A frequency prediction-based hydroelectric generator power active control method, system, device and medium

Abstract

The application discloses a kind of based on frequency prediction's hydroelectric generating set power initiative control method, system, equipment and medium, belongs to power control technical field, including the real-time acquisition of hydroelectric generating set regulating system's basic operation data, and the basic operation data is preprocessed;Extract the time sequence characteristics of power grid frequency data, generate power grid frequency prediction sequence in future time window using neural network;Power grid frequency prediction sequence is converted into dynamic power feedforward compensation, and feedforward compensation optimization calculation is carried out, and feedforward compensation is superimposed to feedback controller output end, forms composite power instruction input value hydroelectric generating set;Using rolling horizon prediction strategy, neural network input sequence is updated in each control cycle, and future prediction result is dynamically corrected, to realize the control amount update based on actual frequency deviation.The application uses online adaptive algorithm to dynamically update compensation weight, considers prediction error and system performance target, and optimizes compensation distribution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power control, in particular to a method, system, device and medium for active power control of a hydroelectric generating set based on frequency prediction. BACKGROUND

[0002] Under the background of large-scale access of renewable energy such as wind power and photovoltaic power to the power grid, the frequency regulation of the power system faces higher dynamic response requirements. As a key resource for frequency support and inertia response of the power grid, the hydroelectric generating set undertakes an important task of power grid regulation due to its rapid start-stop and regulation capacity. In the scenario of high proportion of renewable energy grid connection, the rapid power response of the hydroelectric generating set not only helps to suppress frequency disturbance, but also improves system stability and safety margin, which is of great significance to the safe operation of the power grid.

[0003] At present, the power regulation of the hydroelectric generating set mainly relies on the hydraulic governor and proportional-integral controller of the generating set, and the feedback regulation of frequency deviation is completed through load distribution strategy and primary droop characteristic. The speed regulation system of the generating set sends increase / decrease instructions by receiving real-time frequency deviation to realize closed-loop control of power. However, due to the inherent time delay of the control link of the generating set and the flow characteristics, the traditional feedback control mode often has response lag when the frequency suddenly changes or fluctuates rapidly, and it is difficult to provide sufficient feedforward support in time. However, the existing methods generally ignore the prediction and utilization of the future situation of the grid frequency, and are still limited to feedback control based on real-time frequency measurement, which leads to the inability of the generating set to compensate in advance in the initial stage of frequency disturbance, and restricts its rapid response capability and support effect. At the same time, although existing research has attempted to apply wind and light power prediction tools to power system dispatching, the accurate prediction of grid frequency disturbance and the power feedforward compensation strategy of the hydroelectric generating set are still blank. Therefore, there is an urgent need for a composite control method that combines short-term prediction of grid frequency and dynamic characteristics of the generating set to break through the phase lag limitation of the traditional control structure and realize active power support of the hydroelectric generating set in the frequency disturbance propagation stage. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the technical problem solved by the present application is: how to solve the problem that the existing methods generally ignore the prediction and utilization of the future situation of the grid frequency, and are still limited to feedback control based on real-time frequency measurement, which leads to the inability of the generating set to compensate in advance in the initial stage of frequency disturbance, and restricts its rapid response capability and support effect. At the same time, although existing research has attempted to apply wind and light power prediction tools to power system dispatching, the accurate prediction of grid frequency disturbance and the power feedforward compensation strategy of the hydroelectric generating set are still blank.

[0006] To solve the above technical problems, the application provides the following technical scheme: a water turbine generator power active control method based on frequency prediction, which comprises the following steps: collecting basic operation data of a water turbine generator regulating system in real time, including a power grid historical frequency sequence, a unit output state and power grid load fluctuation information, and pre-processing the basic operation data; extracting time sequence characteristics of the power grid frequency data, generating a power grid frequency prediction sequence in a future time window by using a neural network; converting the power grid frequency prediction sequence into a dynamic power feedforward compensation amount, combining a preset power-frequency mapping relationship and a unit regulating constraint condition to perform feedforward compensation amount optimization calculation, superimposing the feedforward compensation amount to an output end of a feedback controller to form a composite power instruction input value water turbine generator; and adopting a rolling time domain prediction strategy to update a neural network input sequence in each control period, dynamically correcting a future prediction result, and realizing control amount updating based on an actual frequency deviation.

[0007] As a preferred scheme of the water turbine generator power active control method based on frequency prediction, the basic operation data of the water turbine generator regulating system is collected in real time, including collecting a power grid historical frequency data sequence to form an original frequency sequence; for abnormal values in the original frequency sequence, a sliding median filtering method is used to eliminate transient disturbances, and the filtered power grid frequency values are normalized; the normalized power grid frequency sequence satisfies a standard normal distribution.

