Model-free adaptive control method for reservoir scheduling and terminal device

Abstract

The application belongs to the technical field of water conservancy regulation, and particularly discloses a reservoir regulation method and terminal equipment based on model-free adaptive control, which comprises the following steps: obtaining an actual observed water level at a time point in a regulation period; determining a disturbance estimation value at the time point in the regulation period according to the actual observed water level, a predicted water level and the disturbance estimation value at the time point in the regulation period; determining a water level error at the time point in the regulation period according to the actual observed water level and the predicted water level at the time point in the regulation period; and determining a control amount at the time point in the regulation period according to the disturbance estimation value at the time point in the regulation period, the water level error at the time point in the regulation period and the control amount at the time point in the regulation period. The application does not depend on a traditional hydrological physical model, and can significantly reduce the difficulty of modeling and parameter identification.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water conservancy regulation, and particularly discloses a reservoir regulation method and terminal equipment based on model-free adaptive control. BACKGROUND

[0002] With the increasing complexity and dynamicity of water resource systems, reservoirs, as the core infrastructure for flood control and regulation, water supply guarantee and ecological regulation, face unprecedented challenges in regulation and management. At present, the mainstream reservoir regulation method mainly relies on optimization control strategies based on physical models, which describe the hydrological process of the reservoir area through the establishment of hydrology and hydrology models, and solve the regulation instructions with the help of mathematical optimization algorithms. However, the actual hydrological system often has strong nonlinearity, strong uncertainty and significant external disturbance, which leads to the following typical problems of these model-based regulation methods:

[0003] Strong dependence on model accuracy, when facing external disturbances such as sudden rainfall, sudden water release upstream, etc. (such as local heavy rainfall or abnormal drainage upstream), the prediction deviation of the traditional model may be significantly amplified, thereby affecting the safety of regulation; complex modeling process, low update frequency, for systems with non-stationary input characteristics, model reconstruction and parameter correction are often time-consuming and laborious, and it is difficult to meet the high-frequency regulation demand under the rapidly changing environment; lack of data-driven mechanism, insufficient intelligence, and unable to fully tap the potential laws of historical and real-time hydrological information for dynamic regulation. SUMMARY

[0004] The purpose of the present application is to provide a reservoir regulation method and terminal equipment based on model-free adaptive control, to solve the technical problems of strong model dependence and lack of data-driven mechanism in existing reservoir regulation methods, which affect the safety of regulation.

[0005] The first aspect of the present application provides a reservoir regulation method based on model-free adaptive control, comprising:

[0006] Step 1, obtaining the actual observed water level at time in the regulation period;

[0007] Step 2, determining the disturbance estimate value at time in the regulation period according to the actual observed water level, the predicted water level and the disturbance estimate value at time in the regulation period;

[0008] Step 3, determining the predicted water level at time in the regulation period according to the actual observed water level and the predicted water level at time in the regulation period; ​the water level error at the time point;

[0009] Step 4, determining the control variable at the time point according to the time point within the dispatching period the disturbance estimation value at the time point, the time point within the dispatching period the water level error at the time point, and the time point within the dispatching period determining the control variable at the time point the time point within the dispatching period the control variable at the time point.

[0010] Preferably, Step 2 is specifically:

[0011] Step 2.1, determining the difference between the actual observed water level and the predicted water level at the time point within the dispatching period; the time point within the dispatching period the difference between the actual observed water level and the predicted water level at the time point;

[0012] Step 2.2, determining the water level variable according to the difference and a preset gain value;

[0013] Step 2.3, determining the disturbance estimation value at the time point within the dispatching period according to the sum of the water level variable and the time point within the dispatching period the sum of the disturbance estimation value at the time point; the time point within the dispatching period the disturbance estimation value at the time point.

[0014] Preferably, after Step 2.3, it further includes:

[0015] when the disturbance estimation value at the time point within the dispatching period is greater than a preset saturation value constant, the time point within the dispatching period the disturbance estimation value at the time point is the saturation value constant; the time point within the dispatching period the disturbance estimation value at the time point is the negative saturation value constant.

