Test and measurement device and control signal adjustment method for smu

Abstract

To easily measure a device under test (DUT).SOLUTION: An SMU comprises a user load 24 including a DUT, a reference model 20, a SMU controller 26, an adaptive control neural network 32, a predicting neural network 34, and the like. The controller 26 generates a control signal to control the generation of a voltage or current to supply to the DUT. The user load 24 receives the control signal to control the voltage and current to the DUT and generates an output signal for detecting the voltage and current of the DUT. The predicting neural network receives the control signal and the output signal of the user load, generates a predicted output signal of the user load, and supplies it to the adaptive control neural network. The adaptive control neural network receives the control signal, the output signal of the reference model, and the predicted output signal, and generates an adjusted output signal and adjusts the control signal to cause the user load to perform like the reference model.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a source measurement unit (SMU), and more particularly to a source measurement unit that can simplify operations using a neural network.

Background Art

[0002] A source measurement unit (SMU) accurately supplies voltage or current to a device under test (DUT) and simultaneously measures the voltage and current. In the design of a single output stage, the output stage supplies a voltage across both a load and a sense resistor R s The load often includes a device under test (DUT), and the detailed characteristics (such as resistance value) of the load are unknown. By using a sense resistor R s with a known resistance value, according to Ohm's law, the SMU can measure the current value flowing through the DUT or force the current value.

[0003] One problem with the conventional analog-based control loop of SMU products is that the controller is embedded in hardware, so the controller cannot be changed during execution. This problem of inability to make this change was solved by implementing the SMU control loop 10 digitally, as shown in a simplified version in FIG. 1. With the digital control loop 10, the controller can be changed during execution. The digital control loop 10 generally includes a certain programmable controller (control circuit) 12 that sets the voltage or current to a target value. An analog-to-digital converter (ADC) 11 is used to measure the voltage across the DUT via an amplifier (buffer) 18. That is, it functions as a voltage detection circuit that detects the voltage applied to the DUT. ADC 13 measures the voltage across a sense resistor R s with a known resistance value via an amplifier (buffer) 19, and this measured sense resistor R sThe voltage across both ends is used to calculate the current value flowing through the DUT. That is, it functions as a current detection circuit for detecting the current flowing through the DUT. The programmable controller 12 operates to generate a control signal for the DAC 15. At this time, the digital control loop 10 drives the digital-to-analog converter (DAC) 15 until the voltage and current of the DUT reach their respective desired levels. The digital control loop 10 receives a desired target value and adjusts the voltage supplied by the DAC 15. That is, the DAC 15 operates as a digitally controlled variable power supply. The voltage source 16 may also be a DAC that functions as a digitally controlled variable power supply, similar to the DAC 15. As is well known, these digitally controlled variable power supplies can function as voltage sources or current sources depending on the circuit configuration including peripheral circuits. Note that the DUT may actually be connected to the SMU using a test fixture or the like.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Non-Patent Documents

[0005]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] Typically, such a digital control type controller 12 requires the user to manually input information about the user's device under test (DUT) through a user interface such as a user interface (U / I) 14. However, since the user may not be well-versed in the device under test (DUT), such as a novice user, the operability of the SMU can be improved if measurements can be made without the user inputting detailed information about the DUT.

Means for Solving the Problem

[0007] Embodiments of the present application utilize the existence of a programmable digital control loop that enables the controller to be changed during execution. By changing the controller during execution, it becomes possible to optimize the digital control loop for a specific user load. Embodiments of the present application include a test measurement device (typically, a source measure unit (SMU)) and a method, which learn while dynamically changing the characteristics of the user load using a neural network. This test measurement device adjusts the control signal to correct different user loads during execution. The "control signal" referred to in the present application means a signal generated by a programmable controller to control the SMU to generate a voltage or current sent to the user load. The term "user load" referred to in the present application originally means the user's device under test (DUT), but in embodiments of the present application, it may substantially mean the DUT peripheral circuit 8 of the SMU including the DUT as shown in FIG. 1.

[0008] The term "processor" referred to in the present application refers to a programmable controller (control circuit). In the description of the present application, the programmable controller is referred to as a processor or a controller, which may include a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other types of controllers capable of executing these functions.

[0009] Embodiments of the present application provide the advantage that the SMU has the ability to learn a method for optimizing its own performance for any user load without requiring the user to manually input detailed information regarding the user load. The user may need to notify the SMU of the input range of the user load and give the SMU time to output a signal for a short time so that the SMU can learn the system dynamics of the user load. When the learning process is completed, the SMU will continuously attempt to optimize the performance of the SMU for the user load.

