A guide vane opening degree self-adaptive control method for a hydroelectric unit

Abstract

The present application relates to model adaptive control technical field, specifically for a kind of guide vane opening adaptive control method for hydroelectric generating set, including the following steps, in the present application, by adopting radial basis function network Gaussian node dynamic addition and deletion and weight optimization back mapping algorithm, guide vane relay actuator displacement instruction and actual execution stroke steady-state error absolute value is forced to converge to zero point zero five millimeter closed set, by adopting Kalman filter covariance recursive estimation and multidimensional space boundary constraint optimization algorithm, limit the speed overshoot under sudden load disturbance condition to be less than the threshold value of rated speed percentage three, to ensure that the frequency fluctuation damping ratio of output electric energy is greater than the target value of zero point two five, by adopting speed deviation feedforward multiplication deduction and steady-state threshold truncation reorganization algorithm, maintain the consistency of electrical control flow during the mutation of multiple unit operating conditions, improve the active power regulation dynamic response stability margin of hydroelectric generating set grid-connected operation.

Description

Technical Field

[0001] This invention relates to the field of model adaptive control technology, and in particular to an adaptive control method for guide vane opening of hydroelectric generators. Background Technology

[0002] The field of model adaptive control technology specifically addresses state convergence control schemes for nonlinear dynamic systems containing unknown and time-varying parameters. The operational steps include constructing a reference model that matches the input and output dimensions of the physical controlled object, setting the desired trajectories for multiple state variables, constructing the output error terms between the controlled object and the reference model, applying the energy decay criterion for nonlinear systems to derive the parameter update matrix operation logic, and then driving the dynamic correction of multiple weight elements of the controller gain matrix. This ensures that the pole configuration vector of the closed-loop system converges to a specified region in the left half-plane of the complex plane, and maintains bounded stability of the closed-loop transfer function output within the range of 20% to 50% of parameter perturbation amplitude.

[0003] An adaptive control method for guide vane opening of hydropower units specifically addresses the combined disturbances of nonlinear dead zone and elastic water hammer effect in the turbine regulating system's actuator. The proposed scheme aims to eliminate integral saturation and closed-loop oscillation divergence in the proportional-integral regulator caused by large-scale deviations in the water flow inertia time constant and the generator rotor inertia time constant. The scheme achieves several quantitative indicators, including forcibly converging the absolute value of the steady-state error between the guide vane relay displacement command and the actual executed stroke to within a 0.05 mm closed set, limiting the speed overshoot under sudden load disturbance conditions to below a 3% threshold of rated speed, and shrinking the transient adjustment time to within a single decaying oscillation period, thereby ensuring that the output power frequency fluctuation damping ratio is configured to be greater than the target value of 0.25.

[0004] The existing state convergence control scheme for nonlinear dynamic systems with unknown and time-varying parameters involves constructing a reference model that matches the input and output dimensions of the physical controlled object, setting the desired trajectory of multiple state variables, and applying the energy decay judgment rule of nonlinear systems to derive the operation logic of the parameter update matrix. This drives the dynamic correction of multiple weight elements of the controller gain matrix, resulting in a severe mismatch in the dynamic correction step size of the controller gain matrix weight elements. The scheme also constructs the output error terms of the controlled object and the reference model and applies a closed-loop system pole placement vector convergence mechanism, requiring the parameter perturbation amplitude to be strictly maintained within a closed interval of 20% to 50%. Once a sudden and severe load disturbance breaks through the restricted perturbation boundary interval, the inherent pole placement vector cannot converge to the specified region in the left half of the complex plane, causing a deep destruction of the bounded stability of the closed-loop transfer function output. The transient amplitude of the generator rotor inertial time constant exceeds the preset upper limit of 50%. The existing fixed-dimensional parameter update matrix cannot track the nonlinear dead zone and discontinuous time-varying characteristics, causing deep integral saturation of the proportional-integral regulator. The steady-state error of the control command and the executed displacement is drastically amplified, exceeding the tolerance of 0.05 mm, which in turn induces a large-scale closed-loop oscillation divergence phenomenon in the turbine regulating system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose an adaptive control method for guide vane opening of hydroelectric generators.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive control method for guide vane opening of hydroelectric generating units, comprising the following steps: S1: Based on the target and actual stroke parameters of the guide vane relay, extract the stroke difference to construct the error variable, solve the square root value of the variable, compare the square root value with the growth threshold to add nodes, check the absolute value accumulation term and the removal threshold to perform deletion, and establish a Gaussian node set. S2: Based on the Gaussian node set, extract the product of the radial basis function network state and the target and subtract the error to obtain the difference. Combine the difference and the leakage penalty term to extract the optimization step size. Add the step size to the node weight variable and substitute it into the piecewise equation to perform reverse mapping to generate the servo compensation matrix. S3: Based on the servo compensation matrix, substitute the time-varying pressure of the elastic water hammer into the reduced-order differential equation to solve the negative exponential integral term, extract the product of the integral term and the proportional coefficient to construct a composite variable, integrate the composite variable into the corresponding element inside the compensation matrix, and obtain the order switching manifold. S4: Based on the servo compensation matrix and the order switching manifold, the manifold elements are substituted into the Kalman filter to extract coefficients, the coefficients and matrix elements are merged to establish coordinates, the coordinates and travel boundaries are separated to extract the constraint set, the set is substituted into the set to solve the equation for the target independent variable, and the target gain sequence is generated. S5: Based on the target gain sequence, extract the difference between the actual and set frequency of the unit to construct the speed deviation, combine the deviation with the product of the first term gain of the sequence to extract the feedforward coefficient, splice the limit threshold and the coefficient to construct the electronic control signal, send the electronic control signal to the servo action terminal, and obtain the comprehensive frequency modulation command.