[0016] when the disturbance estimation value at the time point within the dispatching period is less than a negative saturation value constant, the time point within the dispatching period the disturbance estimation value at the time point is the negative saturation value constant. Preferably, Step 3 is specifically:

[0017] determining the water level error at the time point within the dispatching period according to the time point within the dispatching period

[0018] the difference between the actual observed water level and the predicted water level at the time point. the time point within the dispatching period the difference between the actual observed water level and the predicted water level at the time point. the time point within the dispatching period the water level error at the time point.

[0019] Preferably, Step 4 is specifically:

[0020] Step 4.1, determining within the scheduling period an estimated value of the dynamic adjustment factor at the time instant;

[0021] Step 4.2, determining a control inheritance item based on within the scheduling period the estimated value of the dynamic adjustment factor at the time instant and within the scheduling period the control quantity at the time instant;

[0022] Step 4.3, determining a disturbance correction item based on within the scheduling period the estimated value of the dynamic adjustment factor at the time instant, within the scheduling period the disturbance estimate at the time instant and within the scheduling period the water level error at the time instant;

[0023] Step 4.4, determining the control quantity at the time instant within the scheduling period based on the sum of the control inheritance item and the disturbance correction item.

[0024] Preferably, the control inheritance item in Step 4.2 is determined according to a first formula, which is:

[0025]

[0026] wherein, is an energy consumption penalty factor; is a weight; is the estimated value of the dynamic adjustment factor at the time instant within the scheduling period is the control quantity at the time instant within the scheduling period

[0027] Preferably, the disturbance correction item in Step 4.3 is determined according to a second formula, which is:

[0028]

[0029] wherein, is the estimated value of the dynamic adjustment factor at the time instant within the scheduling period is an energy consumption penalty factor; is a weight; is the water level error at the time instant within the scheduling period is​​​​​​ Within the scheduling period The actual observed water level at any given time for Within the scheduling period Predicted water level at any given time; for Within the scheduling period The estimated value of the disturbance at time.

[0030] Preferably, step 4.1 specifically includes:

[0031] Sure Within the scheduling period Time-varying water level error and rate of change of control quantity;

[0032] according to Within the scheduling period Determination of time-dependent water level error, rate of change of control variables, and estimated values ​​of dynamic adjustment factors. Within the scheduling period The estimated value of the dynamic adjustment factor at any given time.

[0033] Preferably, the method further includes the following steps before step 1:

[0034] The reservoir Using historical control values ​​from previous scheduling cycles as input and historical water level values ​​as output, a machine learning model is trained to obtain... The initial control variable for the scheduling cycle.

[0035] A second aspect of the present invention provides a terminal device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described model-free adaptive control reservoir scheduling method.

[0036] The model-free adaptive control reservoir scheduling method and terminal equipment of the present invention have the following advantages compared with the prior art:

[0037] The reservoir scheduling method of the model-free adaptive control based on a machine learning model and an extended state observer fully integrates the prediction ability of machine learning, the disturbance estimation compensation mechanism and the data-driven control algorithm, and breaks through the bottleneck problems of the traditional scheduling mode, such as strong dependence on system modeling, response lag, and unpredictable disturbance. The machine learning model is used to realize efficient prediction of the initial control quantity, the extended state observer is used to dynamically estimate external disturbances such as sudden inflow and rainfall, and the model-free adaptive controller is used to realize iterative optimization of the scheduling instruction according to error feedback, so as to finally form a closed-loop control mechanism with adaptability, robustness and real-time performance. The algorithm is stable, can be embedded in the existing reservoir scheduling automation platform, is suitable for reservoir scheduling scenes of different scales and different operation targets, and has significant engineering promotion value and practical application significance.