[0008] As a preferred scheme of the water turbine generator power active control method based on frequency prediction, the time sequence characteristics of the power grid frequency data are extracted, and a neural network is used to generate a power grid frequency prediction sequence in a future time window, including defining a sliding window length and a prediction time domain according to a historical observation step; the sequence is divided by a sliding window to generate a training sample pair represented as:

[0009]

[0010] wherein, is the first group of input samples of the historical frequency window data; is the first group of output labels of a future frequency single-step prediction target; is a frequency sample index number, ; is a total sample number at a sampling time; is a sliding window length; is a real number set; is a order real number set matrix; is the first group of normalized power grid frequencies, is the first group of normalized power grid frequencies, normalized grid frequency of the first group normalized grid frequency of the first group normalized grid frequency of the first group normalized grid frequency of the first group

[0011] As a preferred scheme of the frequency prediction based active power control method for hydroelectric generating units, the conversion of the grid frequency prediction sequence into the dynamic power feedforward compensation amount comprises: dividing the grid frequency samples into data sets, letting the former group of samples be the training set and the latter group of samples be the validation set; using a neural network to perform frequency prediction, and the output grid frequency prediction sequence is represented as:

[0012]

[0013] wherein, is the grid frequency prediction sequence, is the grid frequency prediction value at time t, is the grid frequency prediction value at time t, is the prediction time domain, is the sampling time; based on the frequency prediction output sequence, the prediction frequency deviation is defined as:

[0014]

[0015] wherein, is the prediction frequency deviation, is the prediction frequency, is the actual frequency; for the frequency deviation sequence within the prediction time domain , the time-varying gain coefficient is designed as:

[0016]

[0017] wherein, is the time-varying gain coefficient, is the speed governor control coefficient, is the nonlinear gain coefficient, is the activation function, is the prediction step length, is the prediction frequency deviation at time t, is the frequency threshold.

[0018] ​​​​​​The preferred scheme converts the frequency prediction result into a timing deviation amount, and combines a time-varying nonlinear gain adjustment mechanism, so that the power compensation response has a dynamic adjustment capability; the designed gain coefficient combined with the activation function can enhance the control effect when the frequency deviation exceeds the threshold, and remain stable when the deviation is small; effectively enhance the adaptive adjustment capability of the hydroelectric generating set to the frequency anomaly, and provide faster response in the initial stage of disturbance.

[0019] As a preferred scheme of the hydroelectric generating set power active control method based on frequency prediction, the conversion of the power grid frequency prediction sequence into a dynamic power feedforward compensation amount further includes generating a feedforward compensation amount introduction time decay factor, dynamically adjusting the compensation weight of different time points in the prediction time domain through exponential weighting, and being expressed as:

[0020]

[0021] wherein, is a decay time constant, is a time decay factor, is a water flow inertia time constant of the water diversion system; and the dynamic feedforward compensation amount is expressed as:

[0022]

[0023] wherein, is a feedforward compensation amount, is a current prediction time.

[0024] The preferred scheme uses an exponential weighting method to introduce a time decay factor to differentially process the confidence of the deviation at different time points in the prediction time domain; can significantly reduce the influence of medium and long-term prediction errors on the stability of power regulation, and strengthen the control weight of short-term prediction results.

[0025] As a preferred scheme of the hydroelectric generating set power active control method based on frequency prediction, the combination of the preset power-frequency mapping relationship and the unit regulation constraint condition for feedforward compensation amount optimization calculation includes embedding the generated feedforward compensation amount into a PID control architecture, and the unit power regulation composite instruction after superimposing the feedforward compensation amount is expressed as:

[0026]

[0027] wherein, is a unit power regulation composite instruction, is a unit original power regulation instruction, is a control deviation, is a water turbine governor controller integral coefficient, is a water turbine governor controller differential coefficient; and The guide vane servo actuator of the input water turbine generator unit and the field frequency feedback quantity jointly constitute a secondary closed loop, and the output is dynamically corrected; the nonlinear relationship between the guide vane opening and the unit power regulation composite instruction is fitted based on the water turbine characteristic curve, the hyperbolic tangent function is introduced to describe the static characteristic, and the following expression is used:

[0028]

[0029] wherein, is the slope of the water turbine guide vane opening-power regulation curve, is a power normalization coefficient, is an opening offset coefficient; is a target guide vane opening; represents the hyperbolic tangent function; the unit guide vane servo system is driven by the pulse width modulation signal issued by the local execution unit, and the duty cycle-opening conversion formula is as follows:

[0030]

[0031] wherein, is the PWM duty cycle, is the minimum opening of the unit guide vane, is the maximum opening of the unit guide vane, is the minimum value of the effective range of the duty cycle, is the maximum value of the effective range of the duty cycle; after the mapping and conversion are completed, the pulse width modulation signal is output to drive the guide vane servo motor, so that the closed loop execution of the power instruction to mechanical action is realized.

