Simulation device and method
The simulation device improves prediction accuracy by training a parameter estimation model with pseudo-generated data to address uncertainties in parameter settings, ensuring precise future process value predictions.
Patent Information
- Application Number
- JP2020146634
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-09-01
- Publication Date
- 2024-11-27
- Estimated Expiration
- 2040-09-01
AI Technical Summary
Existing simulation devices suffer from decreased prediction accuracy due to uncertainties in parameter settings, which cannot be directly measured.
A simulation device and method that incorporates a parameter estimation model trained using pseudo-generated data to improve prediction accuracy by learning the relationship between input and output values of a simulation model, allowing for accurate parameter setting.
Enhances prediction accuracy by calculating and setting actual parameter values in the simulation model, thereby improving the reliability of future process value predictions.
Smart Images

Figure 0007590729000001 
Figure 0007590729000002 
Figure 0007590729000003
Abstract
Description
[Technical field]
[0001] The present invention relates to a simulation device and a method for predicting a process value at a future time by using a simulation model that models a target process. [Background technology]
[0002] Conventionally, simulation devices have been used to perform simulations of processes in various plants (see Patent Documents 1 and 2). In an actual plant, operating state information (operating process variable values) can only be obtained from a limited number of installed sensors (pressure sensors, temperature sensors, flow rate sensors, etc.), but a simulation device can obtain process information in any plant device (pipes, heat exchangers, tanks, control valves, etc.), and can be effectively used not only to estimate future operating states but also as an estimation sensor that indirectly measures process variables at any position in the plant.
[0003] An example of the application of a simulation device is the prediction of the discharge pressure, discharge temperature, and motor power consumption for driving the compressor. Such compressors are used in the refrigerant circuits of freezers, and the prediction of the discharge temperature, discharge pressure, and motor power consumption is essential for the design of the refrigerant circuit or for preventive maintenance during operation (particularly, the discrepancy between the predicted power consumption value and the actual value indirectly indicates the change in compressor efficiency, suggesting aging or abnormality of the compressor).
[0004] 9 is a diagram showing the operation of the compressor 100. The compressor 100 pumps out process gas by the driving force of a motor 101. 10 is a diagram for explaining a simulation for predicting the discharge pressure, discharge temperature, and compressor motor power consumption of the compressor 100. In this simulation, the current discharge pressure Y1(t), the current discharge temperature Y2(t), the current compressor motor power consumption Y3(t), the current gas flow rate X1(t), the current suction pressure X2(t), the current suction temperature X3(t), the predicted value X1(t+1) of the gas flow rate at a future time for predicting the discharge pressure, discharge temperature, and compressor motor power consumption, and the compressor efficiency are input to a compressor model 200 that models the compressor 100, and an estimated value Y1(t+1) of the discharge pressure, an estimated value Y2(t+1) of the discharge temperature, and an estimated value Y3(t+1) of the compressor motor power consumption at a future time are obtained.
[0005] In order to estimate the discharge pressure, discharge temperature, and compressor motor power consumption, it is necessary to set the compressor efficiency as an adjustment parameter. However, since the compressor efficiency cannot be measured, there is a possibility that the estimation accuracy of the discharge pressure, discharge temperature, and compressor motor power consumption may decrease depending on the set value of the compressor efficiency. Such parameter setting problems occur not only in compressor simulations but also in other simulations. [Prior art documents] [Patent documents]
[0006] [Patent Document 1] JP 2018-55299 A [Patent Document 2] JP 2019-21032 A Summary of the Invention [Problem to be solved by the invention]
[0007] The present invention has been made to solve the above-mentioned problems, and has an object to provide a simulation device and method that can avoid a decrease in prediction accuracy that is dependent on parameter settings and improve prediction accuracy. [Means for solving the problem]
[0008] The simulation device of the present invention includes a first storage unit configured to store a simulation model that models a target process and a parameter estimation model that models a relationship between an input value and an output value of the simulation model and a parameter value of the simulation model, and a data generation unit configured to generate a plurality of pseudo pairs of a process value and the parameter value to be input to the simulation model while changing at least a parameter value during learning of the parameter estimation model, and a calculation unit configured to calculate an output value of the simulation model for each pair of a process value and a parameter value when the process value generated by the data generation unit is input in a state in which the parameter values generated by the data generation unit are set in the simulation model during learning of the parameter estimation model. the simulation unit is configured to calculate an output value of the simulation model using a current value of a process value acquired from outside as an input when a simulation is executed; a second storage unit is configured to store, when learning the parameter estimation model, the process value and parameter value generated by the data generation unit and an output value of the simulation model for the process value and parameter value as learning data; a learning unit configured to learn the parameter estimation model using the learning data; and a parameter setting unit configured to calculate a parameter value that is an output of the parameter estimation model using a current value of the process value and a past actual value of the process value as an input before a simulation is executed, and set the parameter value in the simulation model. the simulation model receives as input a current value of a process value to be predicted and a predicted value at a future time of a related process value that affects the process value to be predicted, and outputs the predicted value of the process value to be predicted at a future time, the data generating unit pseudo-generates a plurality of sets of the current value of the process value to be predicted, the predicted value of the related process value, and the parameter value while varying at least the parameter value, the simulation unit calculates an output value of the simulation model when the current value of the process value to be predicted and the predicted value of the related process value generated by the data generating unit are input during learning of the parameter estimation model, and calculates an output value of the simulation model when the current value of the process value to be predicted and the predicted value of the related process value obtained by the outside are input during execution of a simulation, and the parameter setting unit calculates a parameter value which is an output of the parameter estimation model when the current value of the process value to be predicted, the current value of the related process value, and a past actual value of the process value to be predicted, which are obtained by the outside, before execution of a simulation, and sets the parameter value in the simulation model. It is characterized by the above.