[0038] The reservoir scheduling strategy can be optimized and controlled without relying on traditional hydrological physical models, the difficulty of modeling and parameter identification is significantly reduced, and the method is suitable for complex, nonlinear and time-varying hydrological system environments.

[0039] The machine learning model is used to process historical hydrological data and real-time hydrological data, and output initial control quantities for controller initialization, so as to effectively reduce the initial control error, improve the control precision and accelerate the convergence speed of the control strategy.

[0040] The extended state observer can be used to estimate external disturbance factors such as sudden rainfall and upstream water inflow changes in real time, and the disturbance factors are used for control compensation, so as to improve the operation stability and scheduling robustness of the reservoir under non-ideal or extreme working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The flowchart of the reservoir scheduling method of the model-free adaptive control of the embodiment of the present application. DETAILED DESCRIPTION

[0042] In the following description, specific details are set forth such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0043] The first aspect of the embodiment of the present application provides a reservoir scheduling method of model-free adaptive control, which can solve the problems that the traditional scheduling method is time-consuming and laborious to establish an accurate model, is affected by external interference, and cannot quickly respond to sudden hydrological changes, and the flowchart is as shown in Figure 1 .

[0044] The embodiment of the present application firstly takes the historical control amount of the reservoir before the dispatching period as input and the historical water level value as output to train the machine learning model, and obtains the initial control amount of the dispatching period . .

[0045] The above-mentioned historical control amount and initial control amount are the inflow and outflow of the reservoir. In the embodiment of the present application, the inflow is the inflow of the reservoir transmitted by the water level station of the tributary and the main stream, the outflow is obtained by measuring the flow meter at the gate of the tributary and the main stream, and the water level is the reservoir water level measured by the radar water level meter.

[0046] The above-mentioned machine learning model can be a random forest model, a support vector machine and a decision tree, etc. Since the random forest model has the advantages of automatically analyzing the influence degree of each feature on the prediction result, high accuracy and robustness, being able to process high-dimensional data, strong anti-noise ability, supporting parallel computing, etc., the embodiment of the present application preferably uses the random forest model. The random forest model outputs the initial control amount for initializing the model-free adaptive control, which is used as the input initial value of the iterative controller, can effectively reduce the error in the initial stage of dispatching, improve the control efficiency and reduce the iteration round.

[0047] The specific process of determining the initial control amount by the random forest model is as follows:

[0048] The regression type random forest model composed of multiple decision trees is trained, and then the vector composed of the water level values at the initial time of different main streams and tributaries in the dispatching period is input into the regression type random forest model, and the water level values are input into each regression tree to respectively predict the control initial value of the dispatching period. , is the control initial value of the i th regression tree in the dispatching period. The prediction results of all regression trees are integrated by averaging to generate the final output, which is specifically as follows:

[0049] (1)

[0050] wherein, is the initial control amount of the dispatching period; is the total number of regression trees in the random forest; indicates the control initial value of the i th regression tree in the dispatching period.

[0051] ​​​​​​​​​​As the initial input for subsequent modelless controllers, it can effectively reduce initial control errors and improve the convergence speed and scheduling stability of iterative control.

[0052] In determining Initial control quantity of the scheduling cycle Afterwards, model-free adaptive control of reservoir scheduling can be carried out, specifically as follows:

[0053] Step 1, Obtain Within the scheduling period Actual observed water level at time .

[0054] Step 2, according to Within the scheduling period Actual observed water level at time Predicting water levels and disturbance estimates Sure Within the scheduling period Perturbation estimate at time Specifically:

[0055] Step 2.1, Determine Within the scheduling period Actual observed water level at time With predicted water level The difference .

[0056] Step 2.2: Based on the difference and preset gain value Determine water level variables .