[0007] As a further embodiment of the present invention, the Gaussian node set includes reserved nodes, state variables, hidden layer center coordinate parameters, and basis width parameters; the servo compensation matrix includes inverse mapping compensation elements, a two-dimensional matrix, and multi-dimensional coordinate corresponding element parameters within the compensation loop; the order switching manifold includes multi-dimensional manifold elements, array arrangement state, and current control beat observation parameters; the target gain sequence includes valley-corresponding equation independent variables, multi-dimensional elements within the independent variables, and the first gain element of the sequence; and the integrated frequency modulation command includes composite data stream, time stamp parameters, and electronic control signals.

[0008] As a further aspect of the present invention, the specific steps for establishing the Gaussian node set are as follows: Based on the target and actual stroke parameters of the guide vane relay, the stroke difference is extracted to construct an error variable, the square root value of the error variable is calculated, the square root value is compared with the growth threshold to extract the corresponding coordinates, the coordinates are substituted to add new node items, and a new node set is established. Based on the newly added node set, the internal output absolute value parameter is extracted, the relationship between the absolute value accumulation term and the set elimination threshold is checked, the node elements corresponding to the elimination threshold are deleted, the state variables of the retained nodes are integrated, and a Gaussian node set is established.

[0009] As a further aspect of the present invention, the specific steps for generating the servo compensation matrix are as follows: Based on the Gaussian node set, the product value of the internal state parameters and the target parameters of the radial basis function network is extracted to obtain the system state following error parameter. The product value and the error parameter are subtracted to extract the corresponding difference. The difference values ​​are combined and arranged into a data array to obtain the node deviation vector. Based on the node deviation vector, the inherent leakage penalty term and the corresponding deviation parameter are extracted, the difference data and the leakage penalty term are merged to extract the optimization gradient, the optimization gradient is combined to calculate the specific optimization step size, the optimization step size is added to the node weight variable, and the optimization weight parameter is established. Based on the optimization weight parameters, the physical dead zone limit values ​​of the pressure regulating valve are extracted to construct a piecewise equation. The weight accumulation variables are substituted into the piecewise equation to perform reverse mapping calculation, obtain the reverse mapping compensation elements, arrange the compensation elements to reorganize the two-dimensional matrix, and generate the servo compensation matrix.

[0010] As a further aspect of the present invention, the radial basis function network, based on a Gaussian node set, extracts multiple hidden layer center coordinate parameters and basis width parameters within the radial basis function network. It calculates the Euclidean distance between the input control vector and the hidden layer center coordinate parameters, substitutes the Euclidean distance values ​​and the basis width parameters into a nonlinear exponential decay formula to perform mapping calculation, obtains the internal state parameters of the radial basis function network, calls a preset desired output trajectory target parameter, and performs a bit-by-bit aligned algebraic multiplication operation between the internal state parameters of the network and the desired output trajectory target parameter to calculate and extract the corresponding product value.

[0011] As a further aspect of the present invention, the specific steps for obtaining the order-switching manifold are as follows: Based on the servo compensation matrix, the time-varying pressure parameter of water hammer is extracted, and the pressure parameter is substituted into the reduced-order differential equation to solve the negative exponential integral term. The integral term and the proportional coefficient are multiplied to extract the product value, and the multidimensional product values ​​are combined to establish a negative exponential integral composite variable. Based on the negative exponential integral composite variable, extract the element parameters corresponding to the multidimensional coordinates inside the compensation link, add the current composite value to the corresponding element parameter to solve the displacement accumulation value, arrange the accumulation value to reconstruct the array arrangement state, and obtain the order switching manifold.

[0012] As a further aspect of the present invention, the specific steps for generating the target gain sequence are as follows: Based on the servo compensation matrix and the order switching manifold, the manifold elements are substituted into the recursive formula of the Kalman filter algorithm to extract historical coefficients. The historical coefficients are merged with the matrix elements to construct node parameters. The node parameters are combined and arranged into a multi-dimensional spatial array to obtain the coefficient mapping coordinates. Based on the coefficient-mapped coordinates, extract the mechanical motion stroke boundary parameters of the relay, subtract the deviation difference between the coordinate values ​​and the parameter solution, compare the deviation difference data to remove the points in the space that exceed the limit, aggregate the parameter data of compliant nodes inside the boundary, and establish an inward constraint set. Based on the set of inward constraints, the penalty term and cost are combined to construct the target equation. The parameters are substituted into the equation to solve for the target valley value. The independent variables corresponding to the valley value are extracted and the multi-dimensional elements inside the independent variables are arranged to generate the target gain sequence.

[0013] As a further aspect of the present invention, the Kalman filtering algorithm extracts multidimensional manifold elements based on the order-switching manifold as the current control beat observation parameters, calls the previous beat filtering state estimation variables and the preset state transition matrix to perform algebraic multiplication to obtain prior state estimation coefficients, extracts process noise covariance parameters and observation noise covariance parameters, combines error covariance parameters and state observation matrices to perform matrix inversion and continuous multiplication to obtain the Kalman gain matrix, substitutes the prior state estimation coefficients and state observation matrix to multiply and derive the predicted observation parameters, subtracts the multidimensional manifold elements and predicted observation parameters to solve for the innovation deviation value, multiplies the innovation deviation value and Kalman gain matrix to extract the state correction step size parameter, accumulates the state correction step size parameter to the prior state estimation coefficients to perform internal filtering state update, and extracts historical coefficients based on the updated state variables.

[0014] As a further aspect of the present invention, the deviation difference between the subtracted coordinate values ​​and the parameter solution is obtained by extracting multiple node coordinate values ​​within the coefficient mapping coordinates based on the coefficient mapping coordinates, extracting the motion elongation limit parameter and motion compression limit parameter included in the mechanical motion stroke boundary parameters of the relay device, and for a single node coordinate value, subtracting the motion elongation limit parameter from the node coordinate value to extract the first boundary algebraic difference, subtracting the node coordinate value from the motion compression limit parameter to extract the second boundary algebraic difference, splicing the first boundary algebraic difference and the second boundary algebraic difference to reconstruct a bidirectional deviation vector, calculating the Euclidean norm parameter corresponding to the bidirectional deviation vector, and integrating the arrangement combination of multiple node coordinate values ​​corresponding to the Euclidean norm parameter to obtain the deviation difference.