[0011] Further, a simulation method of the present invention includes a first step of generating a plurality of pseudo pairs of a process value and a parameter value to be input to a simulation model while varying at least a parameter value during learning of a parameter estimation model that models a relationship between an input value and an output value of a simulation model and a parameter value of the simulation model; a second step of calculating an output value of the simulation model for each pair of a process value and a parameter value when the process value generated in the first step is input in a state in which the parameter value generated in the first step is set in the simulation model during learning of the parameter estimation model; The method includes a third step of storing, during learning, the process value and parameter value generated in the first step and an output value of the simulation model for these process value and parameter value as learning data; a fourth step of learning the parameter estimation model using the learning data; a fifth step of calculating, before execution of a simulation, a parameter value which is an output of the parameter estimation model using a current value of the process value acquired from outside and a past actual value of the process value as input, and setting the parameter value in the simulation model; and a sixth step of calculating, during execution of a simulation, an output value of the simulation model using the current value of the process value as input. the simulation model receives as input a current value of a process value to be predicted and a predicted value at a future time of a related process value that affects the process value to be predicted, and outputs the predicted value of the process value to be predicted at a future time; the first step includes a step of pseudo-generating a plurality of sets of the current value of the process value to be predicted, the predicted value of the related process value, and the parameter value while varying at least the parameter value; the second step includes a step of calculating an output value of the simulation model when the current value of the process value to be predicted and the predicted value of the related process value generated in the first step are input during learning of the parameter estimation model; the fifth step includes a step of calculating a parameter value that is an output of the parameter estimation model using as input the current value of the process value to be predicted, the current value of the related process value, and a past actual value of the process value to be predicted, which are acquired from outside, before execution of a simulation, and setting the parameter value in the simulation model; and the sixth step includes a step of calculating an output value of the simulation model using as input the current value of the process value to be predicted and the predicted value of the related process value, which are acquired from outside, during execution of a simulation. It is characterized by the above. Effect of the Invention
[0014] According to the present invention, a parameter estimation model is trained using pseudo-generated process values, parameter values, and a simulation model, and before executing a simulation, current values and past actual values of the process values are input into the trained parameter estimation model to calculate parameter values, and these parameter values are set in the simulation model, thereby improving the prediction accuracy of the simulation. [Brief description of the drawings]
[0015] [Figure 1]FIG. 1 is a block diagram showing a configuration of a simulation device according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a flowchart illustrating the learning operation of the simulation device according to the embodiment of the present invention. [Diagram 3] FIG. 3 is a diagram for explaining the process of generating learning data in the embodiment of the present invention. [Figure 4] FIG. 4 is a flowchart illustrating a simulation operation of the simulation device according to the embodiment of the present invention. [Diagram 5] FIG. 5 is a diagram for explaining the process of parameter value calculation using the parameter estimation model in the embodiment of the present invention. [Figure 6] FIG. 6 is a diagram for explaining the process of a simulation in the embodiment of the present invention. [Figure 7] FIG. 7 is a diagram for explaining a specific example of a simulation in the embodiment of the present invention. [Figure 8] FIG. 8 is a block diagram showing an example of the configuration of a computer that realizes a simulation apparatus according to an embodiment of the present invention. [Figure 9] FIG. 9 is a diagram illustrating the operation of the compressor. [Figure 10] FIG. 10 is a diagram for explaining a simulation for predicting the discharge pressure and discharge temperature of the compressor and the compressor motor power consumption. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0016] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a block diagram showing a configuration of a simulation device according to an embodiment of the present invention. The simulation device includes a model storage unit 1 (first storage unit), a simulation unit 2, a data accumulation unit 3, a data generation unit 4, a learning data storage unit 5 (second storage unit), a learning unit 6, and a parameter setting unit 7.
