Delay compensation systems, methods, and programs

Abstract

This invention provides a control device that adaptively compensates for delays in target devices with delay characteristics using machine learning technology. [Solution] A delay compensation system for a target device is provided, comprising: a target value acquisition unit for acquiring a target value; a state value acquisition unit for acquiring predetermined state values ​​related to the target device; a prediction processing unit for generating a predicted output value, which is a predicted value of the output value after a dead time has elapsed due to the delay characteristics of the target device, by inputting the target value, the state value, and past output values ​​fed back with respect to the target device into a learned model; a correction amount generation unit for generating a correction amount by applying a control gain to the deviation between the target value and the predicted output value; and a correction target value generation unit for generating a correction target value for controlling the target device by adding the target value and the correction amount.

Description

Technical Field

[0001] This invention relates to a control system, particularly a delay compensation system for a device having a delay characteristic.

Background Art

[0002] Conventionally, a control device for controlling a driving device or the like has been known. Also, it is generally practiced to incorporate this control device into a feedback control system and perform control within the feedback control system. Further, in recent years, for adaptive control, attempts have been made to incorporate a learned model obtained by machine learning into the feedback control system (Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, it is known that some control devices have a delay characteristic. For example, when a hydraulic control valve is used as the control device, a delay occurs in the output response due to various reasons such as the viscosity of the hydraulic oil.

[0005] If machine learning technology can be successfully applied to a feedback control system including a control device having a delay characteristic, it is possible to adaptively compensate for the delay characteristic by prediction.

[0006] However, no specific findings have been provided so far on how to apply machine learning technology to appropriately compensate for the delay of a target device having a delay characteristic in a feedback control system. <​The present invention was made against the technical background described above, and its objective is to adaptively compensate for delays in target devices that have delay characteristics using machine learning techniques. [Means for solving the problem]

[0008] The technical challenges described above can be solved by a delay compensation system having the following configuration.

[0009] In other words, the delay compensation system according to the present invention is a delay compensation system for a target device, comprising: a target value acquisition unit for acquiring a target value; a state value acquisition unit for acquiring predetermined state values ​​relating to the target device; a prediction processing unit for generating a predicted output value, which is a predicted value of the output value after the dead time elapsed due to the delay characteristics of the target device, by inputting the target value, the state value, and past output values ​​fed back with respect to the target device into a learned model; a correction amount generation unit for generating a correction amount by applying a control gain to the deviation between the target value and the predicted output value; and a correction target value generation unit for generating a correction target value for controlling the target device by adding the target value and the correction amount.

[0010] With this configuration, it is possible to predict the output value after the dead time has elapsed based at least on the target value, the status value related to the target device, and the past output value fed back for the target device, generate a deviation from the target value, and generate a corrected target value based on that deviation. Therefore, machine learning techniques can be used to adaptively compensate for delays in target devices that have delay characteristics.

[0011] The control gain may be a proportional gain, an integral gain, and / or a differential gain.

[0012] With this configuration, the correction amount can be generated using a highly reliable method.

[0013] The state value may directly or indirectly indicate the viscosity of the fluid that causes the lag characteristic in the target device.

[0014] With this configuration, predictions are made based on values ​​that directly or indirectly indicate the viscosity of the fluid that causes the delay characteristics in the target device, thereby enabling more appropriate delay compensation.

[0015] The trained model may be a model obtained by training a predetermined learning model on time series data in which the target value, the state value, and the output value are used as explanatory variables, and the output value is shifted to an earlier time by the amount of wasted time as the dependent variable.

[0016] With this configuration, the output value is shifted to a earlier time point (or to the point of inputting the target value) for training purposes. As a result, the trained model can immediately predict the value after the dead time has elapsed, and lag compensation can be performed from an early point in time.

[0017] The target device may also be a hydraulic control valve.

[0018] With this configuration, it is possible to compensate for the delay characteristic of a hydraulic control valve that is affected by the hydraulic fluid.