[0057] Step 2.3: Based on water level changes and Within the scheduling period Perturbation estimate at time The sum is determined Within the scheduling period Perturbation estimate at time As shown in formula (2):

[0058] (2)

[0059] The disturbance terms in this embodiment of the invention include, but are not limited to, external disturbance sources such as localized heavy rainfall, sudden changes in upstream inflow, and fluctuations in the operation of dispatching facilities. Step 2 above is implemented by an extended state observer, thereby estimating the external disturbance terms existing in the system during the dispatching process, which facilitates subsequent disturbance compensation for control commands. The extended state observer iteratively determines the disturbance estimate, avoiding the reliance on modeling the disturbance source and improving the control system's adaptability to sudden changes.

[0060] To avoid excessively large disturbance estimates leading to control oscillations, this embodiment of the invention further includes the following after step 2.3:

[0061] when Within the scheduling period Perturbation estimate at time Greater than the preset saturation constant hour, Within the scheduling period The perturbation estimate at time t is the saturation constant. ;when Within the scheduling period Perturbation estimate at time Less than the negative saturation constant hour, Within the scheduling period The perturbation estimate at time t is a negative saturation constant. The details are as follows:

[0062] (3)

[0063] Disturbance estimate It is passed to the controller to compensate for the error term, thereby improving the stability and robustness of the controller under non-ideal operating conditions.

[0064] Step 3, according to Within the scheduling period Actual observed water level at time and predicted water level Determining the difference Within the scheduling period Water level error at time As shown in formula (4):

[0065] (4)

[0066] Step 4, according to Within the scheduling period Perturbation estimate at time , Within the scheduling period Water level error at time as well as Within the scheduling period Control of time ,Sure Within the scheduling period The amount of control at any given moment.

[0067] The embodiment of the present application uses a controller to implement step 4, and the controller implements control update based on error driving, without system modeling, with fast convergence speed, and suitable for complex and time-varying hydrological scheduling scenarios.

[0068] The step 4 is specifically:

[0069] Step 4.1, determining the disturbance estimation value in the scheduling period at the moment .

[0070] The step 4.1 is specifically:

[0071] determining the disturbance estimation value in the scheduling period at the moment and the change rate of the control amount .

[0072] According to the disturbance estimation value in the scheduling period at the moment , the change rate of the control amount and the estimation value of the dynamic adjustment factor , the estimation value of the dynamic adjustment factor in the scheduling period at the moment is determined , as shown in formula (5):

[0073] (5)

[0074] In the formula, is the adjustment step; is the update smoothing factor.

[0075] The dynamic adjustment factor is used to balance the control speed, and it is unpredictable itself, and the embodiment of the present application performs online estimation on it by minimizing the cost function of formula (6), so as to obtain :

[0076] (6)

[0077] In the formula, is the cost function; is the dynamic adjustment factor in the scheduling period at the moment ; is the estimation value of the dynamic adjustment factor in the scheduling period at the moment .

[0078] Step 4.2, according to the disturbance estimation value in the scheduling period estimated value of the dynamic adjustment factor at the time point and control amount at the time point within the scheduling period control amount at the time point within the scheduling period determining a control inheritance term as shown in equation (7):

[0079] (7)

[0080] wherein, is an energy consumption penalty factor for limiting the scheduling intensity; is a weight; is estimated value of the dynamic adjustment factor at the time point within the scheduling period estimated value of the dynamic adjustment factor at the time point within the scheduling period is control amount at the time point within the scheduling period control amount at the time point within the scheduling period

[0081] Step 4.3, determining a disturbance correction term according to the estimated value of the dynamic adjustment factor at the time point within the scheduling period estimated value of the disturbance at the time point within the scheduling period estimated value of the disturbance at the time point within the scheduling period , estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period determining a disturbance correction term as shown in equation (8):

[0082] (8)

[0083] wherein, is the estimated value of the dynamic adjustment factor at the time point within the scheduling period is an energy consumption penalty factor; is a weight; is estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period estimated value of the water level error at the time point within the scheduling period

[0084] Step 4.4, determining a control inheritance term according to the control inheritance term and the disturbance correction term The sum of the water level error and the disturbance estimation value is determined In a dispatch cycle The control variable at the time instant, as shown in equation (9):

[0085] (9)

[0086] The control inheritance item of the embodiment controls the inheritance degree of the current dispatch instruction to the previous time instant dispatch instruction, and the disturbance correction item corrects the dispatch instruction based on the water level error and the disturbance estimation value.