[0010] As used in the present application, "optimal performance" means that the SMU controller is designed to optimally control a reference model and matches the performance of this reference model. The type of controller used is irrelevant to the adaptive control of the neural network. The SMU controller is designed to optimally control the reference model. However, there will inevitably be a difference between the reference model and the user load due to the non-linearity and error of the reference model. Generally, the neural network aims to force the user load to operate like the reference model by referring to the control signal from the SMU controller, the output signal of the reference model, and the current output signal of the user load, and adjusts the control signal.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

DETAILED DESCRIPTION OF THE INVENTION

[0012] FIG. 2 shows an SMU controller, a reference model, a neural network, and a user load. The SMU controller 26 is an embodiment of the programmable controller 12 of the SMU of FIG. 1. The user load 24 corresponds to the DUT peripheral circuit 8 of the SMU including the DUT as shown in FIG. 1 in this embodiment. The user's DUT may be connected to the SMU by the user using, for example, a test fixture. Also, in this embodiment, the "output signal of the user load" may mean corresponding to a set of two output signals, i.e., the output signal of the ADC 11 and the output signal of the ADC 13, or may mean corresponding to each of the output signal of the ADC 11 or the output signal of the ADC 13.

[0013] In this description, the SMU control function is separated from the neural network 22, and the SMU controller 26 and the neural network 22 are referred to as separate components. However, physically, the SMU controller 26 and the neural network 22 may reside on the same device, such as an FPGA, or within the same digital signal processor (DSP) or ASIC. As noted above, the SMU controller 26 is similar to the programmable controller 12 of FIG. 1 , except that it is designed to optimally control the reference model 20. The reference model 20 is implemented in hardware, software, firmware, or a combination of these to operate similarly to the SMU DUT peripheral circuit 8 of FIG. 1 , except that the DUT included in the reference model 20 is configured as a load with “ideal” characteristics, and the response of the reference model 20 is referred to as a “desired” output signal or response in this description. Control signals from the SMU controller 26 are sent to the reference model 20 and the user load 24. The neural network 22 receives the output signal of the reference model 20 and the output signal and control signal of the user load 24, and supplies an output signal (adjusted output signal) to the adder 23 that adjusts the control signal so that the user load 24 operates like the reference model 20. The adder 23 receives a control signal from the SMU controller 26, and adds or combines the adjusted output signal to the control signal to generate an adjusted control signal, which is supplied to the user load 24.

[0014] As described in more detail below, the neural network 22 that generates the adjusted output signal constitutes an adaptive neural network. The adaptive neural network continuously learns by backpropagating its output signal error through the network, updating its adjustable parameters. However, the neural network's output signal error is typically not known by directly observing the control signal output. While waiting for the user load to respond is an option, waiting for a response can unnecessarily slow system execution. The challenge of not waiting for a response necessitates the need for a second neural network that can accurately predict the user load's output signal based on the input control signal. However, this second predictive neural network is technically optional and may be unnecessary.

[0015] As shown in Figure 3, the second neural network, referred to herein as a predictive neural network 30, receives the last N input signals to the user load and the last M output signals of the user load corresponding to the N input signals. In the present description, the N input signals to the user load and the M output signals of the user load are sometimes collectively referred to as the state of the user load. The predictive neural network 30 is trained to predict the next output signal of the user load based on the current signals and the current state of the user load.

[0016] More specifically, regarding predictive neural network learning, the predictive neural network 30 learns with respect to any given user load. The SMU controller generates a randomized control signal to input to the user load. During the learning process, by tracking the previous N input signals (control signals) and the M output signals corresponding to these N input signals, a pair of the control signal and the output signal resulting from the user load is formed. The learning data set of the predictive neural network is created by the control signal and the resulting output signal. When the predictive neural network is made to learn with reference to the previous N input signals and M output signals of the user load, the predictive neural network can predict the next output signal of the user load.