[0019] The target value may include the target pressure of the hydraulic control valve, the state value may include the measured temperature of the hydraulic fluid, and the output value may include the output pressure related to the hydraulic control valve.

[0020] With this configuration, predictions are made based on the temperature of the hydraulic fluid, which is associated with the lag characteristics, thus enabling predictions that appropriately take the lag characteristics into account.

[0021] The aforementioned state value may further include the coolant measurement temperature related to the hydraulic control valve and the rotational speed of the hydraulic pump engine related to the hydraulic control valve.

[0022] With this configuration, it is possible to make more accurate predictions based on a wide range of information.

[0023] Viewed from a different perspective, the present invention can be conceived as a delay compensation method. That is, the delay compensation method according to the present invention is a delay compensation method for a target device executed using an information processing device, and comprises: a target value acquisition step of acquiring a target value; a state value acquisition step of acquiring a predetermined state value relating to the target device; a prediction processing step of generating a predicted output value, which is a predicted value of the output value after the dead time elapsed due to the delay characteristics of the target device, by inputting the target value, the state value, and past output values ​​fed back with respect to the target device into a learned model; a correction amount generation step of generating a correction amount by applying a control gain to the deviation between the target value and the predicted output value; and a correction target value generation step of generating a correction target value for controlling the target device by adding the target value and the correction amount.

[0024] Viewed from a different perspective, the present invention can be conceived as a computer program. That is, the computer program according to the present invention causes the computer to function as a delay compensation system for a target device, comprising: a target value acquisition unit for acquiring a target value; a state value acquisition unit for acquiring predetermined state values ​​related to the target device; a prediction processing unit for generating a predicted output value, which is a predicted value of the output value after the dead time elapsed due to the delay characteristics of the target device, by inputting the target value, the state value, and past output values ​​fed back with respect to the target device into a learned model; a correction amount generation unit for generating a correction amount by applying a control gain to the deviation between the target value and the predicted output value; and a correction target value generation unit for generating a correction target value for controlling the target device by adding the target value and the correction amount. [Effects of the Invention]

[0025] According to the present invention, machine learning techniques can be used to adaptively compensate for delays in target devices that have delay characteristics. [Brief explanation of the drawing]

[0026] [Figure 1] Figure 1 is an overall hardware configuration diagram. [Figure 2] Figure 2 is an overall diagram of the feedback control system. [Figure 3] Figure 3 is a graph showing the response of the output value to the target value in a target device with a delay characteristic. [Figure 4] Figure 4 is an explanatory diagram illustrating an example of a training dataset used to train a lag compensation model. [Figure 5] Figure 5 is an operation flowchart of the feedback control system. [Modes for carrying out the invention]

[0027] Hereinafter, one embodiment of the present invention will be described in detail with reference to the attached drawings.

[0028] (1. First Embodiment) The first embodiment will be described with reference to Figures 1 to 5. In the first embodiment, an example of applying the present invention to a feedback control system that compensates for delay in a target device will be described.

[0029] In this embodiment, the target device is an electromagnetically driven hydraulic control valve used to control a hydraulic drive device. The target device can be any device that has a delay characteristic (or a delay in its output response). For example, it could be another device using a fluid that similarly exhibits a delay characteristic. Furthermore, the controlled variable may be a parameter other than pressure; for example, flow rate may be used as the controlled variable.

[0030] (1.1 Composition) (Control system configuration) Figure 1 is an overall diagram of the hardware configuration according to this embodiment. In the figure, the control device 30 is connected to the hydraulic drive unit 10 and controls the hydraulic drive unit 10. More specifically, the control device 30 is connected to the detector 12 attached to the hydraulic drive unit 10 and the hydraulic control valve 13. The detector 12 is a type of sensor that acquires various state values ​​of the hydraulic drive unit 10, the hydraulic control valve 13, or related devices. In this embodiment, the state value is the measured temperature (or oil temperature) Tm of the hydraulic fluid related to the hydraulic control valve 13. t , target temperature Tt of the hydraulic fluid t , Coolant measurement temperature Tc t , hydraulic pump engine rotation speed Rev t And so on. The control device 30 is also connected to the hydraulic control valve 13. As will be described later, the hydraulic control valve 13 is controlled based on the correction target value (delayed-compensated hydraulic target value) provided by the control device 30. In other words, this hydraulic control valve 13 is the controlled device having a delay characteristic.

