A coating machine oven monitoring system and method

Abstract

The present invention relates to the technical field of temperature monitoring, and specifically to a system and method for monitoring the oven of a coating machine, the method comprising: obtaining the oven temperature at the last moment; according to the oven temperature, using the best strategy of the current oven environment obtained in advance, obtaining the PID control parameters in the current oven; according to the PID control parameters in the current oven, calculating the corresponding control output signal, and sending the new control output signal to the heating system, and the PLC controller controls the working state and power of the heating system to monitor the temperature of the oven. That is, the solution of the present invention can more accurately control the temperature in the oven to realize the monitoring of the oven of the coating machine.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature monitoring, and more specifically, to a baking oven monitoring system and method for a coating machine. Background Art

[0002] The thousand-layer cake coating and baking machine includes a grouting machine, a grouting machine transmission part, a heating system and a control system, etc. The grouting machine adopts a large gear transmission, which causes less damage to the bubbles in the cake paste, and the grouting machine transmission part adopts seamless steel rails to make each layer of cake more uniform; the heating system adopts far-infrared heating tubes to make baking more uniform; the control system controls the transmission, speed and temperature to ensure the quality of the cake.

[0003] Among them, one of the key factors affecting the quality of cakes is temperature. The control system of the existing crepe cake baking machine usually adopts a closed-loop feedback adjustment system, such as a PLC control system, that is, the temperature sensor detects the oven temperature and converts the temperature into an electrical signal and transmits it to the PLC. The PLC sets different control strategies as needed, such as proportional control, proportional-integral-differential (PID) control, etc., to determine how to adjust the output signal based on the temperature input signal to achieve the desired temperature target. The PLC uses a specific control algorithm to calculate and adjust the temperature internally, calculates the corresponding control output signal based on the measured temperature signal, and sends it to the heating system. The PLC controls the working status and power of the heating system; the PLC monitors and feedback controls the temperature, continuously monitors the temperature value through real-time communication with the temperature sensor, and makes adjustments based on the measurement results to achieve precise temperature control.

[0004] Generally speaking, the proportional-integral-differential (PID) control is usually used to control the ambient temperature, that is, the temperature of the system is adjusted through the combination of proportional (P), integral (I) and differential (D) control.

[0005] However, the parameters of the PID algorithm are difficult to adjust, and it is necessary to rely on experience and repeated experiments to find the optimal parameter combination, which is often time-consuming and cumbersome. In addition, static PID parameters are difficult to adapt to changing temperature environments and have poor robustness.

[0006] Therefore, it is particularly important to quickly and accurately control the temperature of the layer cake baking machine environment. Summary of the invention

[0007] The purpose of the present invention is to propose a coating machine oven monitoring system and method to solve the problems of poor robustness of existing temperature control and cumbersome adjustment process; to this end, the present invention provides solutions in the following two aspects.

[0008] In a first aspect, the present invention provides a method for monitoring a coating machine oven, comprising:

[0009] Get the temperature inside the oven at the last moment;

[0010] According to the temperature in the oven, the PID control parameters in the current oven are obtained by using the best strategy of the current oven environment obtained in advance;

[0011] According to the current PID control parameters in the oven, the corresponding control output signal is calculated, and the new control output signal is sent to the heating system. The PLC controller controls the working state and power of the heating system to realize the temperature monitoring of the oven.

[0012] In the above scheme, the temperature in the oven of the coating machine can be monitored, and the temperature in the current oven can be automatically adjusted through the best strategy obtained in advance, so as to realize the temperature monitoring of the oven.

[0013] In one embodiment, the process of obtaining the optimal strategy is as follows:

[0014] Determine the state space, action space and action strategy function in the Monte Carlo method; the state space is the characteristic value of the PID control curve in the historical oven extracted; the action space is , , , is the action adjustment step size, P, I, and D are the proportion, integral, and differential in the control system respectively; the action strategy function is positively correlated with the current weight and the behavior value function of the current state respectively; the behavior value function is the average value of the value function of the current state in multiple complete state sequences; the current weight for: , ; Indicates state action pair < > The weight of a complete state sequence, is the nth state, , Status The corresponding action;

[0015] The action sequence corresponding to the maximum action strategy function is selected as the optimal strategy.