[0032] As a preferred scheme of the water turbine generator unit power active control method based on frequency prediction, the rolling time domain prediction strategy is adopted to update the neural network input sequence at each control period, dynamically correct the future prediction result, realize the control amount update based on the actual frequency deviation, and the prediction is corrected in real time by combining the power grid frequency dynamic equation, as follows:

[0033]

[0034] wherein, is the system inertia constant, is the load damping coefficient, is the gain coefficient of the water turbine output power and the working water head and the guide vane opening, is the unit guide vane opening, is the unit working water head, is the frequency deviation.

[0035] The application provides a water turbine generator unit power active control system based on frequency prediction.

[0036] To solve the above technical problems, the application provides the following technical scheme: a frequency prediction-based active power control system for a hydroelectric generating unit, comprising: a power grid frequency acquisition and preprocessing module, a multi-scale time sequence prediction module, a dynamic compensation amount generation module and a composite control execution module; the power grid frequency acquisition and preprocessing module is used for acquiring basic operation data of a hydroelectric generating unit regulating system in real time, including a power grid historical frequency sequence, a unit output state and power grid load fluctuation information, and pre-processing the basic operation data; the multi-scale time sequence prediction module is used for extracting time sequence features of the power grid frequency data, and generating a power grid frequency prediction sequence in a future time window by using a neural network; the dynamic compensation amount generation module is used for converting the power grid frequency prediction sequence into a dynamic power feedforward compensation amount, performing feedforward compensation amount optimization calculation in combination with a preset power-frequency mapping relationship and a unit regulating constraint condition, superimposing the feedforward compensation amount to an output end of a feedback controller to form a composite power instruction input value hydroelectric generating unit; and the composite control execution module is used for updating a neural network input sequence in each control period by using a rolling time domain prediction strategy, dynamically correcting a future prediction result, and realizing control amount updating based on an actual frequency deviation.

[0037] The application provides a computer device, comprising a memory and a processor, and the memory stores a computer program, characterized in that the processor implements the steps of the frequency prediction-based active power control method for a hydroelectric generating unit when executing the computer program.

[0038] The application provides a computer readable storage medium, which stores a computer program, characterized in that the computer program is executed by a processor to implement the steps of the frequency prediction-based active power control method for a hydroelectric generating unit.

[0039] The application has the beneficial effects that: the multi-scale LSTM structure is adopted to model the frequency features of different time scales respectively, to realize joint prediction of macroscopic trends and microscopic fluctuations. The online adaptive algorithm is used to dynamically update the compensation weight, to optimize the compensation amount distribution by taking into account the prediction error and the system performance target. The frequency deviation information can be obtained before the actual disturbance occurs through the advanced prediction of the short-term fluctuation of the power grid frequency. The feedforward compensation amount is generated by dynamically adjusting the prediction compensation weight and superimposed to the closed-loop control loop, so that the hydroelectric generating unit can implement power increase or decrease in advance during the frequency disturbance propagation stage. The mechanism breaks through the phase lag limitation of the traditional PI control, shortens the unit power response time under the condition of 0.1pu load step by 11.2 seconds, and effectively improves the response speed of the hydroelectric generating unit to the power grid frequency disturbance. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and all other drawings obtained by those skilled in the art without creative effort based on these drawings should also fall within the protection scope of the present application.

[0041] Figure 1 A general flowchart of a frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application.

[0042] Figure 2 A comparison chart of LSTM prediction results and analysis data of a frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application.

[0043] Figure 3 A comparison of hydroelectric generating set power instruction signals under load step disturbance using different control methods of a frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application. Figure 1 .

[0044] Figure 4 A comparison of hydroelectric generating set power instruction signals under load step disturbance using different control methods of a frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application. Figure 2 .

[0045] Figure 5 A comparison chart of guide vane opening using different control methods of a frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application.

[0046] Figure 6 A comparison chart of hydroelectric generating set output power under load step disturbance using different control methods of a frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the above objectives, features and advantages of the present application more apparent and understandable, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and all other embodiments obtained by those skilled in the art without creative effort based on these embodiments should also fall within the protection scope of the present application.

[0048] Embodiment 1, refer to Figure 1 A frequency prediction-based active control method of power of a hydroelectric generating set according to an embodiment of the present application, which comprises:

[0049] S1, real-time acquisition of basic operation data of hydroelectric generating unit regulating system, including power grid historical frequency sequence, unit output state and power grid load fluctuation information, and preprocessing of the basic operation data.

[0050] S2, extracting time sequence characteristics of power grid frequency data, and generating power grid frequency prediction sequence in future time window by using neural network.

[0051] S3, converting the power grid frequency prediction sequence into dynamic power feedforward compensation amount, combining preset power-frequency mapping relationship and unit regulating constraint condition to perform feedforward compensation amount optimization calculation, superimposing the feedforward compensation amount to the output end of feedback controller to form a composite power instruction input value hydroelectric generating unit.

