A monitoring and adjusting method for a rice crispies production line

Abstract

The present application relates to rice crisp candy production technical field, specifically disclose a kind of monitoring and adjusting method of rice crisp candy production line, comprising the following steps: according to the set value and measured value of oil temperature, using incremental PID algorithm dynamically adjusts oil heating heater power;Monitoring sugar solution temperature and viscosity, fusion temperature threshold and viscosity change trend, using fuzzy control algorithm dynamically adjusts sugar heating heater power.The present application improves the degree of automation of rice crisp candy production, reduces the dependence on manual experience, ensures the quality of large-scale production of rice crisp candy, conducive to building rice crisp candy production digital wisdom factory.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rice crisp candy production, and particularly to a monitoring and adjusting method for a rice crisp candy production line. BACKGROUND

[0002] Rice crisp candy is a famous traditional snack, which is sweet and tasty, has a rice crisp fragrance, and is deeply loved by people. The main process of making rice crisp candy is puffing, oil and sugar melting, oil and sugar ingredient stirring, rice crisp stirring, shaping and cutting, and sub-packaging, and the oil and sugar melting has a greater impact on the quality of the finished rice crisp candy. At present, for the oil and sugar melting in the production process of rice crisp candy, although some mechanical equipment is used, the degree of automatic control is low, and the following problems exist:

[0003] 1. During oil melting, the control and adjustment of oil temperature is not fine enough, and usually only the fire is lowered when the oil temperature is high, and the fire is raised when the oil temperature is low;

[0004] 2. During sugar melting, the sugar color is mainly observed by manual observation to determine the temperature and viscosity of the sugar solution, which is easy to cause the sugar solution to be scorched.

[0005] It can be seen that in the above key process steps, the control precision is poor, and the dependence on manual experience is very high. If the operator is inexperienced or does not observe in time, the quality of the rice crisp candy will be seriously affected. SUMMARY

[0006] In order to solve the problems of insufficient precision of oil and sugar temperature adjustment and insufficient degree of automatic production in the production process of rice crisp candy in the prior art, the present application provides a monitoring and adjusting method for a rice crisp candy production line.

[0007] The technical solution adopted by the present application is:

[0008] A monitoring and adjusting method for a rice crisp candy production line, comprising the following steps:

[0009] According to the set value and the measured value of the oil temperature, the incremental PID algorithm is used to dynamically adjust the oil melting heater power, so that the oil temperature is maintained within the appropriate range;

[0010] According to the monitored temperature and viscosity of the sugar solution, the temperature threshold and the viscosity change trend are fused, and the fuzzy control algorithm is used to dynamically adjust the sugar melting heater power to prevent the sugar solution from being scorched;

[0011] The neural network is introduced to optimize the parameters of the PID algorithm and the fuzzy control algorithm in real time. The neural network is trained based on historical production data and real-time sensor data, and is used to predict the optimal control parameters.

[0012] Preferably, the incremental PID algorithm for dynamically adjusting the oil melting heater power comprises the following steps:

[0013] Obtain the set value of oil temperature The measured oil temperature at the current moment. Measured oil temperature at the previous moment Measured oil temperature at the previous moment The power of the oil heater at the previous moment Calculate the oil temperature deviation at the current moment. Calculate the oil temperature deviation at the previous moment. Calculate the oil temperature deviation at the previous time step. ;

[0014] Calculate the control increment of the oil heating heater power The formula is:

[0015] ;

[0016] in, is the proportionality coefficient, representing the strength of the system's response to the current deviation; These are the integral coefficients used to eliminate steady-state errors; These are the differential coefficients, used to suppress overshoot;

[0017] Based on the control increment Adjust the power of the oil heating heater to control its current output power. .

[0018] Preferably, the parameters of the PID algorithm are dynamically optimized using a neural network. , and This includes the following steps:

[0019] Historical oil temperature data, ambient temperature data, oil volume data, and production batch information were collected as training datasets.

[0020] The system is trained using a multilayer perceptron neural network, with the input layer including the current oil temperature deviation. Historical oil temperature data and ambient temperature data are used as the output layer for optimization. , and ;

[0021] The neural network training uses the backpropagation algorithm, and the loss function is the mean squared error, as shown in the formula:

[0022]

[0023] in, This is the loss value. For the sample size, For the first The true value of each sample For the first The neural network prediction value for each sample;

[0024] Deploy the trained neural network, input the current sensor data in real time, and output adaptive PID parameters.