[0015] As a further aspect of the present invention, the specific steps for obtaining the integrated frequency modulation command are as follows: Based on the target gain sequence, extract the actual and set frequency parameters of the unit, subtract the frequency parameters to extract the speed deviation value and extract the first gain element of the sequence, multiply the deviation and the first gain element to solve the product, merge the product data to extract the action parameters, and obtain the feedforward control coefficient. Based on the feedforward control coefficients, the steady-state integral limit threshold parameter is extracted and the limit threshold and coefficient values ​​are concatenated to construct a composite data stream. The time label parameter is added to the data stream to generate an electronic control signal. The electronic control signal is transmitted to the servo action port to obtain the comprehensive frequency modulation command.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by employing a radial basis function network Gaussian node dynamic addition and deletion and weight optimization reverse mapping algorithm, and substituting the weight variables into the piecewise equation to perform reverse mapping to generate the servo compensation matrix, the nonlinear dead zone perturbation interference of the turbine regulating system actuator is overcome, and the absolute value of the steady-state error between the guide vane servo actuator displacement command and the actual execution stroke is forced to converge to a closed set of 0.05 mm. In this invention, by employing Kalman filter covariance recursive estimation and multidimensional space boundary constraint optimization algorithm, the overshoot of the speed under sudden load disturbance is limited to less than 3% of the rated speed threshold, and the adjustment time of the transient process is shortened to within a single decay oscillation period, thereby ensuring that the output power frequency fluctuation damping ratio is configured to be greater than the target value of 0.25. In this invention, by employing the feedforward multiplication method for speed deviation deduction and the steady-state threshold truncation and recombination algorithm, the difference between the actual and set frequencies of the unit is extracted to construct the speed deviation. This avoids the high-frequency control signal chattering phenomenon caused by the nonlinear extreme value optimization process, cuts off the integral saturation accumulation path of the control loop, maintains the consistency of the electrical control flow and the hydraulic servo conversion layer during multiple sudden changes in the operating conditions of the unit, and improves the dynamic response stability margin of the active power regulation of the hydropower unit in grid-connected operation. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Example 1 Please see Figure 1 This invention provides a technical solution: an adaptive control method for guide vane opening of hydroelectric generating units, comprising the following steps: S1: Based on the target and actual stroke parameters of the guide vane relay, extract the stroke difference to construct the error variable, solve the square root value of the variable, compare the square root value with the growth threshold to add nodes, check the absolute value accumulation term and the removal threshold to perform deletion, and establish a Gaussian node set. S2: Based on the Gaussian node set, extract the product of the radial basis function network state and the target and subtract the error to obtain the difference. Combine the difference and the leakage penalty term to extract the optimization step size. Add the step size to the node weight variable and substitute it into the piecewise equation to perform reverse mapping to generate the servo compensation matrix. S3: Based on the servo compensation matrix, substitute the time-varying pressure of the elastic water hammer into the reduced-order differential equation to solve the negative exponential integral term, extract the product of the integral term and the proportional coefficient to construct a composite variable, integrate the composite variable into the corresponding element inside the compensation matrix, and obtain the order switching manifold. S4: Based on the servo compensation matrix and the order switching manifold, the manifold elements are substituted into the Kalman filter to extract coefficients. The coefficients and matrix elements are merged to establish coordinates. The coordinates and travel boundaries are separated to extract the constraint set. The set is substituted into the set to solve the equation for the target independent variable and generate the target gain sequence. S5: Based on the target gain sequence, extract the difference between the actual and set frequency of the unit to construct the speed deviation, combine the deviation with the product of the first term gain of the sequence to extract the feedforward coefficient, splice the limit threshold and coefficient to construct the electronic control signal, send the electronic control signal to the servo action end, and obtain the comprehensive frequency modulation command.

[0020] 2. The adaptive control method for guide vane opening of hydropower units according to claim 1, characterized in that the Gaussian node set includes reserved nodes, state variables, hidden layer center coordinate parameters and base width parameters; the servo compensation matrix includes inverse mapping compensation elements, a two-dimensional matrix and multi-dimensional coordinate corresponding element parameters within the compensation link; the order switching manifold includes multi-dimensional manifold elements, array arrangement state and current control beat observation parameters; the target gain sequence includes valley corresponding equation independent variables, multi-dimensional elements within the independent variables and the first gain element of the sequence; and the integrated frequency modulation command includes composite data stream, time tag parameters and electronic control signals.