[0052] S4, adopting rolling time domain prediction strategy to update neural network input sequence in each control period, dynamically correcting future prediction results, and realizing control amount update based on actual frequency deviation.

[0053] Further, in step S1, the power grid historical frequency data sequence is collected, the sampling period is T s , and the original frequency sequence is formed:

[0054]

[0055] wherein, is the original frequency time sequence, is the power grid measured frequency at the i th sampling moment, i.e. the i th sampling sample of the power grid measured frequency, i.e. is the power grid measured frequency at the 1st moment, is the power grid measured frequency at the 2nd moment, is the power grid measured frequency at the i th sampling moment, is the total sample number at the sampling moment, is a real number set.

[0056] For the abnormal value in the historical frequency sequence, the sliding median filtering method is adopted to eliminate transient disturbance:

[0057]

[0058] wherein, is the median function, is the half-width of the filtering window, is the filtered frequency value, is the power grid measured frequency at the i th moment, is the power grid measured frequency at the i th moment.

[0059] To avoid the impact of numerical differences on model training, the filtered power grid frequency values ​​are normalized:

[0060]

[0061] in, This represents the mean of the filtered power grid frequency sequence; The standard deviation of the filtered power grid frequency sequence; This represents the normalized frequency value.

[0062] Normalized power grid frequency sequence It follows a standard normal distribution.

[0063] Furthermore, in step S2, the power grid frequency samples are constructed based on a sliding window. The normalized power grid frequency sequence is then... The dataset is converted into a supervised learning sample set to capture temporal dynamics. First, the sliding window length is defined based on the historical observation step size. With prediction time domain Considering the characteristics of hydropower units participating in power grid frequency regulation, and the time required for real-time communication and online calculation, then ≤10.

[0064] Sequence segmentation using a sliding window Generate training sample pairs :

[0065]

[0066] in, The first of the historical frequency window data Group input samples; The first step of the future frequency prediction target Group output labels; This represents the number of frequency sample indices. ; This represents the total number of samples at the sampling time. The length of the sliding window; It is the set of real numbers; for A matrix of order real numbers; For the normalized first The group's power grid frequency, For the normalized first The group's power grid frequency, For the normalized first The group's power grid frequency, For the normalized first The group's power grid frequency.

[0067] Further, the power grid frequency samples are divided into data sets in step S3, and the former group of samples is the training set, and the latter group of samples is the validation set, and a stratified random division strategy is adopted to ensure that the training / validation set covers different frequency fluctuation modes.

[0068] To achieve accurate ultra-short-term prediction of power grid frequency, a deep network model with long-term memory capability is established as follows. First, the basic structure of the network topology is designed, the variable sequence dimension of the input layer, the gating mechanism of the hidden layer, and the data type of the output layer are determined. The gating mechanism formula of the Long Short-Term Memory (LSTM) is as follows:

[0069]

[0070] wherein, is the input feature at time t, i.e. ; is the hidden state at time t; is the cell state at time t; is the input item weight matrix, is the hidden state item weight matrix; is the bias term, is the dimension of the hidden layer, is the dimension of the input feature, is the candidate cell state vector, represents the Hadamard product, is the Sigmoid function:

[0071]

[0072] is the hyperbolic tangent function:

[0073]

[0074] The output layer adopts a certain dropout probability to randomly shield part of the neurons to prevent overfitting, and the output layer mapping formula is:

[0075]

[0076] wherein, is the output weight matrix; is the output bias; is the output layer cell state.

[0077] The network parameters are set and optimized to improve the prediction accuracy and adaptability of the frequency sequence. The optimizer parameter update law is as follows:​​​

[0078]

[0079] where, is the network parameter of the th iteration; is the learning rate; denotes the bias-corrected first moment estimate, is the first moment of parameter gradient, used to estimate the gradient mean, accelerate convergence and reduce oscillation; denotes the bias-corrected second moment estimate, , denote the first and second moment estimate decay rate, respectively, is the second moment of parameter gradient, used to estimate the gradient variance, adaptively adjust the learning rate.

[0080] The training process loss function is defined as follows:

[0081]

[0082] where, is the sample size of each training step; denotes the predicted output of the model for the input frequency sequence , and is the actual output for the input frequency sequence .

[0083] The grid frequency prediction sequence output in this step will be used as the reference input for power feedforward compensation, providing the data basis for active control of the subsequent control loop. Among them, is the grid frequency prediction sequence, is the grid frequency prediction value at time , and is the grid frequency prediction value at time , is the prediction time domain, is the sampling time, i.e. the sample point at the sampling time.

[0084] Based on the frequency prediction output sequence, the predicted frequency deviation is defined as follows:

[0085]

[0086] where, is the predicted frequency deviation, is the predicted frequency, is the actual frequency.