[0025] Preferably, dynamically adjusting the power of the sugar boiling heater using a fuzzy control algorithm includes the following steps:

[0026] Calculate the temperature deviation of the sugar solution and the viscosity change rate of sugar solution Temperature deviation of sugar solution and the viscosity change rate of sugar solution Perform fuzzification processing, divide fuzzy sets, design a fuzzy rule base, and perform inference based on the fuzzy rule base;

[0027] Defuzzification generates control output, which dynamically adjusts the power of the sugar boiling heater.

[0028] Preferably, the temperature deviation of the sugar solution Calculate using the following steps:

[0029] Obtain the set value of sugar solution temperature. and the measured value of the sugar solution temperature at the current moment. ;

[0030] Calculate the temperature deviation of the sugar solution ;

[0031] viscosity change rate Calculate using the following steps:

[0032] Use a viscometer to collect viscosity data of the sugar solution at a preset frequency;

[0033] Take the nearest The viscosity data at each time point were fitted with a linear trend, and the derivative was calculated using the following formula:

[0034]

[0035] in, For the first The time at a specific point in time; for The time average at each time point; For the first Viscosity values ​​at each time point; for The average viscosity at each time point.

[0036] Preferably, a neural network-enhanced fuzzy control algorithm is used, comprising the following steps:

[0037] An adaptive neuro-fuzzy inference model is constructed, whose input layer receives the temperature deviation of sugar solution and the rate of change of viscosity of sugar solution ;

[0038] The first hidden layer of the adaptive neuro-fuzzy inference model performs fuzzification of the input variables, mapping and to fuzzy sets, where each input variable corresponds to 3-5 Gaussian membership functions, with the function form being:

[0039]

[0040] where, denotes the input variable or ; is the center point of the th membership function; is the width of the th membership function; is the membership degree, denotes the membership degree of the variable to the fuzzy set ;

[0041] The second hidden layer of the adaptive neuro-fuzzy inference model realizes the antecedent part of fuzzy rules, with each node corresponding to a fuzzy rule, and the output being the rule firing strength :

[0042]

[0043] where, denotes the th rule; is the firing strength of the th rule; and correspond to the fuzzy sets of temperature deviation and rate of change of viscosity, respectively;

[0044] The third hidden layer of the adaptive neuro-fuzzy inference model realizes rule normalization, calculating the normalized firing strength :

[0045]

[0046] where, is the normalized firing strength of the th rule; is the total number of fuzzy rules; is the firing strength of the th rule;

[0047] The fourth hidden layer of the adaptive neuro-fuzzy inference model performs rule consequent calculation, each rule corresponding to a linear function:

[0048]

[0049] wherein, , , is the consequent parameter of the i-th rule;

[0050] The output layer of the adaptive neuro-fuzzy inference model calculates the final control variable

[0051]

[0052] wherein, is the total number of fuzzy rules;

[0053] The adaptive neuro-fuzzy inference model is trained using historical production data, and the training target is to minimize the loss function

[0054]

[0055] wherein, is the expert experience control variable; is the actual control variable calculated by the adaptive neuro-fuzzy inference model according to the current real-time input; is the regularization coefficient, is the number of training samples, represents the i-th training sample.

[0056] Preferably, it also includes an online learning mechanism:

[0057] A dual neural network architecture is deployed, including a main neural network and a shadow neural network;

[0058] The main neural network performs real-time control, and the shadow neural network continuously learns the latest production data;

[0059] Every 24 hours, the performance indicators of the main neural network and the shadow neural network are compared, and when the mean square error of the shadow neural network is more than 10% lower than that of the main neural network, the network is automatically switched;

[0060] The online learning adopts a momentum gradient descent algorithm, and the update formula is:

[0061]

[0062] wherein, represents the model parameter; , ,​​​​ respectively represent second, second and second iteration of the model parameters; is a learning rate; is a momentum coefficient; is a gradient.

[0063] The beneficial effects of the present application are:

[0064] In the oil boiling stage, the temperature of lard is fed back in real time by infrared temperature measurement, and the fire power is dynamically adjusted by PID algorithm; based on the sugar liquid viscosity sensing unit and temperature feedback, the fire power is adjusted by fuzzy control method to prevent sugar liquid from coking; by automatic transformation of the existing equipment, the automation degree of rice crisp candy production is improved, the dependence on manual experience is reduced, and the quality of large-scale production of rice crisp candy is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 is a schematic diagram of the overall process of the monitoring and adjusting method in the embodiment of the present application. DETAILED DESCRIPTION

[0066] The present application will be described in detail below in conjunction with the drawings and examples.