[0017] The model storage unit 1 stores a simulation model 10, which is a mathematical model constructed using software that models a target process, and a parameter estimation model 11, which is a mathematical model constructed using software that models the relationship between the input values and output values of the simulation model 10 and the parameter values of the simulation model 10.
[0018] First, there will be described the operation of learning the parameter estimation model 11. Fig. 2 is a flow chart for explaining the learning operation of the simulation device of this embodiment. The data generator 4 generates a process value Y1 to be input to the simulation model 10. p (t),,,Ym p (t), X1 p (t),,,Xn p (t), X1 p (t+1),,,Xn p (t+1) and the parameter value α of simulation model 10 p A plurality of pairs of the process value and the parameter value are generated while varying at least the parameter value (Step S100 in FIG. 2).
[0019] Process value Y1 p (t),,,Ym p (t) (m is an integer equal to or greater than 1) is a pseudo-generated current value of the process value to be predicted. In the compressor simulation example, it is the current values of the discharge pressure, discharge temperature, and compressor motor power consumption. p (t),,,Xn p (t) (n is an integer equal to or greater than 1) is a pseudo-generated current value of a related process value that affects the process value to be predicted. In the compressor simulation example, it is the current values of the gas flow rate, suction pressure, and suction temperature. p (t+1),,,Xn p (t+1) is a pseudo-generated predicted value of the relevant process value at a future time, which is the predicted value of the gas flow rate in the compressor simulation example. pis a temporary value of a parameter value set in the simulation model 10, and in the compressor simulation example, is the compressor efficiency.
[0020] The process value may be varied within a predefined input range, and the parameter value may be varied within a predefined parameter range.
[0021] The data generating unit 4 generates the process value Y1 p (t),,,Ym p (t), X1 p (t),,,Xn p (t), X1 p (t+1),,,Xn p (t+1) and parameter value α p A plurality of sets of the process value and the parameter value are generated while varying only the parameter value or while varying both the process value and the parameter value.
[0022] The simulation unit 2 inputs a parameter value α p When the process value Y1 p (t),,,Ym p (t), X1 p (t),,,Xn p (t), X1 p (t+1),,,Xn p The output values Y1(t+1), .., Ym(t+1) of the simulation model 10 when (t+1) is input are calculated for each pair of process value and parameter value (step S101 in FIG. 2). In the compressor simulation example, the output values Y1(t+1), .., Ym(t+1) are predicted values of the discharge pressure, discharge temperature, and compressor motor power consumption at a future time.
[0023] The learning data storage unit 5 stores the process value Y1 p (t),,,Ym p (t), X1 p (t),,,Xn p (t), X1 p (t+1),,,Xn p(t+1) and parameter value α p and the corresponding output values Y1(t+1),,,Ym(t+1) of the simulation model 10 are stored as learning data (FIG. 2, step S102). In this way, multiple sets of learning data, each of which is made up of a process value, a parameter value, and an output value, are generated. This process of generating learning data is shown in FIG.
[0024] Next, the learning unit 6 calculates the process value Y1 included in the learning data stored in the learning data storage unit 5. p (t),,,Ym p (t), X1 p (t),,,Xn p (t), X1 p (t+1),,,Xn p (t+1) and output values Y1(t+1),,,Ym(t+1) are input variables of the parameter estimation model 11, and the parameter value α p is used as an output variable of the parameter estimation model 11, and machine learning of the parameter estimation model 11 is performed so as to obtain a target output variable (Step S103 in FIG. 2). This is the end of the learning operation. Note that since the machine learning method is a well-known technique, a detailed description will be omitted.
[0025] 4 is a flow chart for explaining the simulation operation of the simulation device of this embodiment. The data storage unit 3 stores process values Y1, . . . Ym, X1, . . . Xn previously acquired from a target plant.