[0087] ​The power response feedforward compensation of hydroelectric generating units is calculated by combining the predicted frequency deviation and the dynamic characteristics of hydroelectric generating units. The power response of the regulating system of hydroelectric generating units is influenced by the water hammer effect of the water diversion system, the mechanical and hydraulic action, and the flow torque inertia, etc. The power response and the frequency disturbance process have nonlinear phase lag characteristics. The dynamic equation reflecting the power output characteristics of the hydroelectric generating units is used to bring the predicted frequency deviation into the equation. The specific formula is as follows:

[0088]

[0089] wherein, is the control coefficient of the governor; is the frequency measurement delay time; is the hydraulic response time constant of the governor; is the flow inertia time constant of the water diversion system; is the Laplace operator, is the frequency deviation, is the power response of the generating unit.

[0090] The frequency deviation sequence in the prediction time domain is designed as follows:

[0091]

[0092] wherein, is the time-varying gain coefficient, is the control coefficient of the governor, is the nonlinear gain coefficient, is the activation function, is the prediction step, is the predicted frequency deviation at the moment, is the frequency threshold, when the predicted deviation exceeds the threshold, the coefficient is nonlinearly increased to enhance the compensation strength for large disturbances.

[0093] The generated feedforward compensation is introduced into the time decay factor, the compensation weight at different time points in the prediction time domain is dynamically adjusted by the exponential weighting, the influence of the uncertainty of long-term prediction on the control stability is reduced, and the short-term predicted deviation with high confidence is preferentially responded:

[0094]

[0095] wherein, is the decay time constant, is the time decay factor, is the flow inertia time constant of the water diversion system.

[0096] The dynamic feedforward compensation is represented as:​

[0097]

[0098] wherein, is the feedforward compensation amount, is the current prediction time.

[0099] The feedforward compensation amount generated in the previous step is embedded into the traditional proportional-integral-derivative (PID) control architecture to form a prediction-feedback collaborative mechanism to solve the response speed limitation of single closed-loop regulation. The unit power regulation composite instruction after superimposing the feedforward compensation amount can be expressed as:

[0100] wherein,

[0101] is the unit power regulation composite instruction, is the original unit power regulation instruction, is the control deviation, is the integral coefficient of the hydro-turbine governor controller, is the differential coefficient of the hydro-turbine governor controller. The feedforward term

[0102] provides a leading regulation effect to compensate for the inherent mechanical delay of the hydro-turbine regulation system. The

[0103] is input into the guide vane servo actuator of the hydroelectric unit, and together with the field frequency feedback amount, forms a secondary closed loop to dynamically correct the output and realize fast tracking of the unit power to the prediction deviation. Based on the hydro-turbine characteristic curve, the nonlinear relationship between the guide vane opening and the composite power instruction is fitted, and the inverse hyperbolic tangent function is introduced to describe the static characteristic:

[0104]

[0105] wherein,

[0106] is the slope of the guide vane opening-power regulation curve of the hydro-turbine, is the power normalization coefficient, is the opening offset coefficient, , , , all are fitting coefficients of the hydro-turbine characteristic curve; is the target guide vane opening; represents the inverse hyperbolic tangent function used to fit the nonlinear saturation characteristic.

[0107] The guide vane servo system of the unit adopts PWM signals issued by the local execution unit for driving, and the duty cycle-opening degree conversion formula is as follows:

[0108]

[0109] wherein, is the PWM duty cycle, is the minimum opening degree of the guide vane of the unit, is the maximum opening degree of the guide vane of the unit, is the minimum value of the effective range of the duty cycle, is the maximum value of the effective range of the duty cycle.

[0110] After the above mapping and conversion, the pulse width modulation (PWM) signal output by the local execution unit is used to drive the guide vane servo motor, so as to realize the closed-loop execution of the power command to the mechanical action.

[0111] The unit actuator mechanism is constrained as follows:

[0112]

[0113] To prevent sudden changes in the draft tube pressure.

[0114] wherein, is the guide vane action state under the unit power regulation command, corresponding to the guide vane action state.

[0115] Further, in step S4, an execution-feedback closed-loop dynamic control framework is constructed to realize the coordinated adaptive regulation of the power grid frequency and the hydroelectric unit power. First, a power grid frequency window sliding mechanism is established: in each control period, the first segment of historical data is deleted, and the latest measured data value is inserted to generate an updated input sequence:

[0116]

[0117] wherein, is the updated input sequence.

[0118] A dynamic truncation strategy is adopted, when a sudden change in system frequency is detected, the historical sequence is emptied, and the window is reconstructed with the current point as the center, to avoid the interference of old data on prediction. Based on the updated , the future =10 step prediction sequence can be represented as: .