[0016] In the above scheme, the state space of the improved reinforcement learning model is characterized by obtaining the control curve of the environment in the oven, and when the policy function is subsequently constructed, multiple complete state sequences can be obtained by analyzing the historical data in the oven, and the policy function is constructed based on the value functions of the multiple complete state sequences and the updated weights, and the optimal policy function is obtained. Compared with the prior art, the introduction of weights can take into account the global changes of the action sequences in the multiple complete state sequences, and then obtain the global optimal strategy, which is beneficial to improve the accuracy of the subsequent oven temperature acquisition.

[0017] In one embodiment, the action strategy function is:

[0018] ;

[0019] in, For the The current weight of a complete state sequence, For state action pair > The behavior value function of a complete state sequence is is the proportional coefficient, which is used to adjust the important ratio of weight and value function, where k is greater than or equal to 2, argmax() is the function to find the maximum independent variable, For the multiplication sign.

[0020] In the above scheme, the behavior value function and The parameter of weight after the iteration stops can take into account the changes in states and actions corresponding to different complete state sequences, and provide a data basis for obtaining the best strategy.

[0021] In one embodiment, the process of acquiring the complete state sequence is as follows:

[0022] Based on the initial state and the terminal state, the actual environmental interaction is simulated multiple times to obtain multiple complete state sequences; the environmental interaction is: starting from the initial state, randomly selecting any action in the action space, the individual interacts with the environment until the terminal state, and a complete state sequence is obtained; each complete state sequence corresponds to a set of action sequences and value functions corresponding to the terminal state; among them, the initial state is set to the state corresponding to the actual temperature in the historical oven, and the terminal state is set to the state corresponding to the target temperature reached by the environment in the historical oven.

[0023] In the above scheme, a complete state sequence is used as the path for regulating the temperature in the oven from the initial state to the end state. Since there are multiple actions in the action space, there are multiple combinations of actions in the process from the initial state to the end state, that is, there are multiple paths. One path corresponds to a group of actions, and the value function of a complete state sequence in the above is the value function updated after each complete state sequence ends. There is no need to update the value function by feeding back an action, which improves the efficiency of obtaining the value function.

[0024] In one embodiment, in a complete state sequence, the update rule of the value function is as follows:

[0025] Set the initial value function and initialize it;

[0026] According to the set reward value function, the corresponding reward value of the agent each time it selects an action is obtained until the cumulative sum of the reward values ​​is maximized, the complete state sequence is obtained, and the initial value function is updated; the specific update is:

[0027] ;

[0028] in, is a discount factor used to attenuate the importance of future rewards, For state action pair > value function, For state action pair >, where the reward function is the ratio of the difference between the current state and the next state to the current state, For state action pair > value function, Status The corresponding action, is the nth state, It is the n+1th state.

[0029] In one embodiment, the state space is In the temperature control system in the oven, the agent obtains the characteristic value from the oven ,in Status The corresponding characteristic value of the control curve is When action a is executed, the state is obtained The corresponding characteristic value of the control curve is ; N is the total number of states.

[0030] In one embodiment, the characteristic value of the control curve is: ;

[0031] Among them, t is the adjustment time required for the output of PID control to change from the initial value to the set value, and h is the overshoot, which refers to the ratio of the maximum value of the difference between the output temperature and the target value during the regulation process to the regulation target value; is the maximum value of the output temperature, is the minimum value of the output temperature, is the target value for regulation, is the oscillation frequency, which refers to the frequency at which the output oscillates before reaching a stable state. , b refers to the number of oscillations before the output reaches a stable state, For the multiplication sign.