[0031] The control device 30 comprises a processor 31, a storage unit 32, a communication unit 33, a display output unit 35, an audio output unit 36, and an I / O unit 37. The processor 31 is an arithmetic unit such as a CPU, which executes programs to realize various operations described later. The storage unit 32 is a storage medium (including non-temporary computer-readable storage media) such as ROM / RAM, flash memory, or hard disk, which stores programs and data to realize various operations described later. The communication unit 33 is a communication unit that enables the exchange of information with external devices. The display output unit 35 processes image information and the like to be displayed on a display device (not shown). The audio output unit 36 ​​processes audio information and the like to be output to a speaker (not shown). The I / O unit 37 processes input and output signals with external devices.

[0032] The processor 31, together with the memory unit 32 and other components, provides various functions.

[0033] FIG. 2 is an overall configuration diagram of a feedback control system 100 realized by a processor 31, a storage unit 32, and the like. The processor 31 acquires a target value r t from the storage unit 32 or an external device and provides it to the feedback control system 100. In the present embodiment, the target value r t is a control target value of the hydraulic pressure of a hydraulic control valve 13 to be controlled.

[0034] The processor 31 inputs the target value r t to a delay compensation model 50 which is a learned model obtained by machine learning. In addition to this, a past output value y t-1 obtained from the hydraulic control valve 13 which is a control target and a state value 120 related to the hydraulic control valve 13 are input to the delay compensation model 50. In the present embodiment, the output value y (or control amount) is the hydraulic pressure related to the hydraulic control valve 13, and the past output value is the output hydraulic pressure one step before.

[0035] In the present embodiment, the state value 120 is the measured temperature (or oil temperature) Tm t of the hydraulic oil related to the hydraulic control valve 13. However, the state value 120 is not limited to such a value. Therefore, for example, it may include any one or a combination of a target temperature Tt t of the hydraulic oil, a measured temperature Tc t of the coolant, a rotation speed Rev t of a hydraulic pump engine, and the like. Further, the state value 120 may be input as it is, or a processed value may be input in place of or in addition to them. For example, a total value of values of one or a plurality of past steps may be used. Also, a difference (gradient) between those values and past values may be included.

[0036] According to such a configuration, more accurate prediction can be performed based on a wider range of information.

[0037] When the target value r t , the state value 120, and the past output value y t-1 are input to the delay compensation model 50, the delay compensation model 50 outputs a predicted output value y' tThe subtractor 52 outputs this predicted output value y'. t and target value r t The deviation (or difference) of (=r t -y' t The function calculates the following and provides it to the correction amount generation unit 51.

[0038] In this embodiment, the term "prediction" refers to the output of the trained model and can be rephrased using other terms. For example, terms such as "inference" may be used.

[0039] The correction amount generation unit 51 performs proportional (P) control, integral (I) control, and differential (D) control on the deviation, and generates a correction amount u t The following is generated: the value obtained by multiplying the deviation by the proportional gain 511, the value obtained by multiplying the deviation by the integral gain 513 and integrating it with the integral element 512, and the value obtained by multiplying the deviation by the differential gain 516 and differentiating it with the differential element 515, are added together by the first adder 518 to form the correction amount u. t Output as follows.

[0040] This correction AKU t and target value r t This is added by the second adder 53 and provided to the hydraulic control valve 13, which is the target of control. That is, a corrected hydraulic control target value is provided. As a result of control based on this control target value, the hydraulic control valve 13, which is the target of control, outputs a value y via the detector 12. t The output hydraulic pressure is obtained as follows. This output value y t This is then fed back into the next time step.