[0017] The randomized control signal has the characteristics (properties) of the actual control signal. Therefore, when the actual control signal is likely to be composed of signals based on functions such as steps, ramps (gradients), exponential functions, quadratic functions, and other functions, the learning signal also needs to have the characteristics (properties) of these functions. The randomized learning control signal can be generated by combining randomized basic functions in the following manner. (1) For each of a plurality of function types (steps, ramps, exponential functions, etc.), a function that is periodically and randomly scaled is generated. (2) One of these basic functions is randomly selected periodically by a multi-input single-output multiplexer. (3) The output signal is periodically and randomly scaled so that the output signal of the multiplexer falls within a predefined range that is safe for the DUT. Note that the randomized learning control signal can also be generated using other methods. Figure 4 shows an example of the randomized control signal described in the present application.

[0018] Once the predictive neural network has learned with respect to a specific user load, without additional learning, by supplying the current state of the user load in the form of previous input signals and output signals and a new single input signal to the predictive neural network using the system, it is possible to predict how any control signal affects the user load.

[0019] Figure 5 shows a more detailed embodiment of the SMU controller, neural network, and user load. Regarding the annotation of signals, the letter "u" indicates a variable control signal, and "y" indicates an output signal. The SMU controller 26 generates the control signal u c . The adjusted output signal "y a " from the adaptive control type neural network 32 adjusts the control signal u c and generates the control signal "u l " that is sent to the user load 24 and the memory 38. This adjustment is usually performed by adding an adjustment to the control signal. The output signal "y l " of the user load is also stored in the memory 38. The predictive neural network 34 receives a predetermined number of user load control signals and user load output signals from the memory 38. Next, the predictive neural network 34 generates a predictive output signal "y p " and provides the predictive output signal to the adaptive control type neural network 32. The adaptive control type neural network 32 also receives a set of reference model output signals from the reference model 20 stored in the memory 36 and a set of control signals generated by the controller. Next, the adaptive control type neural network 32 generates an adjustment output signal "y c " for adding an adjustment to the control signal u a generated by the SMU controller 26. This adjustment output signal is added / synthesized to the control signal in the adder 23. Since the SMU controller 26 is designed to optimally control the reference model 20, with this adjustment, the SMU can adjust its performance based on the user load 24 and make the user load 24 operate like the reference model 20.

[0020] The adaptive control type neural network 32 continuously learns during the operation of the SMU. Continuous learning means that at each processing stage, the system backpropagates the error of the output signal (the propagation back to the adaptive control type neural network 32) through the neural network and updates the adjustable parameters of the adaptive control type neural network 32. The error of the output signal of the adaptive control type neural network 23 cannot be known by directly observing the output of the control signal. In the case of arranging the complete system integrally, by using the prediction neural network, the failure of the learning function output signal of the adaptive control type neural network can be evaluated.

[0021] Figure 6 shows a flowchart of an embodiment of the control signal adjustment process. The SMU controller generates a control signal in step 40 and transmits the control signal to the adaptive control type neural network in step 42. The SMU controller further causes the prediction neural network to transmit the output signal from the user load to the adaptive control type neural network and also causes the output signal from the reference model to be transmitted to the adaptive control type neural network in step 42. The adaptive control type network generates an adjusted output signal in step 44. The SMU controller adjusts the control signal based on the adjusted output signal in step 46. This adjustment process is repeated as necessary until the user removes the DUT or terminates this adjustment process.

[0022] FIG. 7 shows an embodiment of an adaptive neural network 32. The adaptive neural network 32 adds or combines an adjustment output signal with a control signal generated by, for example, an SMU controller, and adjusts the control signal so that the combined control signal (adjusted control signal) causes the user load to operate in a manner similar to the reference model. The input to the adaptive neural network 32 is similar to that of the predictive neural network, with the addition of an area having a magnitude P of the output signal from the reference model 20 in FIG. 5. The output signal of the adaptive neural network is composed of a signal that is added to or adjusted with the control signal generated by the SMU controller.

[0023] The adaptive neural network learns "online." That is, each time the adaptive neural network generates an output signal (adjusted output signal), it backpropagates the error between the actual output signal and the desired output signal and adjusts the adjustable parameters of the adaptive neural network. This error causes the adaptive neural network to learn. Using a predictive neural network, the system can predict how the control signal will affect the user load. The resulting output signal of the adaptive neural network (adjusted output signal) is combined with the output signal of the SMU controller (control signal), and the combined control signal (adjusted control signal) is forwarded to the adaptive neural network via the predictive neural network. In this way, the system can predict how the combined control signal (adjusted control signal) will affect the user load. As mentioned above, the goal of the adaptive neural network is to make the user load behave in the same way as the reference model. Therefore, the error of the adaptive neural network can be calculated as the absolute difference between the predicted output signal and the output signal of the reference model (Equation 1). As a result of learning, the adaptive neural network will make adjustments to the control signal so that the user load behaves in the same way as the reference model. [Number 1] Error = |y p - y r |

[0024] In this way, the SMU with digital control function can adjust the control signal from the SMU controller using a neural network and make the user load operate like a reference model. Since the performance of the SMU is optimized for each of various user loads, the user only needs to set the input range of the DUT to the SMU so that the output signal of the DUT falls within a safe output range.