[0032] When using the PID algorithm for temperature control, different control ratios will result in different control curves, and the control curve can intuitively reflect the temperature change process in the oven when the current PID parameters are used. Therefore, analyzing the control curve and extracting the characteristic value of the control curve is helpful to determine whether the current PID parameters are suitable.

[0033] In a second aspect, the present invention further provides a coating machine oven monitoring system, comprising:

[0034] processor;

[0035] The memory stores computer instructions for monitoring the oven of a coating machine. When the computer instructions are executed by the processor, the system executes the method for monitoring the oven of a coating machine in the first aspect.

[0036] The beneficial effects of the present invention are:

[0037] The solution of the present invention can adaptively find suitable PID parameters in the control system and control the oven temperature by using the environmental reinforcement learning model, thereby improving the accuracy of temperature monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0039] Figure 1 A flowchart of a method for monitoring a coating machine oven in this embodiment is schematically shown;

[0040] Figure 2 Schematic diagrams schematically showing multiple complete state sequences in this embodiment;

[0041] Figure 3 A schematic diagram of a control curve is shown schematically;

[0042] Figure 4 The structural block diagram of a coating machine oven monitoring system in this embodiment is schematically shown. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0044] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0045] The scenario targeted by the present invention is the monitoring of the ambient temperature in the oven of the existing baking machine for making cakes. Since the quality of the baked cake is affected by the temperature, it is necessary to monitor the environment in the oven. The existing oven temperature control system usually uses a PID controller to ensure temperature control. The controller uses a feedback mechanism to keep the temperature stable. However, the parameters of the PID algorithm need to rely on experience and repeated experiments to find the optimal parameter combination, which is often time-consuming and cumbersome.

[0046] Based on the above technical problems, the present invention proposes a baking oven monitoring system and method, which can more accurately control the temperature of the oven when baking cakes by obtaining the optimal temperature adjustment strategy.

[0047] Specifically, taking a coating machine oven as an example, a coating machine oven monitoring method of the present invention is introduced.

[0048] The coating baking oven of this embodiment includes a grouting machine, a grouting machine transmission part, a heating system, a control system, etc. The control system is a PLC controller, which is used to execute a control algorithm to monitor the temperature in the baking oven.

[0049] like Figure 1 As shown, a coating machine oven monitoring method in this embodiment includes the following steps:

[0050] In step S1, the temperature in the oven at the last moment is obtained; according to the temperature in the oven, the best strategy for the current environment in the oven obtained in advance is adopted to obtain the PID control parameters in the current oven. Specifically, in this embodiment, the temperature in the oven at the last moment is obtained to determine the best strategy, that is, the oven environment data at the historical moment is used to determine the best strategy.

[0051] In this embodiment, the Monte Carlo method is used to obtain the optimal strategy for the environment inside the oven.

[0052] Specifically, the Monte Carlo method is a learning strategy in reinforcement learning that enables the agent to select the action that is likely to get the maximum reward in the future in the current state. For example, at the nth step, the agent is in state Then take action , the environment rewards the agent , and the state of the agent becomes The Monte Carlo method is a model-free reinforcement learning method in reinforcement learning. It estimates the true value of the state by sampling several complete state sequences (Trajectory / episode) with complete experiences. The so-called complete experience means that this sequence must reach the end. For example, if an initial state and a terminal state are determined, starting from the initial state, any action in the action space is randomly selected, and the individual interacts with the environment until the terminal state, a complete state sequence can be obtained.

[0053] In this embodiment, the state corresponding to the actual temperature in the oven is used as the initial state, and the state corresponding to the target temperature to be reached by the environment in the oven is used as the final state. Figure 2 As shown, there are multiple groups of action sequences from a set initial state to a terminal state, each group of action combinations is different, and one group of action combinations corresponds to a complete state sequence, so that multiple complete state sequences can be obtained. Figure 2 The initial state and the end state of each complete state sequence in are unchanged, that is, the starting point and the end point in different complete state sequences are the same.