[0041] Note that the hardware configuration is not limited to the configuration according to this embodiment. Therefore, other configurations may be adopted, for example, it may be configured as a server-client system, or a separate server may be provided for storing and providing data.

[0042] Furthermore, in this embodiment, the computer program may be provided as a computer program product or a recording medium for storing the computer program. The functions corresponding to the processor 31 may also be implemented circuit-wise using an IC such as an FPGA.

[0043] (Training of 50 delayed compensation models) Next, a method for generating a trained model related to the delay compensation model 50 used in the feedback control system 100 according to this embodiment will be described.

[0044] Figure 3 shows the target value r in the target device having a delay characteristic. t Output value y t This graph shows the response to the oil temperature Tm. Figure (a) shows the response to the oil temperature Tm. t The output response is shown when the oil temperature Tm is high, and Figure (b) shows the output response when the oil temperature Tm is high. t This figure shows the output response when the time step t is low. In this figure, the horizontal axis represents the time step t, and the vertical axis represents the output value y. t This represents the hydraulic pressure related to the hydraulic control valve 13. The dashed line represents the input, i.e., the target value r. t This shows the actual output value y. t This indicates that.

[0045] As is clear from the figure, in the hydraulic control valve 13, generally, the oil temperature Tm t The higher the oil temperature Tm, the smaller the delay time. t The lower the value, the greater the delay time. More specifically, oil temperature Tm t The lower the value of Tm, the longer the dead time before the power rises (the period when the power value is 0), and the smaller the slope of the power rise after it starts up. t This is because a lower value results in increased viscosity of the hydraulic fluid.

[0046] Thus, oil temperature Tm t Since the delay time changes in accordance with the change in the delay time t, d This also fluctuates. Specifically, in the case of Figure (a), the delay compensation time t dAs the value decreases, in the case of Figure (b), the time t for which delay compensation should be performed becomes smaller. d It becomes larger. Here, in this embodiment, the delay compensation time t d The target value is r t From the moment the input is received until the predetermined value y is output t It is defined as the time until the value increases. For example, in the example in Figure 3, the output value y t This is defined as the time it takes for the value to rise to 50% of its maximum value. Note that this predetermined value is not limited to 50% and may be any other percentage. Alternatively, it may be a predetermined value rather than a percentage. As will be described later, in this embodiment, the time t for which this delay compensation should be applied is defined as follows: d Regardless, immediately the predicted output value y' after the wasted time has elapsed. t This generates a pre-trained model that can generate [the desired output].

[0047] Figure 4 is an explanatory diagram showing an example of a training dataset used to train the lag compensation model 50. Figure (a) shows the hydraulic fluid temperature Tm t The training dataset for when the temperature is 50°C is shown in Figure (b), and the oil temperature Tm t This shows the training dataset for 10°C.

[0048] As is clear from Figure (a), oil temperature Tm t The target value is r at 50℃. t If we set it to 3, the output value y t If we have data that starts to rise with a 10ms delay, the explanatory variables of the trained model are the oil temperature Tm at each time step t. t , target value r t , and output value y t It is said that the dependent variable at that time is the output value y. t The value of the dead time (10ms in the example shown in the figure) (or the target value r) t This value is considered to be a value shifted towards earlier time (up to the time of input).

[0049] The same applies to Figure (b). That is, oil temperature Tm t The target value is r at 10℃ t If we set it to 3, the output value y tIf we have data that starts to rise with a 30ms delay, the explanatory variables of the trained model are the oil temperature Tm at each time step t. t , target value r t , and output value y t It is said that the dependent variable at that time is the output value y. t The value of the dead time (30ms in the example shown in the figure) (or the target value r) t This value is considered to be a value shifted towards earlier time (up to the time of input).