[0021] The specific steps for establishing the Gaussian node set are as follows: Based on the target and actual travel parameters of the guide vane relay, a self-organizing mapping spatial distance calculation mechanism is used to extract the travel difference and construct an error variable. Multiple initialization parameters, including a learning rate of 0.01 and a spatial decay constant of 1.5, are configured. The difference between the target parameter of 450 mm and the actual travel parameter of 442 mm is calculated, outputting an 8 mm absolute difference as the error variable. The microprocessor numerical approximation instruction set is called to perform a power-of-half algebraic iteration on the 8 mm absolute difference input source, outputting a 2.828 mm square root value. A preset growth threshold is loaded. Specifically, the preset growth threshold reads historical displacement time series data from 1000 generator sets operating at full load within the terminal, uses a sliding time window to extract the average of the maximum steady-state deviation within 100 sampling periods, defines a fixed benchmark of 2.500 mm, and executes a hardware comparison instruction to determine whether the 2.828 mm square root value is greater than the 2.500 mm fixed benchmark. The current 50-millisecond control cycle time stamp and the real-time elastic water hammer pressure value of 1.2 MPa are extracted by the pulse trigger. The current guide vane opening command is combined to form a three-dimensional input coordinate. The Gaussian topology network dynamic expansion algorithm is used to add new node items to the three-dimensional input coordinate. The maximum boundary address offset of the current storage area plus 1 is configured to insert the memory block index parameter. The data column width alignment mode is set to strict three-dimensional floating point matching. The three-dimensional input coordinate values ​​are appended to the internal multi-dimensional structure with an initial memory size of zero bytes according to the continuous memory allocation mode. A high-frequency activity fixed weight of 0.85 is added to the input values ​​to form a spatially independent new row vector to serve as the new node item. A globally unique node identifier code is assigned to generate a sequence auto-incrementing integer value and write it to the underlying cache page. The parsing of multiple data streams within 10 consecutive 50-millisecond control cycles is performed in a loop. Multiple associated coordinates are substituted to complete the continuous filling and arrangement operation of the new node item, generating a set of new nodes. Based on the newly added node set, a sparse network pruning and order reduction algorithm is used to extract the internal output absolute value parameters. The activation response amplitudes of 50 storage nodes within the set are read column-wise. The floating-point unit is called to remove the positive and negative flag bits and extract the internal output absolute value parameters. The sliding time window length is configured to be 20 sampling periods. Scalar accumulation operations are performed on multiple absolute value parameters within the time window, outputting a 0.015 absolute value accumulation term. The relationship between the 0.015 absolute value accumulation term value and the set rejection threshold value is verified. Specifically, the rejection threshold is set by collecting 500 sets of radial basis hidden layer response peak data through an offline simulation system full-load load shedding operation experiment. The average energy decay ratio of the low-frequency invalid response is calculated and written into the solidified memory to generate a constant value of 0.020, triggering the microprocessor logic comparator for comparison. The absolute value of the accumulated item is 0.015, and the constant value is 0.020. The logic judgment level signal is received when the value is below the removal threshold. The memory linked list disconnection and release rule is used to delete the associated node elements below the removal threshold. The physical storage address offset of the non-compliant node element is located according to the signal. The memory pointer of the predecessor node is modified to bypass the invalid address and directly point to the starting physical position of the successor node. The 16-byte memory block occupied by the associated node elements below the removal threshold is reclaimed and the physical erase operation is performed to completely delete the corresponding items. For the remaining multiple nodes greater than or equal to the removal threshold, the original address memory mapping content is extracted and the continuous block rearrangement is performed to block the memory fragmentation phenomenon. After integration and rearrangement, the internal dynamic activation coordinates of multiple retained nodes are stripped and merged into a one-dimensional floating-point array structure. The state variables are extracted from the internal structure to generate a Gaussian node set.

[0022] The specific steps for generating the servo compensation matrix are as follows: Based on a Gaussian node set, the Hadamard product matrix algebraic calculation formula is used to extract the product values ​​of the internal state parameters and target parameters of the radial basis function network. The bottom-level floating-point unit bit width parameter is configured to be 64 bits, and a single instruction multiple data stream concurrent execution mode is enabled. The system reads 50 sets of internal state parameter values ​​of the radial basis function network and 50 sets of preset expected running trajectory target parameter values ​​from the memory in row priority order. Specifically, the preset expected running trajectory target parameters are obtained by reading the historical command sequence of power-on no-load issued by the hydropower station monitoring host computer, extracting 10 discrete sampling points within 500 milliseconds in the time dimension, writing them into static random access memory to generate target parameters, aligning the state parameters and target parameters bit by bit with the input sources, and performing floating-point multiplication algebraic operations. The system outputs a product value of 0.450, calls the hardware subtractor to read the product value of 0.450 and subtracts the input reference value of 0.050 to extract the difference value of 0.400, and obtains the system state following error parameter. Subtracting the product value and the error parameter, the corresponding difference value is extracted. The arithmetic logic unit is called to perform the operation of subtracting the system state following error parameter value of 0.400 from the product value of 0.450 and outputting the corresponding difference value of 0.050. The dynamic memory block arrangement instruction set is used to combine the difference arrangement data array. The base address offset and the continuous 16-byte step size write instruction are issued to the direct memory access controller. The 50 0.050 corresponding difference values ​​are pushed into the stack space in sequence according to the one-dimensional continuous physical address mapping rule to reassemble the one-dimensional floating-point row arrangement data array and obtain the node deviation vector. Based on the node deviation vector, the Lyapunov gradient descent adaptive optimization algorithm is used to extract the inherent leakage penalty term and its corresponding deviation parameter. The microcontroller's flash memory is used to read the 0.001 inherent leakage penalty term constant value. Specifically, the inherent leakage penalty term constant value is generated by substituting the nonlinear system energy decay law equation into offline simulation software, performing 1000 iterations, truncating the lower bound constant of convergence, and storing it in read-only memory. The 0.050 corresponding deviation parameter within the node deviation vector is extracted column-by-column. The difference data and the leakage penalty term are merged to extract the optimization gradient. The accumulator register is triggered to perform the addition of the 0.001 inherent leakage penalty term value to the 0.050 corresponding deviation parameter, outputting a composite value of 0.051. The composite value of 0.051 is multiplied by the pre-allocated network learning rate constant of 0.020 to output a value of 0.00102 to generate the optimization gradient. The optimization gradient is combined with the specific optimization step size to calculate. The 0.00102 optimization gradient value is substituted into the step size ratio allocation integral equation. The accumulation step parameter of the integral term is configured to be 10 steps. The numerical integration operation is performed to extract the specific optimization step size of 0.0102. The optimization step size is appended to the node weight variable. The baseline value of 0.500 of the network node weight variable in the current control cycle is read. The floating-point adder is called to accumulate the specific optimization step size of 0.0102 and the baseline value of 0.500 to output an updated value of 0.5102. The updated value is written to the address of the network node weight variable register in the current control cycle to establish the optimization weight parameter. Based on the optimization weight parameter, a nonlinear dead-zone inverse mapping piecewise equation algebraic solution is used to extract the physical dead-zone limit value of the pressure regulating valve and construct a piecewise equation. The servo controller hardware configuration register is read to extract the forward opening dead-zone limit value of 0.020 mm and the reverse closing dead-zone limit value of -0.020 mm. Logical judgment statements are configured to divide the forward interval, reverse interval, and zero interval, and piecewise equations are constructed. The weight accumulation variable is substituted into the piecewise equation to perform inverse mapping solution, extracting the 0.5102 weight accumulation variable inside the optimization weight parameter. A floating-point comparison instruction is triggered to determine that the 0.5102 weight accumulation variable is greater than zero, and the forward interval algebraic operation is initiated. The system calls the arithmetic logic unit to perform a 0.5102 weight accumulation variable plus a 0.020 mm positive dead zone limit value operation, outputting a 0.5302 value to obtain the reverse mapping compensation element. It uses a two-dimensional array memory mapping rearrangement mechanism to arrange the compensation element and reorganize the two-dimensional matrix. The matrix row number parameter is configured as 50 and the column number parameter is configured as 1. It requests a 50-row, 1-column contiguous physical memory block from the memory management unit, sets the data access pointer offset step size to 8 bytes per access, and cyclically extracts 50 0.5302 reverse mapping compensation element values ​​and writes them sequentially to the corresponding row address and column address intersection point of the contiguous physical memory block to complete the multi-dimensional spatial element arrangement and generate the servo compensation matrix.