[0054] In this embodiment, the process of obtaining the best strategy includes the following steps:

[0055] Step S11, determining the state space, action space and action strategy function in the Monte Carlo method. In this embodiment, the Monte Carlo method is used to perform multiple simulations, and the value of the strategy function is evaluated according to the simulation results to obtain the best strategy.

[0056] Specifically, the specific settings of the Monte Carlo method are as follows:

[0057] (1) Environmental information: temperature inside the oven.

[0058] (2) State space : The characteristic value of the control curve corresponding to the current PID ratio It should be noted that there is a one-to-one mapping relationship between the eigenvalues ​​of the state space and the control curve: For example, the state space In the temperature control system inside the oven, the agent obtains the characteristic values ​​from the environment inside the oven ,in Status The corresponding characteristic value of the control curve is Execute action when , get the status The corresponding characteristic value of the control curve is ; N is the total number of states, so in state Execute action when The mapping of states and actions is called a state-action pair. >.

[0059] The characteristic value of the control curve in the above It is obtained according to the PID control curve, specifically:

[0060] ;in is the adjustment time, which refers to the time required for the output to change from the initial value to the target temperature value. The smaller the adjustment time t, the better the response speed of the regulation; h is the overshoot, which refers to the ratio of the maximum amplitude of the temperature output exceeding the regulation target value during the regulation process to the regulation target value. The calculation formula is: ; is the maximum value of the output temperature, is the minimum value of the output temperature, is the target value for regulation, see Figure 3 The smaller the overshoot h is, the better the control capability is. In this embodiment, the overshoot h needs to be maintained within 5%. is the oscillation frequency, which refers to the frequency at which the output oscillates before reaching a stable state, that is, ; b refers to the number of oscillations before the output reaches a stable state; the smaller the oscillation frequency, the better the stability of the system. Among them, when using the PID algorithm for temperature control, different control ratios will obtain different control curves, and the control curve can intuitively reflect the process of ambient temperature changes when using the current PID parameters. Therefore, the control curve is analyzed and the characteristic value of the control curve is extracted to determine whether the current PID parameters are suitable. Among them, the smaller the characteristic value Z of the control curve, the better the control stability under the PID proportional control and the stronger the control ability. Therefore, the characteristic value is used to reflect the control situation.

[0061] (3) Action Space : 6 actions: , , ,in Adjust the step length for the action and take the experience value =0.2.

[0062] (4) Set reward value : ; Among them, the reward value function Indicates in status Execute a specific action The rewards for timely feedback. Status The corresponding eigenvalues ​​in the state Execute action when , the obtained status , the corresponding eigenvalue is .

[0063] The above reward value function represents the characteristic value gain during state transition. Since the smaller the characteristic value of the control curve, the better the control is, The bigger the action The greater the gain, the more likely it is to make the state Moving towards a better state ;like Less than zero, indicating action The system control ability decreases, and the state conversion reward is negative.

[0064] (5) Setting value function and weight : Used to evaluate the expected return of performing an action in a given state, i.e., the value function Indicates in status When you select Action The expected value of the sum of all possible future rewards from then on to the final state; weight Indicates in status Select action The weight size.

[0065] The reason for setting the median function is that since the calculation of the reward function only depends on the information of the current state and the next state, only local information is obtained. If the reward function is used as a strategy to guide the selection of each action, it is possible to fall into the local optimal solution. Therefore, in order to obtain the global optimal solution, it is necessary to set the value function and weights to characterize the global reward accumulation value.

[0066] (6) Strategy function :Strategy According to the current status , calculate the action that should be chosen , specifically: ;

[0067] in, For the The current weight of a complete state sequence, For state action pair > The behavior value function of a complete state sequence is is the proportional coefficient, which is used to adjust the important ratio of weight and value function. is greater than or equal to 2, argmax() is the function that maximizes the independent variable, For the multiplication sign.