[0025] Aspects of the disclosed technology may operate on specially created hardware, firmware, digital signal processors, or specially programmed general-purpose computers, including processors that operate according to programmed instructions. The terms "controller" or "processor" herein contemplate microprocessors, microcomputers, ASICs, and dedicated hardware controllers, among others. Aspects of the disclosed technology may be implemented with computer-usable data and computer-executable instructions, such as one or more program modules, executed by one or more computers (including a monitoring module) or other devices. Generally, program modules include routines, programs, objects, components, data structures, etc., which, when executed by a processor in a computer or other device, perform particular tasks or implement particular abstract data types. Computer-executable instructions may be stored in computer-readable storage media, such as hard disks, optical disks, removable storage media, solid-state memory, RAM, etc. Those skilled in the art will appreciate that the functionality of the program modules may be combined or distributed as desired in various embodiments. Furthermore, such functionality may be embodied in whole or in part in firmware or hardware equivalents, such as integrated circuits, field programmable gate arrays (FPGAs), etc. Certain data structures may be used to more effectively implement one or more aspects of the disclosed technology, and such data structures are considered within the scope of the computer-executable instructions and computer-usable data described herein.

[0026] The disclosed embodiments may, in some cases, be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may be implemented as instructions carried or stored by one or more computer-readable media readable by and executable by one or more processors. Such instructions may be referred to as a computer program product. As used herein, a computer-readable media means any media accessible by a computing device. By way of example, and not limitation, a computer-readable media may include computer storage media and communication media.

[0027] A computer storage media means any media that can be used to store computer-readable information. By way of example, and not limitation, computer storage media may include random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital video disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, and any other removable or non-removable media implemented by any technology for storing volatile or nonvolatile information. Computer storage media excludes signals per se and transient forms of signal transmission.

[0028] A communication media means any media that can be used to communicate computer-readable information. By way of example, and not limitation, communication media may include coaxial cable, fiber optic cable, air, or any other media suitable for communicating electrical, optical, radio frequency (RF), infrared, acoustic, or other forms of signals. Examples

[0029] Examples that are helpful for understanding the technology disclosed in this application are presented below. Embodiments of this technology may include one or more of the examples described below and any combination thereof.

[0030] Example 1 is a test measurement device, a user load to which a device under test (DUT) is connected, one or more neural networks, one or more processors and includes, wherein the user load is a power supply that receives a control signal and supplies a voltage or current to the DUT, a current detection circuit for detecting the current flowing through the DUT, and a voltage detection circuit for detecting the voltage applied to the DUT and has, wherein the one or more processors perform a process of generating the control signal supplied to the user load and a reference model, perform a process of transmitting the control signal, the output signal of the user load based on the control signal, and the output signal of the reference model to the one or more neural networks to generate an adjusted output signal, and perform a process of operating the user load in the same manner as the reference model by adjusting the control signal with the adjusted output signal and is configured to execute a program that causes the one or more processors to perform the above processes.

[0031] Example 2 is the test measurement device of Example 1, wherein the one or more neural networks include at least one adaptive control type neural network.

[0032] Example 3 is the test measurement device of Example 2, wherein the one or more neural networks include a prediction neural network.

[0033] Example 4 is the test and measurement device of Example 3, wherein the predictive neural network is configured to receive the control signal and at least one output signal from the user load as input signals and to generate a predicted output signal of the user load based on the control signal, and the adaptive control neural network is configured to use the predicted output signal as the output signal from the user load.

[0034] Example 5 is the test and measurement device of Example 3, wherein the predictive neural network receives as input signals a predetermined number of previous control signals and previous output signals from the user load corresponding to the predetermined number of previous control signals and performs learning.

[0035] Example 6 is the test and measurement instrument of example 4, wherein the adaptive neural network continuously trains using a difference between the predicted output signal and the reference model output signal.