[0067] Embodiment

[0068] In the production process of rice crisp candy, especially in the oil boiling and sugar boiling process, the precision of oil temperature and sugar temperature adjustment is insufficient, and the dependence on manual experience is high, which is not conducive to improving the quality of rice crisp candy and large-scale production. In order to improve the automation degree of rice crisp candy production and improve the quality of rice crisp candy automatic production, the present application provides a monitoring and adjusting method for rice crisp candy production line, as shown in Figure 1 The method comprises the following steps:

[0069] According to the set value and the measured value of the oil temperature, the incremental PID algorithm is used to dynamically adjust the power of the oil boiling heater, so that the oil temperature is maintained within the appropriate range;

[0070] According to the monitored sugar liquid temperature and viscosity, the temperature threshold and viscosity change trend are fused, and the fuzzy control algorithm is used to dynamically adjust the power of the sugar boiling heater to prevent the sugar liquid from coking;

[0071] The neural network is introduced to optimize the parameters of the PID algorithm and the fuzzy control algorithm in real time. The neural network is trained based on historical production data and real-time sensor data, and is used to predict the optimal control parameters.

[0072] A hybrid intelligent control system is constructed by combining incremental PID algorithm, fuzzy control algorithm and neural network. The method fundamentally improves the automation level of production and reduces the absolute dependence on human experience, providing a basic method framework for the standardized, large-scale and high-quality production of rice crisp candy.

[0073] The dynamic adjustment of the oil heating heater power by using incremental PID algorithm includes the following steps:

[0074] Obtaining the set value of oil temperature , the measured value of oil temperature at the current time , the measured value of oil temperature at the last time , the measured value of oil temperature at the time before the last time , and the power of the oil heating heater at the last time , calculating the oil temperature deviation at the current time , calculating the oil temperature deviation at the last time , calculating the oil temperature deviation at the time before the last time ;

[0075] Calculating the control amount increment of the oil heating heater power , the formula is:

[0076] ;

[0077] Wherein, is the proportional coefficient, indicating the response strength of the system to the current deviation; is the integral coefficient, used to eliminate steady-state error; is the differential coefficient, used to suppress overshoot;

[0078] According to the control amount increment , the power of the oil heating heater is adjusted, and the current output power of the oil heating heater is controlled as .

[0079] The oil temperature is accurately, quickly and stably controlled. Compared with traditional on-off control or position PID, the incremental PID algorithm can effectively avoid the sharp fluctuation of power output, prevent overshoot, and make the oil temperature smoothly maintain in the optimal puffing temperature range (such as 180℃-190℃), so as to ensure the uniformity and quality of rice crisp puffing, and reduce the waste rate caused by improper oil temperature.

[0080] According to the puffing characteristics of rice crisp, the set value of oil temperature is generally 180℃-190℃, such as 185℃, the initial value of the proportional coefficient is generally set as 8.0, the initial value of the integral coefficient is generally set as 0.05, and the initial value of the differential coefficient The initial value of the coefficient can be set to 2.0. The coefficient can also be adjusted according to the heating stage, such as increasing the proportional coefficient during the heating-up stage to quickly approach the set temperature; and decreasing the proportional coefficient during the steady-state stage to 6, and increasing the integral coefficient to 0.1 to suppress overshoot.

[0081] Adopting neural network dynamic optimization PID algorithm parameters 、 and , including the following steps:

[0082] Collecting historical oil temperature data, ambient temperature data, oil volume data and production batch information as training data set;

[0083] Using a multilayer perceptron neural network for training, the input layer includes the current oil temperature deviation , historical oil temperature data, ambient temperature data, and the output layer is the optimized 、 and ;

[0084] The neural network training adopts the back propagation algorithm, and the loss function is the mean square error, the formula is:

[0085]

[0086] Wherein, is the loss value, is the number of samples, is the true value of the th sample, is the neural network prediction value of the th sample;

[0087] Deploy the trained neural network, input the current sensor data in real time, and output adaptive PID parameters.