[0068] The strategy function in the above Based on the current state in a sequence of multiple complete states , calculate the action that should be chosen .

[0069] Of course, as another implementation method, the strategy function in the above may also be:

[0070] ;

[0071] in, For the The current weight of a complete state sequence, For state action pair > The behavior value function of a complete state sequence is is the multiplication sign, is the proportional coefficient, which is used to adjust the important ratio of weight and value function, and is taken according to the empirical value. . represents the sum of the action strategy values ​​of other complete state sequences except the current complete state sequence. Use a smaller random number to avoid a denominator of 0, such as a number in the range of 0.001 to 0.1.

[0072] The above strategy function is in the current state Select an action , which depends not only on the value function but also on the weight. Combining the behavior value function and the weight, the strategy function finds an optimal path that converges as quickly as possible (that is, the action sequence in the optimal strategy), and the end point of the action sequence is the end state under the control of the optimal PID parameters.

[0073] Since the policy function defines the action mode of the agent, telling the agent how to react when facing different states, the policy function used in this embodiment has a faster learning strategy convergence speed than the epsilon-greedy learning strategy used in the traditional reinforcement learning algorithm, and can handle continuous action space.

[0074] The behavior value function in this embodiment for: ;in, For the The value function of a complete state sequence, K is the total number of complete state sequences.

[0075] The behavior value function in the above is used to reflect the state-action pair in different complete state sequences. >It is possible to obtain the central tendency of the expected value of the reward and use it to guide the policy function to predict the best action.

[0076] It should be noted that each complete state sequence corresponds to a value function, namely The state-action pairs in a complete state sequence < >The cumulative value of possible future rewards is For the behavior value function, after obtaining each complete state sequence, the average of all the fragments before the current complete state sequence and the current value function is used as the current state action pair. > function.

[0077] The weight update rules in this embodiment are as follows:

[0078] ; ;

[0079] in, Indicates state action pair < > The weight of a complete state sequence, is the nth state, , a is the state Corresponding actions, weights Indicates state action pair < > The weight of a complete state sequence. In particular, after the first iteration, the action pair on this segment < The weight value of > is the inverse of the fragment length, which is .

[0080] The above weights are the state-action pairs at the kth iteration < >Fragment 1 to Fragment The reciprocal of the mean length of .

[0081] Among them, since the shorter the length of the complete state sequence, the faster the convergence and the better the control ability of the system, the action selection strategy tends to select the action with the shorter expected length of the complete state sequence. Therefore, using the sequence length as the weight, compared with the direct PID adjustment which can only guarantee the convergence of the characteristic curve, the scheme of the present invention can combine the weight with the value function of the characteristic curve, and can ensure the stability of the PID adjustment effect.

[0082] The current weight in the above scheme is obtained by updating the weight after the last iteration is stopped. Therefore, the weight can represent the global situation of the state and action changes from the first iteration to the current iteration.

[0083] Step S12, select the action sequence corresponding to the maximum action strategy function as the optimal strategy. In this embodiment, based on the updated weight and behavior value function after obtaining the current complete state sequence, the strategy function is calculated. , the action sequence corresponding to the maximum value of the policy function is taken as the optimal policy.

[0084] Specifically, in this embodiment, based on obtaining the current complete state sequence, the weight and behavior value function are updated, and the specific process of calculating the strategy function is as follows:

[0085] Step 121, initialize the state space , initialize the value function and weight, let, ; = 0. Among them, the initialization state space Specifically, it is obtained through the initial control parameters of the current PID. Taking the initial control parameters (0.5, 0.5, 0.5) as an example, the initial characteristic value can be obtained according to the control curve corresponding to the initial control parameters: , taking the initial eigenvalue as the initial state .