[0036] Example 7 is the test and measurement instrument of Example 3, wherein the one or more processors are further configured to execute a program that causes the one or more processors to perform a process for training the predictive neural network.

[0037] Example 8 is the test and measurement device of Example 7, wherein the program that causes the one or more processors to perform the process of training the predictive neural network includes a process of accessing a predetermined number of previous input signals to the user load and an output signal from the user load corresponding to the predetermined number of previous input signals, a process of generating a randomized control signal, a process of inputting the randomized control signal to the user load, a process of pairing the randomized control signal and the output signal from the user load in response to the randomized control signal as the predetermined number of previous input signals and the corresponding previous output signal to create a training data set, and a process of training the predictive neural network using the training data set.

[0038] Example 9 is the test measurement device of Example 8, and the program that causes the one or more processors to perform the process of generating the randomized control signal includes a program that causes the one or more processors to perform a process of scaling the randomized control signal so that the output signal of the DUT falls within a safe range for the DUT.

[0039] Example 10 is the test measurement device of any one of Examples 1 to 9, and further includes one or more memories for storing one or more of the output signal of the reference model, the control signal, the output signal of the user load, and the predicted output signal.

[0040] Example 11 is the test measurement device of any one of Examples 1 to 10, and the one or more neural networks are configured by a program executed by the one or more processors.

[0041] Example 12 is a method for automatically adjusting a control signal transmitted to a user load to which a device under test (DUT) is connected in a source measurement unit to control a voltage or current supplied to the DUT, a process of generating the control signal supplied to the user load and the reference model, a process of transmitting the control signal and the output signal of the user load and the output signal of the reference model based on the control signal to the one or more neural networks to generate an adjusted output signal, a process of operating the user load in the same manner as the reference model by adjusting the control signal with the adjusted output signal and includes.

[0042] Example 13 is the method of Example 12, further comprising a process of generating a predicted output signal from the user load by transmitting at least one of the control signals and at least one output signal from the user load as inputs to a predictive neural network, and a process of transmitting the predicted output signal to an adaptive control type neural network and using it as an output signal from the user load.

[0043] Example 14 is the method of Example 13, further comprising a process in which the predictive neural network receives, as input signals, a predetermined number of previous control signals and previous output signals from the user load corresponding to the predetermined number of previous control signals.

[0044] Example 15 is the method of Example 13, further comprising a process of continuously training the adaptive control type neural network using the difference between the predicted output signal and the output signal of the reference model.

[0045] Example 16 is the method of Example 13, further comprising a process of training the predictive neural network.

[0046] Example 17 is the method of Example 16, wherein the process of training the predictive neural network includes a process of accessing a predetermined number of previous input signals to the user load and previous output signals of the corresponding user load, a process of generating a randomized control signal, a process of inputting the randomized control signal to the user load, and a process of pairing the randomized control signal and the output signal from the user load corresponding to the randomized control signal as a pair with the predetermined number of previous input signals and the corresponding previous output signals to create a training data set, and a process of training the predictive neural network using the training data set.

[0047] Example 18 is the method of example 17, wherein generating the randomized control signal includes scaling the randomized control signal to place the output signal of the DUT within a safe range for the DUT.

[0048] Example 19 is the method of any of Examples 12 to 18, further comprising storing one or more of the output signal of the reference model, the control signal, the output signal from the user load, and any predicted output signals.

[0049] Additionally, the description of this application refers to specific features. It should be understood that the disclosure herein includes all possible combinations of these specific features. When a specific feature is disclosed in connection with a particular aspect or example, that feature can also be used in connection with other aspects and examples, to the extent possible.

[0050] Furthermore, when this application refers to a method having two or more defined steps or processes, these defined steps or processes may be performed in any order or simultaneously, unless the circumstances do not preclude this possibility.

[0051] All features disclosed in the specification, claims, abstract and drawings, and all steps in any disclosed method or process, may be combined in any combination, except where at least some of such features or steps are mutually exclusive combinations. Each feature disclosed in the specification, abstract, claims and drawings may be replaced by an alternative feature serving the same, equivalent or similar purpose, unless expressly stated otherwise.