[0119] The prediction is corrected in real time in combination with the power grid frequency dynamic equation:

[0120]

[0121] wherein, is the system inertia constant, is the load damping coefficient, is the gain coefficient of the turbine output power and the working water head and guide vane opening, is the unit guide vane opening, used to correct the power regulation dynamic in real time according to the unit state, is the unit working water head, unit: meter, generally regarded as a constant in the dynamic equation, is the frequency deviation.

[0122] Embodiment 2, with reference to Figures 2-6 , an active control method for a hydroelectric generating unit power based on frequency prediction is provided. In order to verify the beneficial effects of the present application, scientific demonstration is carried out through experiments.

[0123] Step 1, through constructing an LSTM time series prediction model under the sliding window mechanism, the accurate extraction and prediction of the dynamic characteristics of the power grid frequency are realized. The specific data preparation is as follows: in the actual system, the phasor measurement unit (PMU) is used to collect the power grid frequency data in real time, and the sampling period is set to ≤100 ms to ensure the capture of high-frequency dynamic characteristics. For noise interference, a sliding median filtering algorithm is used, with a window width of 6 sampling points, effectively eliminating transient impulse noise. Missing data is completed within a limited time window by linear interpolation method to avoid the influence of long-term data fault on model training.

[0124] Typical power grid frequency disturbance time series data is selected as an example to verify the effectiveness of the prediction method. Z-score method is used for data standardization, and global mean and standard deviation are used to normalize the given frequency data to ensure consistent data distribution for the input model. In the sliding window mechanism, the input window length is set to 20 sampling points, and the output step is 5 steps (predicting the frequency change in the future 500 ms), which fully balances the capture ability of short-term fluctuation characteristics and long-term trend.

[0125] The LSTM prediction model is designed as a single-input single-output structure, and the hidden layer contains 500 units. A large number of experiments have verified that this scale achieves the optimal balance between prediction accuracy and computational efficiency. In the training process, the Adam optimizer is used, and the initial learning rate is set to 0.1, combined with the early stopping strategy to prevent overfitting. Selecting typical power fluctuation data from a new energy station, a total of 800 groups of samples, the first 600 groups of samples are selected as the training set, and the remaining 200 groups are reserved as the validation set to ensure the model generalization ability. The LSTM prediction results are compared with the analysis data as shown in Figure 2 . As can be seen from the figure, the overall prediction results of the LSTM are highly consistent with the trend of the actual analysis data.

[0126] Step 2, dynamic adjustment coefficient design and power feedforward compensation mapping. The specific data preparation is as follows: based on the predicted frequency deviation, combined with the water hammer effect and mechanical inertia characteristics of the hydroelectric generating set, a dynamic adjustment coefficient design and feedforward compensation mechanism is designed. Taking a power station as an example, the water hammer time constant of the water conduit is = 2.2s, the governor time constant is = 0.3, based on this characteristic, the dynamic adjustment coefficient is designed as a time-varying gain, and the basic regulation gain is set to the original load frequency regulation gain of the generating set. The power amount to be compensated per Hz deviation when the generating set is at full load.

[0127] After detecting the grid frequency fluctuation, the adaptive adjustment is made according to the cumulative amplitude of the predicted deviation, the sensitivity coefficient is set to 0.3, and the maximum frequency deviation threshold is 0.02 Hz. When the predicted deviation exceeds the threshold, nonlinear mapping is made to enhance the compensation strength for large disturbances. At the same time, a time decay factor is introduced in the feedforward compensation amount calculation, and the decay time constant = 1.2s, the recent prediction value weight is higher, and the long-term prediction weight is exponentially attenuated, effectively suppressing the interference of prediction uncertainty on the control command.

[0128] Step 3, composite power instruction generation and closed-loop control. The specific data preparation is as follows: based on the traditional PID controller, the feedforward compensation amount is generated by weighted superposition to generate the composite power instruction. After determining the original PID regulation coefficient of the generating set, the guide vane opening control adopts PWM modulation technology, and the carrier frequency is dynamically adjusted according to the frequency deviation. The guide vane opening PWM under the comparison of the conventional PID control law and the active power control method based on frequency prediction is shown in Figure 3 and Figure 4 , wherein Figure 3 is the power instruction signal for 0.1 load step disturbance under the original control structure, Figure 4 is the power instruction signal using the active control method based on grid prediction proposed in the present application. As can be seen from the figure, the active power control method increases the action time of the power given instruction signal, thereby realizing the rapid tracking of the generating set power given value.