[0023] The specific steps to obtain the order-switching manifold are as follows: Based on the servo compensation matrix, the Euler numerical integral differential discretization approximation algorithm is used to extract the time-varying pressure parameters of water hammer. The sampling frequency constant of the data acquisition channel is configured to be 100 Hz. The analog-to-digital converter hardware low-level instructions are called to read the 24-bit binary time-varying data from the pressure sensor port. The binary data is input to the floating-point conversion module to perform a fixed-point to floating-point operation and output the 1.5 MPa water hammer time-varying pressure parameter. The pressure parameter is substituted into the reduced-order differential equation to solve for the negative exponential integral term. The discretization step size parameter of the reduced-order differential equation is configured to be set to 0.0. At 1 second, the microprocessor's arithmetic unit is invoked to read the time-varying pressure parameter of 1.5 MPa water hammer. The water flow inertia time constant is set to 2.5 seconds. The reciprocal of 2.5 seconds is executed and multiplied by the time-varying pressure parameter of 1.5 MPa water hammer, outputting an intermediate variable of 0.6. The current timestamp of 5.0 seconds is read from the system clock register to extract the negative exponential time decay base. The underlying natural logarithm base nonlinear exponentiation instruction is invoked to perform a base exponentiation operation on the 5.0-second timestamp to extract the negative exponential decay constant of 0.0067. The intermediate variable of 0.6 is then input and multiplied by 0.006. The negative exponential decay constant is added to the accumulator register. Ten consecutive values ​​are integrated and accumulated, outputting a value of 0.0402 as the negative exponential integral term. This integral term is multiplied by the proportional coefficient to extract the product. A preset proportional coefficient is then loaded. Specifically, the preset proportional coefficient is calculated by reading 100 sets of overshoot peak values ​​from the historical full-load step response test records of the hydro-generator governor, using the arithmetic mean method, and multiplying this average by a fixed gain proportional constant of 0.8. This result is then written to read-only memory to generate a proportional coefficient of 0.4. Finally, a floating-point multiplier is called with an input of 0. The .0402 value and the 0.4 proportional coefficient perform single-precision floating-point multiplication algebraic operation to output a product value of 0.01608. The multidimensional product values ​​are merged, and the direct memory access channel parameters are configured to set the source address to the current product value register and the destination address to the first address of the continuous multidimensional array. The length of a single data block is set to 16 bytes. The memory transfer start instruction is activated to control the data bus to sequentially extract 50 product values ​​of 0.01608 and push them into the stack memory space in row priority order to create a negative exponential integral composite variable. Based on the negative exponential integral composite variable, a multidimensional matrix element addressing mapping and phase space reconstruction rule are used to extract the corresponding element parameters of the multidimensional coordinates within the compensation stage. The memory management unit address offset divided by the address word length parameter is configured to be in 8-byte alignment mode. An indirect addressing instruction sequence is initiated, reading the servo compensation matrix base address pointer, adding the current loop count value multiplied by the row offset constant to locate the target physical memory block address, and calling a load instruction to push the corresponding physical address stored value into the CPU floating-point register queue cache. The corresponding element parameter of 0.5302 is output, and the current composite value is added to the corresponding element parameter to calculate the shift accumulation value. The negative exponential integral composite variable internally stores the composite value of 0.01608 from the stack memory. The arithmetic logic unit is configured to execute a saturation addition instruction, inputting the composite value of 0.01608 and 0. The .5302 parameter corresponds to the element parameter, which controls the hardware adder to perform binary addition arithmetic operations and outputs a composite value of 0.54628. The value comparison circuit is called to determine whether the composite value of 0.54628 has overflowed. The preset 64-bit floating-point maximum limit value is used as the displacement accumulation value. The accumulated values ​​are arranged to reconstruct the array arrangement state. The multi-dimensional space state is set to reconstruct the grid dimension parameter, which is a 50-row, 1-column two-dimensional tensor topology. The microcontroller memory data block movement instruction is called, and the destination address is specified as the pre-cleared initial grid memory page. The memory write handshake level signal is sent 50 times in a loop. The 50 displacement accumulation values ​​of 0.54628 are filled into the physical cell of the corresponding row and column coordinate intersection point of the pre-cleared initial grid memory page in ascending order of index number. The full value filling is completed to form the array arrangement state and the order switching manifold is obtained.