[0026] Before executing the simulation, the parameter setting unit 7 calculates an output value α of the parameter estimation model 11 when the current values Y1(t), .., Ym(t) of the prediction target process values and the current values X1(t), .., Xn(t) of the related process values obtained from the target plant as shown in FIG. 5, and the past actual values Y1(t-1), .., Ym(t-1) of the prediction target process values and the past actual values X1(t-1), .., Xn(t-1) of the related process values stored in the data storage unit 3 are input (step S200 in FIG. 4).
[0027] That is, the current values X1(t), .., Xn(t) of the related process values, the actual values Y1(t-1), .., Ym(t-1) of the prediction target process values, and the actual values X1(t-1), .., Xn(t-1) of the related process values correspond to input values of the simulation model 10 at a past time point. The current values Y1(t), .., Ym(t) of the prediction target process values correspond to output values of the simulation model 10 at a past time point. The current values of the related process values are the current value X1(t) of the gas flow rate, the current value X2(t) of the suction pressure, and the current value X3(t) of the suction temperature in the compressor simulation example. The actual values of the related process values are the past actual value X1(t-1) of the gas flow rate, the past actual value X2(t-1) of the suction pressure, and the past actual value X3(t-1) of the suction temperature in the compressor simulation example. In the compressor simulation example, the actual values of the process values to be predicted are the actual discharge pressure value Y1(t-1), the actual discharge temperature value Y2(t-1), and the actual compressor motor power consumption value Y3(t-1). In the compressor simulation example, the current values of the process values to be predicted are the current discharge pressure value Y1(t), the current discharge temperature value Y2(t), and the current compressor motor power consumption value Y3(t). Therefore, by inputting these values into the parameter estimation model 11, it is possible to calculate the parameter value α. The time interval between the past time point at which the actual process value was measured and the present time point is equal to the time interval between the present time point and a future time at which the process value is to be predicted.
[0028] The parameter setting unit 7 sets the parameter value α calculated using the parameter estimation model 11 in the simulation model 10 (Step S201 in FIG. 4).
[0029] After setting the parameter value α, the simulation unit 2 acquires the current values Y1(t), . . . Ym(t) of the process value to be predicted and the current values X1(t), . . . Xn(t) of the related process values, and also acquires the predicted values X1(t+1), . . . Xn(t+1) of the related process values. The current values Y1(t), . . . Ym(t), X1(t), . . . Xn(t) can be acquired from sensors in the plant. The predicted values X1(t+1), . . . Xn(t+1) can be acquired from an external prediction system (for example, a weather prediction system). Then, the simulation unit 2 calculates the output values Y1(t+1), . . . Ym(t+1) of the simulation model 10 when the current values Y1(t), . . . Ym(t) of the process value to be predicted, the current values X1(t), . . . Xn(t) of the related process values, and the predicted values X1(t+1), . . . Xn(t+1) of the related process values are input (step S202 in FIG. 4).
[0030] In this way, the predicted values Y1(t+1),,,Ym(t+1) at future times can be obtained as shown in Fig. 6. If Fig. 6 is specifically described for the compressor simulation, it becomes as shown in Fig. 7.
[0031] In this embodiment, a parameter estimation model is trained using pseudo-generated process values, parameter values, and a simulation model, and before executing a simulation, the current values and past actual values of the process values are input into the trained parameter estimation model to calculate parameter values, and these parameter values are set in the simulation model, thereby improving the prediction accuracy of the simulation.
[0032] It is not essential for the present invention to use all the process values described in this embodiment. For example, a configuration that does not require the current value of the related process value may be used. In this case, the simulation model 10 receives the current value of the prediction target process value and the predicted value of the related process value as input. The data generating unit 4 generates a plurality of pseudo sets of the current value of the prediction target process value, the predicted value of the related process value, and the parameter value. The simulation unit 2 calculates an output value of the simulation model 10 when the current value of the prediction target process value and the predicted value of the related process value generated by the data generating unit 4 are input during learning of the parameter estimation model 11. The parameter setting unit 7 calculates an output of the parameter estimation model 11, which receives the current value of the prediction target process value acquired from outside, the current value of the related process value, and the past actual value of the prediction target process value as input.
[0033] The simulation device described in this embodiment can be realized by a computer equipped with a CPU (Central Producing Unit), a storage device, and an interface, and a program that controls these hardware resources. An example of the configuration of this computer is shown in FIG.