[0088] The data-driven adaptive capability is introduced, which solves the pain points of fixed traditional PID parameters, and cannot adapt to different production stages (such as heating-up stage, steady-state stage) and environmental changes (such as ambient temperature, oil volume). Through real-time parameter adjustment of neural network, the system can automatically adapt to complex working conditions, respond faster, overshoot smaller, steady-state accuracy higher, significantly improving the intelligent level and robustness of control.

[0089] Adopting fuzzy control algorithm to dynamically adjust the power of sugar melting heater includes the following steps:

[0090] Calculating the temperature deviation of the sugar solution and the viscosity change rate of the sugar solution , the temperature deviation of the sugar solution Rate of change of viscosity of the sugar solution The fuzzy processing is performed, the fuzzy set is divided, the fuzzy rule base is designed, and the inference is performed based on the fuzzy rule base.

[0091] The control amount output is generated by defuzzification, and the power of the sugar boiling heater is dynamically adjusted.

[0092] An effective solution is provided for the characteristics of multivariable and nonlinearity of the sugar boiling process. By simultaneously fusing the two key parameters of temperature deviation and viscosity change trend, the expert's comprehensive decision-making process is simulated, and the sugar solution is predicted and prevented from coking earlier and more accurately than single temperature control, thereby improving the foresight and reliability of the control.

[0093] Temperature deviation of the sugar solution The temperature deviation of the sugar solution is calculated by the following steps:

[0094] Obtaining a set value of the temperature of the sugar solution and a measured value of the temperature of the sugar solution at the current time ;

[0095] Calculating the temperature deviation of the sugar solution ;

[0096] Rate of change of viscosity The rate of change of viscosity is calculated by the following steps:

[0097] Using a viscometer to collect viscosity data of the sugar solution at a preset frequency;

[0098] Taking the viscosity data of the last time points, fitting a linear change trend to derive, and the formula is as follows:

[0099]

[0100] Wherein, is the time of the th time point; is the time average of the th time point; is the viscosity value of the th time point; is the viscosity average of the th time point.

[0101] An accurate and quantifiable input feature calculation method is defined. In particular, the rate of change of viscosity is calculated by linear fitting and derivation, which can effectively capture the dynamic change trend of the viscosity, filter out the random noise of the sensor data, provide a more stable and reliable input signal for subsequent fuzzy inference, and improve the anti-interference ability and decision accuracy of the entire system.

[0102] The neural network enhanced fuzzy control algorithm comprises the following steps:

[0103] An adaptive neuro-fuzzy inference model is constructed, the input layer of which receives the temperature deviation of sugar solution and the rate of change of viscosity of sugar solution ;

[0104] The first hidden layer of the adaptive neuro-fuzzy inference model performs fuzzification of the input variables, mapping and to fuzzy sets, wherein each input variable corresponds to 3-5 Gaussian membership functions, the function form being:

[0105]

[0106] wherein represents the input variable or ; is the center point of the th membership function; is the width of the th membership function; is the membership degree, indicating the membership degree of the variable to the fuzzy set ;

[0107] The second hidden layer of the adaptive neuro-fuzzy inference model realizes the antecedent part of the fuzzy rules, each node corresponding to a fuzzy rule, and the output being the rule triggering strength :

[0108]

[0109] wherein represents the th rule; is the triggering strength of the th rule; and correspond to the fuzzy sets of the temperature deviation and the rate of change of viscosity respectively;

[0110] The third hidden layer of the adaptive neuro-fuzzy inference model realizes rule normalization, calculating the normalized triggering strength :

[0111]

[0112] wherein is the normalized triggering strength of the th rule; is the total number of fuzzy rules; is the The trigger strength of the rule;

[0113] The fourth hidden layer of the adaptive neural fuzzy inference model performs rule consequent computation, with each rule corresponding to a linear function:

[0114]

[0115] in, , , For the first The consequent parameter of the rule;

[0116] The output layer of the adaptive neural fuzzy inference model calculates the final control quantity. :

[0117]

[0118] in, The total number of fuzzy rules;

[0119] An adaptive neural fuzzy inference model is trained using historical production data, with the training objective being to minimize the loss function. :

[0120]

[0121] in, The quantity is controlled by expert experience; This is the actual control quantity calculated by the adaptive neural fuzzy inference model based on the current real-time input; The regularization coefficient is . The number of training samples, Indicates the first Training samples.