[0024] The specific steps for generating the target gain sequence are as follows: Based on the servo compensation matrix and the order-switching manifold, a Kalman filter state-space recursive estimation algorithm is used to extract historical coefficients by substituting manifold elements into the recursive formula of the Kalman filter algorithm. The initial state error covariance scalar value is set to 0.1, the process noise variance constant is set to 0.01, and the observation noise variance constant is set to 0.05. The microcontroller memory read instruction is called to extract the value of 0.54628, which is the element contained in the order-switching manifold. The floating-point multiplication and addition arithmetic logic unit is called to input the scalar value of 0.1 and the process noise variance constant of 0.01 to calculate and extract the prior state covariance value of 0.11. The prior state covariance value of 0.11 and the observation noise variance constant of 0.05 are input to perform a combined division and multiplication operation to output the Kalman gain constant of 0.6875. The Kalman gain constant of 0.6875 is multiplied by the measurement residual value and added to the prior estimated state number. The output value of 0.521 is used as the historical coefficient for the posterior state estimation result. The arithmetic logic unit vector addition instruction is used to merge the historical coefficient and matrix elements to construct node parameters. The value of the matrix element of 0.5302 in a specific physical storage unit inside the servo compensation matrix is ​​read. The floating-point adder hardware circuit is triggered to input the historical coefficient of 0.521 and the value of the matrix element of 0.5302 to perform binary floating-point addition operation and output the value of 1.0512 to construct node parameters. The tensor cross multiplication dimension reduction and recombination mechanism is used to combine the node parameters and arrange them into a multidimensional spatial array. The instruction to request a continuous storage block is sent to the underlying memory management unit to obtain a memory address segment containing 50 floating-point units. The direct memory access controller channel is started, and the single displacement step size parameter is set to 8 bytes. The 50 node parameter values ​​of 1.0512 are pushed into the continuous memory address segment to reconstruct the state of the one-dimensional row vector spatial array and obtain the coefficient mapping coordinates. Based on coefficient-mapped coordinates, a hard-boundary hyperplane truncation constraint algorithm is used to extract the mechanical motion travel boundary parameters of the relay. Preset relay mechanical motion travel boundary parameters are loaded, specifically as follows: When the hydro-generator unit is in a shutdown and physically disconnected state, maximum opening and full closing commands are issued to the electro-hydraulic servo distribution valve. Real-time feedback endpoint level signals from linear displacement sensors are collected and converted into 500 mm elongation limit action values ​​and 0 mm compression limit action values. These values ​​are then written to a designated sector of a read-only memory chip to generate the preset relay mechanical motion travel boundary parameters. The bus read protocol is called to extract the 500 mm elongation limit action values ​​and 0 mm compression limit action values. The deviation difference between the subtracted coordinate values ​​and the calculated parameters is then calculated. The 1.0512 coordinate value within the coefficient-mapped coordinates is multiplied by the dimension conversion constant 200, outputting an equivalent travel value of 210.24 mm. The central processing unit's logical subtraction instruction is called to subtract the 210.24 mm equivalent travel value from the 500 mm elongation limit action value, outputting a positive deviation of 289.76 mm. The value is calculated by subtracting the 0 mm compression limit action value from the 210.24 mm equivalent travel value, and outputting a 210.24 mm reverse deviation difference. The hardware comparator is called to compare the deviation difference data and perform the operation of stripping the over-limit space point. The positive deviation difference value of 289.76 mm and the reverse deviation difference value of 210.24 mm are input to the two channels of the comparator. If it is determined that the difference values ​​on both sides are greater than the 0 mm absolute over-limit reference, a low-level compliance signal is output to indicate that the over-limit state is not exceeded. If a negative value difference of less than 0 mm is detected, a high-level interrupt flag is triggered and a memory erase instruction is called to clear the internal value of the associated physical address to complete the stripping of the over-limit space point. The memory fragment reorganization and aggregation mechanism is used to aggregate the internal compliance node parameter data of the boundary. Multiple compliance low-level flags are scanned inside the continuous storage page and associated with the original 1.0512 node parameter values. The continuous memory data block transfer instruction is started. The source address is set as the discrete compliance physical storage node, and the destination address is set as the starting offset of the newly allocated continuous stack memory. 50 consecutive compliance data loop push and transfer operations are performed to aggregate multiple floating-point values ​​and establish an inward constraint set. Based on the inward constraint set, a quadratic cost functional derivative-free optimization algorithm is used to combine penalty terms and costs to construct the target equation. Preset penalty term values ​​and cost weight values ​​are loaded. Specifically, the preset operation involves calling the host computer simulation workstation to run 1000 sets of Monte Carlo parameter perturbation random target experiments on the generator set's no-load closed-loop control model. The product of the settling time and overshoot is extracted and associated with a 0.85 cost weight constant and a 0.15 control input penalty constant. These are then sent to the programmable logic controller's internal static random access memory to generate the preset penalty term values ​​and cost weight values. The controller's multiplication and addition instruction set is called to extract the square values ​​of the internal parameters of the 1.0512 inward constraint set, multiplying them by the 0.85 cost weight constant, and adding the square values ​​of the independent variables to multiply by the 0.15 control input penalty constant. These are combined and added to construct a parabolic target equation. The parameters are then substituted into the equation to solve for the target valley value. The lower bound of the independent variable optimization interval constant is configured to be 1.0. With an upper bound of 10.0, the single-iteration discrete search step size constant is configured to be 0.1. The loop traversal operator is started, and the discrete values ​​of multiple independent variables are substituted into the parabolic target equation to calculate the substitution value and output the multiple result values. The array minimum value hardware searcher array is called to compare the multiple result values ​​and extract the minimum cost associated with the 0.025 minimal floating-point value as the target valley value. The valley value is extracted and associated with the independent variables of the equation, and the multi-dimensional elements inside the independent variables are arranged. The storage address of the 0.025 minimal floating-point value is read along with the associated index mapping table to extract the mapped 2.5 independent variable value. The 2.5 independent variable value is parsed and split into 2.0 proportional control elements and 0.5 integral control elements multi-floating-point variables. The direct memory access controller is called to configure the base address offset addressing mode. The 2.0 proportional control elements are pushed into the first address of the contiguous memory page, and the 0.5 integral control elements are pushed into the first address of the contiguous memory page with an additional 4-byte offset position. The one-dimensional gain row arrangement array structure is reorganized to generate the target gain sequence.