[0050] Using this training data, a trained model, i.e., a lag-compensated model 50, can be generated by performing supervised learning on a predetermined learning model, such as a neural network.

[0051] In this embodiment, the model used for learning is a neural network, but the present invention is not limited to such a configuration. Therefore, other models can also be used. For example, models such as linear models, support vector machines (SVMs), Bayesian models, decision trees, random forests using tree structures, and gradient boosting may be used.

[0052] With this configuration, the output value is shifted to a earlier time point (or to the point of inputting the target value) for training purposes. As a result, the trained model can immediately predict the value after the dead time has elapsed, and lag compensation can be performed from an early point in time.

[0053] Furthermore, with this configuration, predictions are made based on values ​​that directly or indirectly indicate the viscosity of the fluid that causes the lag characteristic in the target device, thus enabling more appropriate lag compensation.

[0054] (1.2 Operation of the control system) The operation of the feedback control system 100 according to this embodiment will be described with reference to Figure 5.

[0055] When control begins, the processor 31 performs initialization processing for the time step t (S11). For example, it performs processing to set the time step t to 0.

[0056] After the initialization process, the processor 31 receives the target value r from the storage unit 32 or an external device. t The processor 31 performs a process to acquire the following (S12). The processor 31 also acquires various state values ​​from the hydraulic drive unit 10 and the hydraulic control valve 13 or their related devices obtained via the detector 12 (sensor) (S12). In this embodiment, the state values ​​include at least the measured temperature (or oil temperature) Tm of the hydraulic fluid related to the hydraulic control valve 13. t Includes.

[0057] With this configuration, predictions are made based on the temperature of the hydraulic fluid, which is associated with the lag characteristics, thus enabling predictions that appropriately take the lag characteristics into account.

[0058] After processing the acquisition of various information, the processor 31 sets the target value r for the delay compensation model 5. t , state value 120 and feedback past output value y t-1 The input is used to perform prediction processing on the lag compensation model 5, and the predicted output value y' is obtained. t Generate (S13).

[0059] Predicted output value y' t After generating the predicted output value y', the processor 31 then generates the predicted output value y' t and target value r t The deviation (or difference) of (=r t -y' t The error ) is calculated and input to the correction amount generation unit 51 (S15). When this error is input, the correction amount generation unit 51 performs proportional (P) control, integral (I) control, and differential (D) control on each error to produce a correction amount u t (S15) The following is generated: the value obtained by multiplying the deviation by the proportional gain 511, the value obtained by multiplying the deviation by the integral gain 513 and integrating it with the integral element 512, and the value obtained by multiplying the deviation by the differential gain 516 and differentiating it with the differential element 515, which are added together by the first adder 518 to form the correction amount u t Output as follows.

[0060] With this configuration, correction quantities can be generated using reliable means based on classical control.

[0061] Correction amount u t After generation, the processor 31 uses the adder 53 to generate the target value r t and correction amount u t The processor 31 performs a process to generate a correction target value by adding the two values ​​(S16). After that, the processor 31 performs a process to provide the correction target value to the hydraulic control valve 13, which is the target of control (S16). As a result, the hydraulic control valve 13 performs control using the correction target value.

[0062] After the control process, the processor 31 performs an increment process for time step t (S18). After this increment process, the process returns to the information acquisition step (S12). This series of processes is repeated until interrupted by an interrupt or the like.

[0063] With this configuration, it is possible to predict the output value after the dead time has elapsed based at least on the target value, the status value related to the target device, and the past output value fed back for the target device, generate a deviation from the target value, and generate a corrected target value based on that deviation. Therefore, machine learning techniques can be used to adaptively compensate for delays in target devices that have delay characteristics.