[0122] Generally, the target temperature for heating the sugar solution is around 120℃. In this embodiment, 120℃ is selected, and the set temperature deviation range is -20℃ to 20℃. The fuzzification of the temperature deviation is shown in Table 1 below:

[0123] Table 1. Fuzzy set partitioning of temperature deviation

[0124]

[0125] For example, the measured value of the sugar solution temperature at the current moment. The temperature is 110℃, with a deviation of [missing information]. At 10℃, the temperature is mainly in the positive small (PS) set.

[0126] Assuming viscosity change rate The range is 0~30 cSt / min, and the fuzzification is shown in Table 2 below:

[0127] Table 2 Fuzzy set partition of viscosity change rate

[0128]

[0129] For example, viscosity change rate is 18 cSt / min, at this time, mainly belongs to the medium (Medium) set.

[0130] Based on the combination of temperature deviation and viscosity change rate, the control strategy is formulated, and the fuzzy rule base is as follows (part of the rules) as shown in Table 3:

[0131] Table 3 Joint fuzzy rule base

[0132]

[0133] Through the ANFIS structure, the fixed fuzzy rules and membership functions that depend on expert experience are converted into neural network parameters that can be automatically learned and optimized through historical data. This enables the system to autonomously discover the optimal control rules, freeing it from a heavy reliance on expert knowledge, enabling self-evolution of the control strategy, and greatly improving the system's intelligence and adaptive ability.

[0134] It also includes an online learning mechanism:

[0135] A dual neural network architecture is deployed, including a main neural network and a shadow neural network;

[0136] The main neural network performs real-time control, and the shadow neural network continuously learns the latest production data;

[0137] Every 24 hours, compare the performance indicators of the main neural network and the shadow neural network. When the mean square error of the shadow neural network is more than 10% lower than that of the main neural network, automatically switch the network;

[0138] The online learning uses the momentum gradient descent algorithm, and the update formula is:

[0139]

[0140] where, denotes the model parameters; , , denote the model parameters after the first, second and third iterations, respectively; is the learning rate; is the momentum coefficient; is the gradient.

[0141] ​The online learning mechanism is a general and system-level architecture, which can be applied to the neural network for optimizing the PID algorithm, the neural network for enhancing the fuzzy control, or both. The main neural network ensures the continuous and stable operation of the production line, and the shadow neural network realizes uninterrupted learning. The periodic evaluation and safe switching mechanism ensures that the production line has a low risk of downtime during continuous improvement, realizes the goal of 'learning while producing', and enables the system to have continuous learning ability and continuously adapt to changes in raw materials and equipment aging.

[0142] The method of the present application can be applied to newly built production lines or automatic transformation of existing production line equipment, improving the automation level of rice crisp candy production, reducing the dependence on manual experience, ensuring the quality of large-scale production of rice crisp candy, and facilitating the construction of a digital and intelligent rice crisp candy production plant.

[0143] The above-described embodiments only express the specific implementation of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent of the present application. It should be noted that, for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application.