[0025] The specific steps to obtain the integrated frequency modulation command are as follows: Based on the target gain sequence, a single-precision floating-point differential feedforward multiplication algorithm is used to extract the actual and set frequency parameters of the generator unit. The data acquisition bus sampling period constant is configured to be 10 milliseconds. The underlying analog-to-digital conversion channel of the synchronous phasor measurement device is called to read the zero-crossing time difference value of the real-time voltage waveform of the generator terminals. The time difference value is converted to generate a 50.05 Hz actual frequency parameter. The preset set frequency parameter is loaded. Specifically, the preset set frequency parameter is generated by writing the standard grid operation reference waveform frequency issued by the active power given node of the power dispatch center into a specified sector of the local controller's read-only memory to generate a fixed value of 50.00 Hz. The arithmetic logic unit is called to perform subtraction frequency parameter extraction to extract the speed deviation value. The 50.05 Hz actual frequency parameter and the 50.00 Hz set frequency parameter are input to the subtractor pin to perform binary floating-point subtraction. The operation outputs a 0.05 Hz value to extract the speed deviation value. It calls the memory addressing pointer to send a base address access instruction to extract the first gain element of the sequence. It reads the 2.0 proportional control element as the first gain element by pointing to the first 4-byte storage offset segment of the target gain sequence. It triggers the hardware multiplier module to perform the product operation of the deviation and the first gain element. It inputs the 0.05 Hz speed deviation value and the 2.0 proportional control element to perform single-precision floating-point multiplication algebra operation to output a 0.10 product value. It uses the register bit splicing logic arrangement mechanism to merge the product data to extract the action parameters. It shifts the 0.10 product value to the high 16 bits of the 32-bit data bus and fills the low 16 bits with the pre-allocated servo proportional valve action identifier code 1011 binary data combination to obtain the combined data stream as the action parameters and obtain the feedforward control coefficient. Based on the feedforward control coefficients, a multi-source data frame encapsulation and hard real-time bus scheduling protocol are used to extract the steady-state integral limit threshold parameter. The preset steady-state integral limit threshold parameter is loaded. Specifically, it involves reading the steady-state control dead zone data of 1000 step disturbance responses from the historical trial operation archives of the hydropower station's unit speed control system, calculating the central extreme point using the arithmetic mean, and storing it in the programmable controller's non-volatile memory to generate a 0.05 mm constant value. A microprocessor flash memory read instruction is called to extract the 0.05 mm constant value as the steady-state integral limit threshold parameter. A shift register array is used to concatenate the limit threshold and coefficient values ​​to construct a composite data stream. The underlying data frame assembly bit width is configured to a 64-bit standard format. The binary code stream containing the 0.05 mm steady-state integral limit threshold parameter is written into the lower 32 bits of the data segment, and the feedforward control coefficient containing the 0.10 value and identification code information is written into the higher 32 bits. The system performs bitwise logical addition operations to construct a 64-bit composite data stream. It then calls the system's real-time clock chip to add a time stamp parameter to the data stream to generate electrical control signals. The system sends and reads handshake levels to extract the current control time node, accurate to milliseconds, and associates it with a 16-bit absolute timestamp as the time stamp parameter. A bit-joining hardware accelerator appends and binds the 16-bit timestamp to the end of the 64-bit composite data stream, generating an 80-bit fixed-length data packet to output the electrical control signal. An Ethernet control automation technology real-time bus network card is configured to transmit the electrical control signal to the servo action port. The bus communication baud rate is set to 100 megabits per second, and the transmission timeout is set to 2 milliseconds. Through the underlying direct memory access engine, the 80-bit electrical control signal is pushed into the network card's first-in-first-out (FIFO) buffer queue and triggers a high-low level conversion pulse on the physical layer's transmit pin. The data packet is then sent along the twisted-pair medium to the signal receiving port pin of the hydraulic servo actuator to obtain the integrated frequency modulation command.

[0026] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An adaptive control method for guide vane opening of a hydroelectric generator, characterized in that, Includes the following steps: S1: Based on the target and actual stroke parameters of the guide vane relay, extract the stroke difference to construct the error variable, solve the square root value of the variable, compare the square root value with the growth threshold to add nodes, check the absolute value accumulation term and the removal threshold to perform deletion, and establish a Gaussian node set. S2: Based on the Gaussian node set, extract the product of the radial basis function network state and the target and subtract the error to obtain the difference. Combine the difference and the leakage penalty term to extract the optimization step size. Add the step size to the node weight variable and substitute it into the piecewise equation to perform reverse mapping to generate the servo compensation matrix. S3: Based on the servo compensation matrix, substitute the time-varying pressure of the elastic water hammer into the reduced-order differential equation to solve the negative exponential integral term, extract the product of the integral term and the proportional coefficient to construct a composite variable, integrate the composite variable into the corresponding element inside the compensation matrix, and obtain the order switching manifold. S4: Based on the servo compensation matrix and the order switching manifold, the manifold elements are substituted into the Kalman filter to extract coefficients, the coefficients and matrix elements are merged to establish coordinates, the coordinates and travel boundaries are separated to extract the constraint set, the set is substituted into the set to solve the equation for the target independent variable, and the target gain sequence is generated. S5: Based on the target gain sequence, extract the difference between the actual and set frequency of the unit to construct the speed deviation, combine the deviation with the product of the first term gain of the sequence to extract the feedforward coefficient, splice the limit threshold and the coefficient to construct the electronic control signal, send the electronic control signal to the servo action terminal, and obtain the comprehensive frequency modulation command.

2. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The Gaussian node set includes reserved nodes, state variables, hidden layer center coordinate parameters, and basis width parameters. The servo compensation matrix includes inverse mapping compensation elements, a two-dimensional matrix, and multi-dimensional coordinate corresponding element parameters within the compensation loop. The order switching manifold includes multi-dimensional manifold elements, array arrangement state, and current control beat observation parameters. The target gain sequence includes valley-corresponding equation independent variables, multi-dimensional elements within the independent variables, and the first gain element of the sequence. The integrated frequency modulation command includes composite data streams, time stamp parameters, and electronic control signals.

3. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The specific steps for establishing the Gaussian node set are as follows: Based on the target and actual stroke parameters of the guide vane relay, the stroke difference is extracted to construct an error variable, the square root value of the error variable is calculated, the square root value is compared with the growth threshold to extract the corresponding coordinates, the coordinates are substituted to add new node items, and a new node set is established. Based on the newly added node set, the internal output absolute value parameter is extracted, the relationship between the absolute value accumulation term and the set elimination threshold is checked, the node elements corresponding to the elimination threshold are deleted, the state variables of the retained nodes are integrated, and a Gaussian node set is established.

4. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The specific steps for generating the servo compensation matrix are as follows: Based on the Gaussian node set, the product value of the internal state parameters and the target parameters of the radial basis function network is extracted to obtain the system state following error parameter. The product value and the error parameter are subtracted to extract the corresponding difference. The difference values ​​are combined and arranged into a data array to obtain the node deviation vector. Based on the node deviation vector, the inherent leakage penalty term and the corresponding deviation parameter are extracted, the difference data and the leakage penalty term are merged to extract the optimization gradient, the optimization gradient is combined to calculate the specific optimization step size, the optimization step size is added to the node weight variable, and the optimization weight parameter is established. Based on the optimization weight parameters, the physical dead zone limit values ​​of the pressure regulating valve are extracted to construct a piecewise equation. The weight accumulation variables are substituted into the piecewise equation to perform reverse mapping calculation, obtain the reverse mapping compensation elements, arrange the compensation elements to reorganize the two-dimensional matrix, and generate the servo compensation matrix.

5. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The radial basis function network, based on a Gaussian node set, extracts multiple hidden layer center coordinate parameters and basis width parameters within the radial basis function network. It calculates the Euclidean distance between the input control vector and the hidden layer center coordinate parameters, substitutes the Euclidean distance values ​​and basis width parameters into a nonlinear exponential decay formula to perform mapping calculation, obtains the internal state parameters of the radial basis function network, calls the preset desired output trajectory target parameters, and performs a bit-by-bit aligned algebraic multiplication operation on the internal state parameters of the network and the desired output trajectory target parameters to calculate and extract the corresponding product values.

6. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The specific steps for obtaining the order-switching manifold are as follows: Based on the servo compensation matrix, the time-varying pressure parameter of water hammer is extracted, and the pressure parameter is substituted into the reduced-order differential equation to solve the negative exponential integral term. The integral term and the proportional coefficient are multiplied to extract the product value, and the multidimensional product values ​​are combined to establish a negative exponential integral composite variable. Based on the negative exponential integral composite variable, extract the element parameters corresponding to the multidimensional coordinates inside the compensation link, add the current composite value to the corresponding element parameter to solve the displacement accumulation value, arrange the accumulation value to reconstruct the array arrangement state, and obtain the order switching manifold.

7. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The specific steps for generating the target gain sequence are as follows: Based on the servo compensation matrix and the order switching manifold, the manifold elements are substituted into the recursive formula of the Kalman filter algorithm to extract historical coefficients. The historical coefficients are merged with the matrix elements to construct node parameters. The node parameters are combined and arranged into a multi-dimensional spatial array to obtain the coefficient mapping coordinates. Based on the coefficient-mapped coordinates, extract the mechanical motion stroke boundary parameters of the relay, subtract the deviation difference between the coordinate values ​​and the parameter solution, compare the deviation difference data to remove the points in the space that exceed the limit, aggregate the parameter data of compliant nodes inside the boundary, and establish an inward constraint set. Based on the set of inward constraints, the penalty term and cost are combined to construct the target equation. The parameters are substituted into the equation to solve for the target valley value. The independent variables corresponding to the valley value are extracted and the multi-dimensional elements inside the independent variables are arranged to generate the target gain sequence.

8. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The Kalman filter algorithm extracts multidimensional manifold elements based on the order-switching manifold as the current control cycle observation parameters. It calls the previous cycle's filtered state estimation variables and the preset state transition matrix to perform algebraic multiplication to obtain prior state estimation coefficients. It extracts process noise covariance parameters and observation noise covariance parameters. It combines the error covariance parameters and the state observation matrix to perform matrix inversion and continuous multiplication to obtain the Kalman gain matrix. It substitutes the prior state estimation coefficients and the state observation matrix to derive the predicted observation parameters. It subtracts the multidimensional manifold elements and the predicted observation parameters to solve for the innovation deviation value. It multiplies the innovation deviation value and the Kalman gain matrix to extract the state correction step size parameter. It accumulates the state correction step size parameter to the prior state estimation coefficients to perform internal filtering state updates. It extracts historical coefficients based on the updated state variables.

9. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 7, characterized in that, The deviation difference between the subtracted coordinate values ​​and the parameter solution is obtained by extracting multiple node coordinate values ​​within the coefficient mapping coordinates based on the coefficient mapping coordinates, extracting the motion elongation limit parameter and motion compression limit parameter included in the mechanical motion stroke boundary parameters of the relay, and extracting the first boundary algebraic difference by subtracting the motion elongation limit parameter from the node coordinate value for each node coordinate value, and extracting the second boundary algebraic difference by subtracting the node coordinate value from the motion compression limit parameter. The first boundary algebraic difference and the second boundary algebraic difference are concatenated to reconstruct a bidirectional deviation vector, and the Euclidean norm parameter corresponding to the bidirectional deviation vector is calculated. The arrangement and combination of the Euclidean norm parameter corresponding to the multiple node coordinate values ​​are integrated to obtain the deviation difference.

10. The adaptive control method for guide vane opening of hydroelectric generating units according to claim 1, characterized in that, The specific steps for obtaining the integrated frequency modulation command are as follows: Based on the target gain sequence, extract the actual and set frequency parameters of the unit, subtract the frequency parameters to extract the speed deviation value and extract the first gain element of the sequence, multiply the deviation and the first gain element to solve the product, merge the product data to extract the action parameters, and obtain the feedforward control coefficient. Based on the feedforward control coefficients, the steady-state integral limit threshold parameter is extracted and the limit threshold and coefficient values ​​are concatenated to construct a composite data stream. The time label parameter is added to the data stream to generate an electronic control signal. The electronic control signal is transmitted to the servo action port to obtain the comprehensive frequency modulation command.

Images