[0034] The computer includes a CPU 300, a storage device 301, and an interface device (I / F) 302. A plant management device, a prediction system, and the like are connected to the I / F 302. In such a computer, a program for implementing the simulation method of the present invention is stored in the storage device 301. The CPU 300 executes the processing described in this embodiment in accordance with the program stored in the storage device 301. [Industrial Applicability]
[0035] The present invention can be applied to a process simulation technique. [Explanation of symbols]
[0036] 1...model storage unit, 2...simulation unit, 3...data accumulation unit, 4...data generation unit, 5...learning data storage unit, 6...learning unit, 7...parameter setting unit, 10...simulation model, 11...parameter estimation model.
Claims
1. a first storage unit configured to store a simulation model that models a target process and a parameter estimation model that models a relationship between an input value and an output value of the simulation model and a parameter value of the simulation model; a data generating unit configured to generate a plurality of pseudo pairs of process values and the parameter values to be input to the simulation model while varying at least the parameter values during learning of the parameter estimation model; a simulation unit configured to calculate, during learning of the parameter estimation model, an output value of the simulation model for each pair of a process value and a parameter value when a process value generated by the data generation unit is input in a state in which the parameter values generated by the data generation unit are set in the simulation model, and to calculate, during execution of a simulation, an output value of the simulation model using a current value of the process value acquired from outside as an input; a second storage unit configured to store, as learning data, the process values and parameter values generated by the data generating unit and output values of the simulation model corresponding to the process values and parameter values during learning of the parameter estimation model; a learning unit configured to learn the parameter estimation model using the learning data; a parameter setting unit configured to calculate, before a simulation is performed, parameter values that are outputs of the parameter estimation model using a current value of the process value and a past actual value of the process value as inputs, and to set the parameter values in the simulation model; the simulation model receives as input a current value of a process value to be predicted and a predicted value at a future time of a related process value that affects the process value to be predicted, and outputs the predicted value of the process value to be predicted at a future time, the data generating unit generates a plurality of pseudo sets of the current value of the prediction target process value, the prediction value of the related process value, and the parameter value while varying at least the parameter value; the simulation unit calculates an output value of the simulation model when the current value of the prediction target process value and the predicted value of the related process value generated by the data generation unit are input during learning of the parameter estimation model, and calculates an output value of the simulation model when the current value of the prediction target process value and the predicted value of the related process value obtained from outside are input during execution of a simulation; the parameter setting unit calculates, before execution of a simulation, parameter values which are outputs of the parameter estimation model using as inputs a current value of the process value to be predicted, the current value of the related process value, and a past actual value of the process value to be predicted, which are acquired from an external source, and sets the parameter values in the simulation model.
2. a first step of generating a plurality of pseudo pairs of process values and parameter values input to the simulation model while varying at least the parameter values, during learning of a parameter estimation model that models a relationship between an input value and an output value of the simulation model and a parameter value of the simulation model; a second step of calculating, during learning of the parameter estimation model, an output value of the simulation model for each pair of a process value and a parameter value when the process value generated in the first step is input, with the parameter value generated in the first step being set in the simulation model; a third step of storing, as learning data, the process values and parameter values generated in the first step and output values of the simulation model corresponding to the process values and parameter values during learning of the parameter estimation model; a fourth step of learning the parameter estimation model using the learning data; a fifth step of calculating parameter values, which are outputs of the parameter estimation model using a current value of a process value obtained from an outside source and a past actual value of the process value, before the execution of a simulation, and setting the parameter values in the simulation model; and a sixth step of calculating, during a simulation run, output values of the simulation model using the current values of the process values as inputs; the simulation model receives as input a current value of a process value to be predicted and a predicted value at a future time of a related process value that affects the process value to be predicted, and outputs the predicted value of the process value to be predicted at a future time, the first step includes a step of generating a plurality of pseudo sets of the current value of the prediction target process value, the prediction value of the related process value, and the parameter value while varying at least the parameter value; the second step includes a step of calculating an output value of the simulation model when the current value of the prediction target process value generated in the first step and the predicted value of the related process value are input during learning of the parameter estimation model; the fifth step includes a step of calculating parameter values, which are outputs of the parameter estimation model, using as inputs a current value of the process value to be predicted, the current value of the related process value, and a past actual value of the process value to be predicted, which are acquired from an outside source, before executing a simulation, and setting the parameter values in the simulation model; the sixth step comprising a step of calculating an output value of the simulation model using as inputs a current value of the prediction target process value obtained from outside and the prediction value of the related process value during execution of a simulation.
Citation Information
Patent Citations
Adaptive control method
JP1991179502A
Method and system for regulating controller
JP1991233702A
Predicting equipment of thermal load
JP1996086490A
Plant simulation device and plant simulation method
JP2018055299A
Simulation device and simulation method
JP2019021032A