[0129] Step 4, a dynamic control framework of execution-feedback closed loop is constructed to realize the coordinated adaptive adjustment of power grid frequency and hydroelectric unit power. The specific data preparation is as follows: in each 100ms control period, the latest measured power grid frequency is injected into the LSTM input sequence to replace the oldest data, and the window length is maintained at 60 points. The current time prediction deviation is corrected by exponential decay to the future 10-step prediction value. Further combining the power grid frequency dynamic equation and the power response characteristics of the hydro turbine regulating system, the dynamic compensation of the hydroelectric unit power regulating amount based on the actual frequency deviation is realized. The comparison of the output power response and the opening degree of the hydroelectric unit adopting the power response active control method is shown in FIGS. 8 and 9. Figure 5 and Figure 6 As shown in the figures, the active power control method can adjust the guide vane opening more quickly after the frequency disturbance occurs, improve the power response speed, and shorten the transition process duration. Compared with the original control structure, the guide vane opening changes more quickly under active control, and the power output decreases more smoothly, effectively improving the dynamic performance and system stability of frequency regulation. In particular in Figure 6 , the active control scheme shortens the response time by about 11.2 seconds, significantly improving the rapidity and accuracy of the regulation system.

[0130] Embodiment 4 is an embodiment of the present application, which provides a frequency prediction based hydroelectric unit power active control system, comprising a power grid frequency acquisition and preprocessing module, a multi-scale time series prediction module, a dynamic compensation amount generation module, and a composite control execution module.

[0131] The power grid frequency acquisition and preprocessing module is used to acquire the basic operation data of the hydroelectric unit regulating system in real time, including the historical frequency sequence of the power grid, the unit output state, and the power grid load fluctuation information, and to preprocess the basic operation data.

[0132] The multi-scale time series prediction module is used to extract the time series features of the power grid frequency data, and a neural network is used to generate the power grid frequency prediction sequence in the future time window.

[0133] The dynamic compensation amount generation module is used to convert the power grid frequency prediction sequence into a dynamic power feedforward compensation amount, and to perform feedforward compensation amount optimization calculation combined with the preset power-frequency mapping relationship and the unit regulating constraint condition, and to superimpose the feedforward compensation amount to the output end of the feedback controller to form a composite power instruction input value hydroelectric unit.

[0134] The composite control execution module is used to update the neural network input sequence at each control period to dynamically correct the future prediction results, and to realize the control amount update based on the actual frequency deviation.

[0135] The embodiment further provides an electronic device suitable for the case of the frequency prediction based active power control method of a hydroelectric generating set, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the frequency prediction based active power control method of a hydroelectric generating set.

[0136] The embodiment further provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to realize the frequency prediction based active power control method of a hydroelectric generating set.

[0137] The storage medium proposed in the embodiment and the frequency prediction based active power control method of a hydroelectric generating set proposed in the above embodiment belong to the same inventive concept, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary universal hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk or an optical disk, etc., including a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for active power control of hydropower units based on frequency prediction, characterized in that: include, Real-time acquisition of basic operating data of hydropower unit regulation system, including historical frequency sequence of power grid, unit output status and power grid load fluctuation information, and preprocessing of basic operating data; Extract the temporal features of power grid frequency data and use a neural network to generate a power grid frequency prediction sequence within a future time window; The power grid frequency prediction sequence is converted into a dynamic power feedforward compensation quantity. The feedforward compensation quantity is optimized by combining the preset power-frequency mapping relationship and the unit regulation constraints. The feedforward compensation quantity is then superimposed on the output of the feedback controller to form a composite power command input value for the hydropower unit. A rolling time-domain prediction strategy is adopted to update the neural network input sequence in each control cycle, dynamically correct future prediction results, and realize the update of control quantity based on actual frequency deviation. The step of optimizing the feedforward compensation amount by combining the preset power-frequency mapping relationship and unit regulation constraints includes, The generated feedforward compensation is embedded into the PID control architecture, and the composite power regulation command of the unit after superimposing the feedforward compensation is expressed as follows: ； in, This is a composite command for unit power regulation. This is the unit's original power regulation command. To control deviation, The integral coefficient of the turbine governor controller. The differential coefficient of the turbine governor controller; Will The input hydro-generator guide vane servo actuator, together with the field frequency feedback, forms a secondary closed loop to dynamically correct the output. Based on the turbine characteristic curves, the nonlinear relationship between the guide vane opening and the combined power regulation command of the unit is fitted. An inverse hyperbolic tangent function is introduced to describe the static characteristics, expressed as: ； in, The slope of the turbine guide vane opening-power regulation curve. This is the power normalization coefficient. This is the opening offset coefficient; The target guide vane opening; Represents the inverse hyperbolic tangent function; The turbine guide vane servo system is driven by pulse width modulation signals issued by the local execution unit. The duty cycle-opening conversion formula is expressed as follows: ； in, For PWM duty cycle, This is the minimum opening of the unit's guide vanes. This represents the maximum opening of the unit's guide vanes. This represents the minimum effective range of the duty cycle. This represents the maximum effective range of the duty cycle. After mapping and conversion are completed, a pulse width modulation signal is output to drive the guide vane servo motor, realizing closed-loop execution from power command to mechanical action.