[0052] While specific embodiments of the disclosed technology have been illustrated and described for purposes of illustration, it will be appreciated that various modifications can be made therein without departing from the spirit and scope of the invention. Accordingly, the disclosed technology should not be limited, except as by the appended claims. [Explanation of symbols]

[0053] 8 DUT peripheral circuits 10 Digital Control Loops 11 Analog-to-digital converter for voltage detection 12 Programmable Controllers 13 Analog-to-digital converter for current detection 14 User Interface (U / I) 15 Digital-to-Analog Converter 16 Voltage Source 17 Variable Amplifier 18 Voltage detection amplifier 19 Current detection amplifier 20 Reference Model 22 Neural Networks 23 Adder 24 user load 26 SMU controllers 30 Predictive Neural Networks 32 Adaptive Control Neural Networks 34 Predictive Neural Networks 36 memory 38 memory

Claims

1. A test measurement device, comprising a user load to which a device under test (DUT) is connected, one or more neural networks, and one or more processors, wherein the user load comprises a power supply that receives a control signal and supplies a voltage or current to the DUT, a current detection circuit for detecting a current flowing through the DUT, and a voltage detection circuit for detecting a voltage applied to the DUT, and the one or more processors are configured to execute a program that causes the one or more processors to perform a process of generating the control signal supplied to the user load and a reference model, a process of transmitting the control signal, an output signal of the user load based on the control signal, and an output signal of the reference model to the one or more neural networks to generate an adjusted output signal, and a process of adjusting the user load to operate in the same manner as the reference model by adjusting the control signal with the adjusted output signal. A test measurement device configured as described above.

2. The one or more neural networks include at least one adaptive control type neural network and a prediction neural network, wherein the prediction neural network is configured to receive, as input signals, the control signal and at least one output signal from the user load, and generate a predicted output signal of the user load based on the control signal, and the adaptive control type neural network is configured to use the predicted output signal as an output signal from the user load. The test measurement device according to Claim 1.

3. The prediction neural network according to Claim 2, wherein the prediction neural network receives, as input signals, a predetermined number of previous control signals and previous output signals from the user load corresponding to the predetermined number of previous control signals, and performs learning.

4. The adaptive control type neural network according to Claim 2, wherein the adaptive control type neural network continuously learns using a difference between the predicted output signal and an output signal of the reference model.

5. A program that causes the one or more processors to perform a process of causing the prediction neural network to learn, wherein the program accesses a predetermined number of previous input signals to the user load and output signals from the user load corresponding to the predetermined number of previous input signals, generates a randomized control signal, and inputs the randomized control signal to the user load. A process of creating a learning data set by pairing the randomized control signal and the output signal from the user load corresponding to the randomized control signal as the predetermined number of previous input signals and the corresponding previous output signals, A process of causing the prediction neural network to learn using the learning data set The test measurement device according to claim 3, including a program for causing the one or more processors to perform the above processes.

6. In a source measure unit (SMU), a method for automatically adjusting a control signal transmitted to a user load to which a device under test (DUT) is connected and controlling the voltage or current supplied to the DUT, comprising: A process of generating the control signal supplied to the user load and the reference model, A process of transmitting the control signal, the output signal of the user load based on the control signal, and the output signal of the reference model to the one or more neural networks to generate an adjusted output signal, A process of operating the user load in the same manner as the reference model by adjusting the control signal with the adjusted output signal A method for adjusting a control signal of an SMU comprising the above.

7. A process of generating a predicted output signal from the user load by transmitting at least one of the control signals and at least one output signal from the user load as inputs to a prediction neural network, and a process of transmitting the predicted output signal to an adaptive control type neural network and using it as an output signal from the user load. The method for adjusting a control signal of an SMU according to claim 6 further comprises the above.

8. The method for adjusting a control signal of an SMU according to claim 7, wherein the prediction neural network further comprises a process of receiving, as input signals, a predetermined number of previous control signals and previous output signals from the user load corresponding to the predetermined number of previous control signals, and causing the prediction neural network to learn.

9. The method for adjusting a control signal of an SMU according to claim 7 further comprises a process of continuously learning the adaptive control type neural network using the difference between the predicted output signal and the output signal of the reference model.

10. The process of causing the prediction neural network to learn is A process of accessing a predetermined number of previous input signals to the user load and the previous output signals of the corresponding user load, A process of generating a randomized control signal The process of inputting the randomized control signal into the user load, The process of pairing the randomized control signal and the output signal from the user load corresponding to the randomized control signal as the predetermined number of previous input signals and the corresponding previous output signals to create a learning data set, The process of training the prediction neural network using the learning data set The method for adjusting the control signal of the SMU according to claim 8, which comprises the above steps.

Images