect value is , then the loss function is: , in The coefficient is used to balance the importance of the two deviations. Through the back propagation algorithm, the gradient is calculated according to the loss function and the weight and bias parameters in the neural network are updated. The training is continuously iterated to gradually reduce the loss function value, so that the model can accurately predict the adjustment amount of humidification power and optimize the comprehensive printing effect according to the input printing-related data and the current humidification status data, and at the same time use the integrated historical trend value information of the printing effect to better adjust the model decision.

[0072] Furthermore, in order to enable the humidity control model to better deal with the situation where the printing effect value does not meet expectations and the changes in the printing effect value, in the method of this step, the humidity control model also includes a printing effect determination sub-model; the printing effect determination sub-model is used to determine the printing effect value according to the color density uniformity, pattern clarity and printing standard of each partial printed pattern, and the color density difference value, pattern clarity difference value and printing standard difference value of each adjacent partial printed pattern, and each printing effect value carries a timestamp; the printing effect determination sub-model is also used to determine the printing effect trend value according to the printing effect value carrying the timestamp; the hidden layer includes special neurons, and the special neurons are also used to input the printing effect value and the printing effect trend value.

[0073] Specifically, the printing effect determination submodel is used to determine the printing effect value based on the color density uniformity, pattern clarity and printing standard of each partial printed pattern, as well as the color density difference value, pattern clarity difference value and printing standard difference value of each adjacent partial printed pattern, and each printing effect value carries a timestamp; the printing effect determination submodel is also used to determine the printing effect trend value based on the printing effect value carrying the timestamp; the hidden layer includes special neurons, and the special neurons are also used to input the printing effect value and the printing effect trend value.

[0074] Specifically, the printing effect determination sub-model includes: , In the formula, E is the printing effect value, k is the number of partial printing patterns, , , are the color density uniformity, pattern clarity and printing standard of the i-th part of the printed pattern, , , are the color density difference value, pattern clarity difference value and printing standard difference value between the i-th printed pattern and the i+1-th printed pattern, respectively. is a pre-set weight coefficient; Printing effect values ​​for history with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the printing effect trend value.

[0075] Based on this, a special neuron can be specifically set in the second hidden layer of the neural network model , the input of the special neuron is the printing effect value and the printing effect trend value, and the output is the weighted calculation result of the two, for example , where is the calculation weight of the printing effect trend value, The calculation weight of the printing effect value.

[0076] This tuning method can also be applied to any other scenarios by analogy to give full play to its advantages.

[0077] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0078] In summary, this application at least has the following beneficial effects: Provided are an intelligent continuous printing device and a control method thereof, which can determine the adjustment amount of input parameters according to multi-dimensional information, which is beneficial to optimizing output conditions. For example, in an intelligent continuous printing scenario, the humidity of printed paper can be accurately controlled, which is beneficial to ensuring the quality of printed products.

[0079] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A tuning method based on input and output characteristic parameters, characterized in that: include: Acquire a first input parameter and a second input parameter in real time, and acquire a plurality of output characteristic parameters in real time; Based on the pre-acquired expected output feature parameters, analyzing the evaluation value parameters of each feature dimension of each output feature parameter, and determining the difference in evaluation values ​​of each feature dimension of every two associated output feature parameters; Inputting the first input parameter, the second input parameter, the evaluation value parameter of each characteristic dimension of each output characteristic parameter, and the evaluation value difference of each characteristic dimension of each two adjacent sequences of output characteristic parameters into a pre-constructed parameter tuning model to obtain a first parameter adjustment amount, wherein the first parameter adjustment amount represents the amount by which the first input parameter needs to be adjusted; The parameter tuning model includes an input layer, multiple hidden layers and an output layer, the input layer includes input nodes, and the input nodes are used to input the first input parameter, the second input parameter and the evaluation value parameter of each output feature parameter; The output layer is used to output the first parameter adjustment amount; The parameter tuning model also includes an output comprehensive evaluation model, which is used to determine an output comprehensive evaluation parameter in combination with the evaluation value parameter of each feature dimension of each output feature parameter, and the comprehensive evaluation parameter carries a timestamp; The output comprehensive evaluation model is also used to determine the effect trend parameter according to the output comprehensive evaluation parameter; The hidden layer includes special neurons, and the special neurons are also used to input the output comprehensive evaluation parameters and effect trend parameters.