[0064] Although embodiments of the present invention have been described above, these embodiments represent only a part of the application examples of the present invention, and are not intended to limit the technical scope of the present invention to the specific configurations of the above embodiments. Furthermore, the above embodiments can be combined as appropriate, as long as no contradictions arise. [Industrial applicability]

[0065] This invention can be used in various industries that utilize control technology and machine learning technology. [Explanation of symbols]

[0066] 10. Hydraulic drive system 12 Detectors (sensors) 13. Hydraulic control valve (controlled object) 30 Control device 31 processors 32 Storage section 33 Communications Department 35 Display Output Section 36 Audio output section 37 I / O Department 50. Delay Compensation Model (Pre-trained Model) 51 Correction amount generation section 511 Proportional Gain 512 Integral Elements 513 Integral Gain 515 Differential Elements 516 Differential Gain 518 Adder 52 Subtractors 53 Adder 100 Feedback Control System 120 Status Value

Claims

1. A delay compensation system for the target device, A unit for acquiring target values, A status value acquisition unit that acquires predetermined status values ​​related to the target device, A prediction processing unit generates a predicted output value, which is a predicted value of the output value after the dead time elapsed due to the delay characteristics of the target device, by inputting the aforementioned target value, the aforementioned state value, and the past output value that has been fed back regarding the target device into a trained model. A correction amount generation unit that generates a correction amount by applying a control gain to the deviation between the target value and the predicted output value, The system includes a correction target value generation unit that generates a correction target value for controlling the target device by adding the aforementioned target value and the aforementioned correction amount, A delay compensation system in which the state value is a value that directly or indirectly indicates the viscosity of the fluid that causes the delay characteristic in the target device.

2. The lag compensation system according to claim 1, wherein the control gain is a proportional gain, an integral gain, and / or a differential gain.

3. The delay compensation system according to claim 1, wherein the trained model is a model obtained by training a predetermined learning model on time series data in which the target value, the state value, and the output value are used as explanatory variables, and the output value is shifted to an earlier time by the amount of wasted time as the dependent variable.

4. The delay compensation system according to claim 1, wherein the target device is a hydraulic control valve.

5. The aforementioned target value includes the target pressure of the hydraulic control valve. The aforementioned state value includes the measured temperature of the hydraulic fluid. The delay compensation system according to claim 4, wherein the output value includes the output pressure related to the hydraulic control valve.

6. The delay compensation system according to claim 5, wherein the state value further includes the coolant measurement temperature related to the hydraulic control valve and the rotational speed of the hydraulic pump engine related to the hydraulic control valve.

7. A delay compensation method for a target device, which is performed using an information processing device, The steps to obtain the target value and the target value acquisition step, A state value acquisition step, which involves acquiring a predetermined state value for the target device, A prediction processing step involves inputting the target value, the state value, and past output values ​​fed back with respect to the target device into a trained model to generate a predicted output value, which is a predicted value of the output value after the dead time elapsed due to the delay characteristics of the target device. A correction amount generation step, which involves applying a control gain to the deviation between the target value and the predicted output value to generate a correction amount, The system includes a correction target value generation step, which generates a correction target value for controlling the target device by adding the aforementioned target value and the aforementioned correction amount, A delay compensation method wherein the state value is a value that directly or indirectly indicates the viscosity of the fluid that causes the delay characteristic in the target device.

8. A computer is a delay compensation system for the target device, A unit for acquiring target values, A status value acquisition unit that acquires predetermined status values ​​related to the target device, A prediction processing unit generates a predicted output value, which is a predicted value of the output value after the dead time elapsed due to the delay characteristics of the target device, by inputting the aforementioned target value, the aforementioned state value, and the past output value that has been fed back regarding the target device into a trained model. A correction amount generation unit that generates a correction amount by applying a control gain to the deviation between the target value and the predicted output value, The system includes a correction target value generation unit that generates a correction target value for controlling the target device by adding the aforementioned target value and the aforementioned correction amount, A computer program that functions as a delay compensation system, wherein the state value is a value that directly or indirectly indicates the viscosity of the fluid that causes the delay characteristic in the target device.

Images