[0086] Step 122, for the first complete state sequence: before obtaining the complete state sequence, each action selected by the agent corresponds to a reward value, until the accumulated reward value is the maximum and reaches the terminal state, the first iteration stops, the complete state sequence is obtained, and the value function and weight are updated to obtain the updated value function and weight. For example, Figure 2 As shown, a complete state sequence is ;in is the initial state, is the final state. The initial state and the final state correspond to the initialization parameters in the oven and the target parameters of the final control, respectively. The parameters are P, I and D.

[0087] Step 123, keep the initial state unchanged, initialize the value function, continue to iterate the second complete state sequence, update the value function and the weight of the first iteration after the second iteration stops; at this time, take the average of the value function of the first iteration and the value function of the second iteration as the behavior value function, and calculate the value of the strategy function.

[0088] Step 124, repeat step 123, perform the third iteration of the complete state sequence, update the value function and the weight of the second iteration, obtain a new behavior value function, and calculate the value of the policy function.

[0089] Step 125, perform multiple iterations until the set number of iterations is reached, and obtain the final behavior value function and the weight corresponding to the current complete state sequence, and then calculate the value of the policy function of the current complete state sequence, and select the action sequence that maximizes the policy function as the optimal policy.

[0090] The length of the complete state sequence in the above is equal to the number of state transitions n in the process from the initial state to the terminal state.

[0091] For an iteration of a complete state sequence, the rule for updating the value function after each iteration stops is as follows:

[0092] ;

[0093] in, is a discount factor used to attenuate the importance of future rewards, For state action pair >Reward value, For state action pair >value function.

[0094] The discount factor in the above The value range is usually between 0 and 1. When it is close to 1, the influence of future rewards is greater; When it is close to 0, only immediate rewards are considered. Here, experience value is taken .

[0095] The stopping condition for each iteration in the above is: setting the accumulated value matrix , record status The accumulated value of rewards obtained: ;like: is the maximum accumulated value in a complete state sequence, is the state where the iteration stops; otherwise, it does not stop.

[0096] The maximum accumulated value in the above can be set by setting a size of The sliding window (here The empirical value is 4), which is used to find the stop state in a complete state sequence (episode). For Window The iteration stops when the maximum value of Not for window The window slides backwards and n'=n+1.

[0097] The solution of the present invention is to deploy the above-mentioned improved optimal strategy into the PLG control system, and input it as the PID control parameters in the initial state space (the characteristic values ​​of the control curve correspond to the PID control parameters), output the optimal strategy (action sequence), and based on the obtained action sequence of the optimal strategy, realize the regulation of the corresponding target value in the control curve, thereby obtaining the PID control parameters in the current oven.

[0098] In step S2, the corresponding control output signal is calculated according to the PID control parameters in the current oven, and the new control output signal is sent to the heating system. The PLC controller controls the working state and power of the heating system to monitor the temperature of the oven.

[0099] The specific regulation process is to obtain the PID control parameters of each state in the current oven according to the action sequence in the optimal strategy, adjust the control curve of the oven, and calculate the corresponding control output signal to stabilize the temperature in the oven at the target temperature.

[0100] It should be noted that, when making the above adjustments, since the specific process of calculating the control output signal is prior art, it will not be described in detail here.

[0101] It should be noted that when the target temperature value corresponding to the terminal state is still changing, it is necessary to find the best strategy again.

[0102] The solution of the present invention characterizes the state space of the improved reinforcement learning model by acquiring the control curve of the environment in the oven, and when subsequently constructing the policy function, multiple complete state sequences are obtained by simulating the process of the same initial state and end state multiple times, and the policy function is evaluated based on the multiple complete state sequences, and the action sequence corresponding to the maximum value of the policy function is used as the optimal policy. Since the weight information of the action changes of multiple complete state sequences is introduced into the optimal strategy, it is possible to analyze from the global information when evaluating the policy function. Compared with the prior art, it avoids falling into the situation of local optimal solution and improves the accuracy of temperature control of the oven.

[0103] The present invention also provides a coating machine oven monitoring system. Figure 4As shown, the system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned baking oven monitoring method of the coating machine according to the present invention is implemented.