[0087] The controller used in the dispatch process of the application uses the objective function as the optimization criterion, and the objective function is as follows:

[0088] (10)

[0089] In the formula, is In a dispatch cycle The control variable at the time instant The objective function of is In a dispatch cycle The water level error at the time instant is In a dispatch cycle The estimation value of the dynamic adjustment factor at the time instant is In a dispatch cycle The change rate of the control variable at the time instant, and ; is In a dispatch cycle The disturbance estimation value at the time instant is the energy consumption penalty factor is the weight is In a dispatch cycle The control variable at the time instant

[0090] The partial derivative of the above formula (10) is taken and is zero, and the optimal control law as shown in formula (9) is obtained.

[0091] The control law of formula (9) can be simplified as follows:

[0092] (11)

[0093] In the formula, is a historical reservation factor, is an error amplification factor, and the above two parameters can adjust the iteration speed and sensitivity of the controller.

[0094] The reservoir scheduling method according to the embodiment of the application determines the scheduling instruction according to the above reservoir scheduling method, including opening degree adjustment of a reservoir gate, realizes water inflow and water outflow adjustment, ensures that the water level of the reservoir is within a set range, and realizes accurate control of the water level of the reservoir and efficient utilization and intelligent management of water resources.

[0095] The embodiment of the application periodically backs up and stores the scheduling operation records and hydrological data, ensures that the data can be restored in case of failure or emergency, and guarantees long-term stable operation of the system. Meanwhile, the scheduling strategy is continuously optimized by analyzing historical hydrological data, and the data is used to adjust parameters in the controller, further improving the scheduling efficiency.

[0096] The method of the application realizes adaptive adjustment of input by real-time output data control, breaks away from the dependence on accurate mathematical models, and greatly improves the flexibility and adaptability of scheduling.

[0097] The method of the application can realize real-time adjustment of the water level of the reservoir, quickly respond to sudden hydrological changes, and ensure the safety and efficiency of the operation of the reservoir.

[0098] The method of the application realizes self-learning and optimized scheduling according to historical data and real-time data, and significantly improves the intelligent management level of the reservoir.

[0099] The method of the application makes scheduling decisions in combination with meteorological forecast data and real-time hydrological data, realizes early response to extreme weather events, and further improves the accuracy of scheduling.

[0100] The second aspect of the application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor realizes the steps of the above-mentioned model-free adaptive control reservoir scheduling method when executing the computer program.

[0101] The application is essentially a model-free adaptive iterative learning control reservoir scheduling method based on a random forest model and an extended state observer, which fully integrates machine learning prediction capability, disturbance estimation compensation mechanism and data-driven control algorithm, and breaks through the bottleneck problems of strong dependence on system modeling, response lag, unpredictable disturbance and the like in traditional scheduling methods. The application realizes efficient prediction of the initial control amount through a random forest model, dynamically estimates external disturbances such as sudden inflow and rainfall through an extended state observer, and realizes iterative optimization of the scheduling instruction according to error feedback by a model-free adaptive controller, finally forming a closed-loop control mechanism with adaptability, robustness and real-time performance. The algorithm of the application is stable in operation, can be widely applied to various types of reservoir scheduling scenes, and has significant engineering popularization value and practical application significance.

[0102] The above merely describes several embodiments of the present application and does not limit the present application in any form. Although the present application is disclosed with the above preferred embodiments, it is not intended to limit the present application. Any skilled person in the art can make some changes or modifications to the disclosed technical contents without departing from the scope of the technical solutions of the present application, and the equivalent embodiments are equivalent to the equivalent embodiments, which are within the scope of the technical solutions.