2. The active power control method for hydropower units based on frequency prediction as described in claim 1, characterized in that: The real-time acquisition of basic operating data of the hydropower unit regulation system includes, Collect historical frequency data sequences of the power grid to form the original frequency sequence; For outliers in the original frequency sequence, the moving median filtering method is used to eliminate instantaneous disturbances, and the filtered power grid frequency values ​​are normalized. The normalized power grid frequency sequence follows a standard normal distribution.

3. The active power control method for hydropower units based on frequency prediction as described in claim 2, characterized in that: The extraction of time-series features from power grid frequency data, followed by the generation of a future time window power grid frequency prediction sequence using a neural network, includes... Define the sliding window length and the prediction time domain based on the historical observation step size; The sequence is segmented using a sliding window, and training sample pairs are generated as follows: ； in, The first of the historical frequency window data Group input samples; The first step of the future frequency prediction target Group output labels; This represents the number of frequency sample indices. ; This represents the total number of samples at the sampling time. The length of the sliding window; It is the set of real numbers; for A matrix of order real numbers; For the normalized first The group's power grid frequency, For the normalized first The group's power grid frequency, For the normalized first The group's power grid frequency, For the normalized first The group's power grid frequency.

4. The active power control method for hydropower units based on frequency prediction as described in claim 3, characterized in that: The process of converting the grid frequency prediction sequence into a dynamic power feedforward compensation quantity includes, Divide the power grid frequency samples into datasets, and let the preceding... The group of samples is the training set, and then... The group of samples serves as the validation set; Frequency prediction is performed using a neural network, and the output power grid frequency prediction sequence is represented as follows: ； in, For power grid frequency prediction sequences, for Predicted power grid frequency at time 10:

00. for Predicted power grid frequency at time 10:

00. To predict the time domain, Sampling time; Based on the frequency prediction output sequence, the prediction frequency deviation is defined as follows: ； in, To predict frequency deviation, To predict frequency, This is the actual frequency; For prediction time domain frequency deviation sequence within The design time-varying gain coefficient is expressed as follows: ； in, For time-varying gain coefficients, This is the speed governor control coefficient. For nonlinear gain coefficients, For activation function, To predict the step size, For the first Predicted frequency deviation at time, This is the frequency threshold.

5. The active power control method for hydropower units based on frequency prediction as described in claim 4, characterized in that: The process of converting the grid frequency prediction sequence into a dynamic power feedforward compensation quantity also includes, The feedforward compensation is generated by introducing a time decay factor. The compensation weights at different time points in the prediction time domain are dynamically adjusted through exponential weighting, as shown below. ； in, The decay time constant, The time decay factor, The inertial time constant of the water flow in the water diversion system; The dynamic feedforward compensation amount is then expressed as: ； in, This is the feedforward compensation amount. This is the current predicted time.

6. The active power control method for hydropower units based on frequency prediction as described in claim 5, characterized in that: The aforementioned rolling time-domain prediction strategy updates the neural network input sequence in each control cycle, dynamically correcting future prediction results to achieve control quantity updates based on actual frequency deviations. The real-time correction prediction based on the power grid frequency dynamic equation is expressed as follows: ； in, Let be the system's inertial constant. This is the load damping coefficient. This is the gain coefficient of the turbine output power in relation to the operating head and guide vane opening. For the turbine guide vane opening, For the unit's working head, This represents the frequency deviation.

7. A frequency-predictive-based active power control system for hydropower units, employing the frequency-predictive-based active power control method for hydropower units as described in any one of claims 1 to 6, characterized in that, include: The module includes a power grid frequency acquisition and preprocessing module, a multi-scale time series prediction module, a dynamic compensation quantity generation module, and a composite control execution module. The power grid frequency acquisition and preprocessing module is used to acquire basic operating data of the hydropower unit regulation system in real time, including historical frequency sequences of the power grid, unit output status and power grid load fluctuation information, and to preprocess the basic operating data. The multi-scale time series prediction module is used to extract the time series features of power grid frequency data and use a neural network to generate a power grid frequency prediction sequence within a future time window. The dynamic compensation quantity generation module is used to convert the power grid frequency prediction sequence into a dynamic power feedforward compensation quantity, and to perform feedforward compensation quantity optimization calculation in combination with the preset power-frequency mapping relationship and unit regulation constraints. The feedforward compensation quantity is then superimposed on the output of the feedback controller to form a composite power command input value hydropower unit. The composite control execution module is used to employ a rolling time-domain prediction strategy to update the neural network input sequence in each control cycle, dynamically correct future prediction results, and realize control quantity updates based on actual frequency deviations.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the active power control method for hydropower units based on frequency prediction as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the active power control method for hydropower units based on frequency prediction as described in any one of claims 1 to 6.

Images