[0104] The system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface, whose configuration and functions are known in the art and thus will not be described in detail here.

[0105] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0106] In the description of this specification, “plurality” means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0107] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for monitoring a coating machine oven, characterized in that: include: Get the temperature inside the oven at the last moment; According to the temperature in the oven, the PID control parameters in the current oven are obtained by using the best strategy of the current oven environment obtained in advance; the process of obtaining the best strategy is as follows: Determine the state space, action space and action strategy function in the Monte Carlo method; the state space is the characteristic value of the PID control curve in the oven extracted; the action space is , , , is the action adjustment step size, P, I, and D are the proportion, integral, and differential in the control system respectively; the action strategy function is positively correlated with the current weight and the behavior value function of the current state respectively; the behavior value function is the average value of the value function of the current state in multiple complete state sequences; the current weight for: , ; Indicates state action pair< > The weight of a complete state sequence, is the nth state, , Status The corresponding action; Select the action sequence corresponding to the maximum action strategy function as the optimal strategy; The characteristic value of the control curve is: ; Among them, t is the adjustment time required for the output of PID control to change from the initial value to the set value, and h is the overshoot, which refers to the ratio of the maximum value of the difference between the output temperature and the target value during the regulation process to the regulation target value; is the maximum value of the output temperature, is the minimum value of the output temperature, is the target value for regulation, is the oscillation frequency, which refers to the frequency at which the output oscillates before reaching a stable state. , b refers to the number of oscillations before the output reaches a stable state, is the multiplication sign; According to the current PID control parameters in the oven, the corresponding control output signal is calculated, and the new control output signal is sent to the heating system. The PLC controller controls the working state and power of the heating system to realize the temperature monitoring of the oven.

2. The method for monitoring the oven of a coating machine according to claim 1, characterized in that: The action strategy function is: ； in, For the The current weight of a complete state sequence, For state action pair > The behavior value function of a complete state sequence is is the proportional coefficient, which is used to adjust the important ratio of weight and value function, where k is greater than or equal to 2, argmax() is the function to find the maximum independent variable, For the multiplication sign.

3. The method for monitoring the oven of a coating machine according to claim 2, characterized in that: The acquisition process of the complete state sequence is: Based on the initial state and the terminal state, the actual environmental interaction is simulated multiple times to obtain multiple complete state sequences; the environmental interaction is: starting from the initial state, randomly selecting any action in the action space, the individual interacts with the environment until the terminal state, and a complete state sequence is obtained; each complete state sequence corresponds to a set of action sequences and value functions corresponding to the terminal state; among them, the initial state is set to the state corresponding to the actual temperature in the oven, and the terminal state is set to the state corresponding to the target temperature to be reached by the environment in the oven.

4. The method for monitoring the oven of a coating machine according to claim 2, characterized in that: In a complete state sequence, the update rule of the value function is as follows: Set the initial value function and initialize it; According to the set reward value function, the corresponding reward value of each action selected by the agent is obtained until the cumulative sum of the reward values ​​is maximized, the complete state sequence is obtained, and the initial value function is updated; the specific update is: ； in, is a discount factor used to attenuate the importance of future rewards, For state action pair > value function, For state action pair >, where the reward function is the ratio of the difference between the current state and the next state to the current state, For state action pair > value function, Status The corresponding action, is the nth state, It is the n+1th state.

5. The method for monitoring the oven of a coating machine according to claim 1, characterized in that: The state space is In the temperature control system in the oven, the agent obtains the characteristic value from the oven ,in Status The corresponding characteristic value of the control curve is When action a is executed, the state is obtained The corresponding characteristic value of the control curve is ; N is the total number of states.

6. A coating machine oven monitoring system, characterized in that: include: processor; A memory storing computer instructions for monitoring a coating machine oven. When the computer instructions are executed by the processor, the system executes the coating machine oven monitoring method according to any one of claims 1 to 5.

Images