Claims

1. A method for reservoir scheduling of model-free adaptive control, characterized in that, Comprising: Step 1, obtaining within a dispatch cycle actual observed water level at the time Step 2, according to the actual observation water level, the predicted water level and the disturbance estimation value at the time point within the dispatching period the actual observation water level, the predicted water level and the disturbance estimation value at the time point within the dispatching period the actual observation water level, the predicted water level and the disturbance estimation value at the time point within the dispatching period the actual observation water level, the predicted water level and the disturbance estimation value at the time point within the dispatching period Step 2.1, determining within a scheduling period a difference between the actual observed water level and the predicted water level at the time instant; Step 2.2, determining a water level variable according to the difference and a preset gain value; Step 2.

3. determining a sum of disturbance estimates for the time instants in the scheduling period based on the water level variable and the scheduling period the sum of disturbance estimates for the time instants in the scheduling period the scheduling period the sum of disturbance estimates for the time instants in the scheduling period Step 3, according to the scheduling period the actual observation water level and the predicted water level at the time determine the scheduling period the water level error at the time, specifically: According to the difference between the actual observed water level and the predicted water level at the time determines the water level error at the time within the dispatch cycle; Step 4, according to the scheduling period the water level error at the time point the scheduling period the water level error at the time point the scheduling period the control amount at the time point the scheduling period the control amount at the time point Step 4.1, determining within a scheduling period the estimated value of the dynamic adjustment factor at the moment, in particular: determining within a scheduling period the rate of change of the water level error and the control variable at the time instant According to Within the dispatch cycle The estimation of the water level error, the rate of change of the control variable and the dynamic adjustment factor at the instant of time determines Within the dispatch cycle The estimation of the dynamic adjustment factor at the instant of time; Step 4.2, according to within the scheduling period the estimated value of the dynamic adjustment factor at the moment and within the scheduling period the control quantity at the moment determines the control inheritance item; Step 4.3, according to within the scheduling period the estimated value of the dynamic adjustment factor at the moment, within the scheduling period the disturbance estimated value at the moment and within the scheduling period the water level error determines the disturbance correction term; Step 4.

4. determining the control variable at the time instant in accordance with the sum of the control inheritance term and the disturbance correction term within a scheduling period the control variable at the time instant.

2. The model-free adaptive control method for reservoir scheduling according to claim 1, wherein, Further comprising after step 2.3: When Within the scheduling period The disturbance estimation value at the time is greater than a preset saturation value constant, Within the scheduling period The disturbance estimation value at the time is a saturation value constant; When within the scheduling period the disturbance estimate at the time instant is less than a negative saturation value constant, within the scheduling period the disturbance estimate at the time instant is the negative saturation value constant.

3. The model-free adaptive control method for reservoir scheduling according to claim 1, wherein, The control inheritance item in step 4.2 is determined according to a first formula, which is: ; wherein is an energy consumption penalty factor; is a weight; is the control quantity at the time instant within the scheduling period the estimated value of the dynamic adjustment factor at the time instant; is the control quantity at the time instant within the scheduling period the control quantity at the time instant within the scheduling period 4. The model-free adaptive control method for reservoir scheduling according to claim 3, wherein, The disturbance correction term in step 4.3 is determined according to a second formula, which is: ； wherein is within a dispatch period an estimate of the dynamic factor at the time instant; is an energy consumption penalty factor; is a weight; is within a dispatch period a water level error at the time instant, is within a dispatch period an actual observed water level at the time instant, is within a dispatch period a predicted water level at the time instant; is within a dispatch period a disturbance estimate at the time instant.

5. The model-free adaptive control method for reservoir scheduling according to claim 1, wherein, Further comprising before step 1: The reservoir is in The machine learning model is trained with the historical control amount before the dispatch cycle as input and the historical water level value as output to obtain The initial control amount of the dispatch cycle.

6. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor implements the steps of the reservoir scheduling method of model-free adaptive control according to any one of claims 1-5 when executing the computer program.

Images