Claims

1. A method of monitoring and regulating a rice crispie production line, characterized in that, Comprising the following steps: According to the set value and the measured value of the oil temperature, the incremental PID algorithm is used to dynamically adjust the power of the oil heating heater, so that the oil temperature is maintained within the appropriate range; According to the monitored sugar liquid temperature and viscosity, the temperature threshold and viscosity change trend are fused, and the fuzzy control algorithm is used to dynamically adjust the power of the sugar heating heater to prevent the sugar liquid from coking; Among them, the fuzzy control algorithm is used to dynamically adjust the power of the sugar heating heater, comprising the following steps: calculating the temperature deviation of the sugar solution and the rate of change of the viscosity of the sugar solution calculating the temperature deviation of the sugar solution and the rate of change of the viscosity of the sugar solution fuzzy processing, dividing fuzzy sets, designing fuzzy rule base, and reasoning based on fuzzy rule base De-fuzzification generates control output, dynamically adjusts the power of the sugar heating heater; wherein the temperature deviation of the sugar solution is calculated by the following steps: acquiring a set value of the temperature of the sugar solution and a measured value of the temperature of the sugar solution at the current time ; Calculating temperature deviation of sugar solution ; viscosity change rate The viscosity change rate is calculated by the following steps: The viscosity instrument is used to collect the viscosity data of the sugar liquid at a preset frequency; Take the viscosity data at the most recent time point, fit a linear trend, and take the derivative, as follows: ； wherein, is the time at the time point; is the average time at the time points; is the viscosity value at the time point; is the average viscosity at the time points; In addition, the neural network enhanced fuzzy control algorithm is adopted, comprising the following steps: An adaptive neuro-fuzzy inference model is constructed, whose input layer receives the temperature deviation of sugar solution and the rate of change of viscosity of sugar solution ; The first hidden layer of the adaptive neural fuzzy inference model performs fuzzification of the input variables, and Mapping to a fuzzy set, where each input variable corresponds to 3-5 Gaussian membership functions, with the following functional form: ； wherein, represents an input variable or ; is the center point of the th membership function; is the width of the th membership function; is the membership degree, represents the membership degree of the variable to the fuzzy set . The second hidden layer of the adaptive neuro-fuzzy inference model implements the antecedent part of the fuzzy rules, each node corresponds to a fuzzy rule, and the output is the rule firing strength : ； wherein, represents the first rule; is the triggering strength of the first rule; and correspond to the fuzzy sets of temperature deviation and viscosity change rate, respectively. The third hidden layer of the adaptive neuro-fuzzy inference model implements rule normalization, calculating normalized firing strengths : ； wherein, is the normalized trigger strength of the i-th rule; is the normalized trigger strength of the i-th rule; is the total number of fuzzy rules; is the trigger strength of the i-th rule; is the trigger strength of the i-th rule; The fourth hidden layer of the adaptive neural fuzzy inference model performs rule consequent calculation, and each rule corresponds to a linear function: ； wherein , , is a consequent parameter of the rule of clause . Adaptive neuro-fuzzy inference model output layer calculates final control variable : ； wherein, is the total number of fuzzy rules; The adaptive neuro-fuzzy inference model is trained using historical production data, with the training objective being to minimize a loss function : ； wherein, is the expert experience control quantity; is the actual control quantity calculated by the adaptive neuro-fuzzy inference model according to the current real-time input; is the regularization coefficient, is the number of training samples, denotes the th training sample; The neural network is introduced to optimize the parameters of the PID algorithm and the fuzzy control algorithm in real time. The neural network is trained based on historical production data and real-time sensor data, and is used to predict the optimal control parameters.

2. The method of claim 1, wherein the monitoring and adjusting of the line is performed by a computer system. The incremental PID algorithm is used to dynamically adjust the power of the oil heating heater, comprising the following steps: acquiring a set value of oil temperature , a measured value of oil temperature at a current time , a measured value of oil temperature at a previous time , a measured value of oil temperature at a time before the previous time , and a previous oil heating heater power , calculating an oil temperature deviation at the current time , calculating an oil temperature deviation at the previous time , calculating an oil temperature deviation at the time before the previous time ; Control quantity increment for calculating the power of a boiler heater , the formula is: ； wherein, is a proportional coefficient, indicating the response strength of the system to the current deviation; is an integral coefficient, used to eliminate steady-state error; is a differential coefficient, used to suppress overshoot; According to the control amount increment The oil heating heater power is adjusted, and the current output power of the oil heating heater power is controlled to be .

3. The method of claim 2, wherein the monitoring and adjusting of the line is performed by a computer system. Adopting neural network dynamic optimization PID algorithm parameter , and comprising the steps of: Collect historical oil temperature data, environmental temperature data, oil quantity data and production batch information as training data set; Using a multi-layer perceptron neural network for training, the input layer includes current oil temperature deviation , historical oil temperature data, ambient temperature data, and the output layer is the optimized 、 and ; The neural network training adopts back propagation algorithm, and the loss function is mean square error, and the formula is: ； wherein, is a loss value, is a number of samples, is a true value of a th sample, is a neural network predicted value of a th sample; Deploy the trained neural network, input the current sensor data in real time, and output adaptive PID parameters.

4. The method of claim 1, wherein the monitoring and adjusting of the line is performed by a computer system. It also includes an online learning mechanism: Deploy a double neural network architecture, including a main neural network and a shadow neural network; The main neural network performs real-time control, and the shadow neural network continuously learns the latest production data; Every 24 hours, compare the performance indicators of the main neural network and the shadow neural network. When the mean square error of the shadow neural network is more than 10% lower than that of the main neural network, automatically switch the network; The online learning adopts momentum gradient descent algorithm, and the update formula is: ； wherein, denote model parameters; , , denote model parameters after , and iterations, respectively; is a learning rate; is a momentum coefficient; is a gradient.

Images