2. The tuning method according to claim 1, characterized in that: Each output feature parameter has evaluation value parameters of three feature dimensions; The output comprehensive evaluation model includes: ； In the formula, E is the output comprehensive evaluation parameter, k is the number of output feature parameters, , , are the first evaluation value parameter, the second evaluation value parameter and the third evaluation value parameter of the i-th output feature parameter, respectively. , , are the first evaluation value difference, the second evaluation value difference and the third evaluation value difference of the i-th output feature parameter and the i+1-th output feature parameter respectively, is a pre-set weight coefficient; Comprehensive evaluation parameters of historical output with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the effect trend parameter.

3. According to the tuning method of claim 1, the expected output feature parameters include a preset evaluation benchmark for each feature dimension of each output feature parameter, and each preset evaluation benchmark carries a dimension attention weight; The step of analyzing the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter to determine the difference in the evaluation value of each characteristic dimension of every two associated output characteristic parameters includes: Determine the evaluation value parameter of each feature dimension of the output feature parameter according to the preset evaluation benchmark of each feature dimension; The evaluation value parameters of each feature dimension of every two associated output feature parameters are compared to determine the corresponding evaluation value difference, and the associated relationship of the output feature parameters is pre-acquired.

4. A process parameter intelligent tuning device, characterized in that: It includes a parameter acquisition module, a parameter processing module and a tuning determination module; The parameter acquisition module is used to acquire the first input parameter and the second input parameter in real time, and acquire multiple output characteristic parameters in real time; The parameter processing module is used to analyze the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter, and determine the difference in evaluation value of each characteristic dimension of every two associated output characteristic parameters; The tuning determination module is used to input the first input parameter, the second input parameter, the evaluation value parameter of each feature dimension of each output feature parameter, and the difference in the evaluation values ​​of each feature dimension of each two adjacent sequences of output feature parameters into a pre-constructed parameter tuning model to obtain a first parameter adjustment amount, wherein the first parameter adjustment amount represents the amount by which the first input parameter needs to be adjusted; The parameter tuning model includes an input layer, multiple hidden layers and an output layer, the input layer includes input nodes, and the input nodes are used to input the first input parameter, the second input parameter and the evaluation value parameter of each output feature parameter; The output layer is used to output the first parameter adjustment amount; The parameter tuning model also includes an output comprehensive evaluation model, which is used to determine an output comprehensive evaluation parameter in combination with the evaluation value parameter of each feature dimension of each output feature parameter, and the comprehensive evaluation parameter carries a timestamp; The output comprehensive evaluation model is also used to determine the effect trend parameter according to the output comprehensive evaluation parameter; The hidden layer includes special neurons, and the special neurons are also used to input the output comprehensive evaluation parameters and effect trend parameters.

5. The device according to claim 4, characterized in that The tuning determination module is further configured such that each output feature parameter has evaluation value parameters of three feature dimensions; The output comprehensive evaluation model includes: ； In the formula, E is the output comprehensive evaluation parameter, k is the number of output feature parameters, , , are the first evaluation value parameter, the second evaluation value parameter and the third evaluation value parameter of the i-th output feature parameter, respectively. , , are the first evaluation value difference, the second evaluation value difference and the third evaluation value difference of the i-th output feature parameter and the i+1-th output feature parameter respectively, is a pre-set weight coefficient; Comprehensive evaluation parameters of historical output with timestamps Perform a difference operation to obtain a difference sequence , calculates the slope of the linear regression model for the differenced series is the effect trend parameter.

6. The device according to claim 4, wherein the parameter processing module is further configured such that the expected output feature parameter includes a preset evaluation benchmark for each feature dimension of each output feature parameter, and each preset evaluation benchmark carries a dimension attention weight; The step of analyzing the evaluation value parameter of each characteristic dimension of each output characteristic parameter based on the pre-acquired expected output characteristic parameter to determine the difference in the evaluation value of each characteristic dimension of every two associated output characteristic parameters includes: Determine the evaluation value parameter of each feature dimension of the output feature parameter according to the preset evaluation benchmark of each feature dimension; The evaluation value parameters of each feature dimension of every two associated output feature parameters are compared to determine the corresponding evaluation value difference, and the associated relationship of the output feature parameters is pre-acquired.

Images