Paddle-guide vane joint control method and system of axial-flow movable paddle hydraulic turbine set and medium of paddle-guide vane joint control method and system

Abstract

The invention provides an axial-flow propeller turbine set blade-guide vane joint control method and system and a medium thereof, and the method comprises the following steps: forming a training set based on the real-time operation data of an axial-flow propeller turbine set under a normal working condition, so as to update the model parameters of a trained feedforward neural network; generating a guide paddle opening degree relation table through a feedforward neural network; the guide blade opening degree relation table is used for representing the mapping relation between the water head height and the unit power of the axial-flow movable blade hydraulic turbine unit and the guide blade opening degree and the blade opening degree; when the axial-flow propeller turbine set is subjected to frequency disturbance, the guide vane opening degree and the blade opening degree target value are obtained based on the current water head height of the axial-flow propeller turbine set and the relation table of the target set power and the guide vane opening degree; and the blades and the guide blades of the axial-flow movable-paddle hydraulic turbine set are controlled through the guide blade opening and the blade opening target value. The method can adapt to the actual condition of each unit and is quick in response.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydropower, and in particular relates to a blade-guide vane coordinated control method, system and medium thereof for an axial-flow propeller turbine unit. Background Art

[0002] Axial-flow propeller turbine is a common hydroelectric power generation equipment, widely used in hydropower stations with low head and large flow. Unlike other types of turbines, axial-flow propeller turbines achieve power regulation through two main components: guide vanes and blades. The guide vanes and blades of existing axial-flow propeller turbines usually work in a coordinated relationship. When the unit is power regulated, the guide vanes start first, and the blades rotate with the guide vanes, thereby ensuring that the water flow enters the runner in the best state and efficiently realizes the process of converting the energy of the water flow into electrical energy. The coordinated relationship of the general unit is formulated by the manufacturer based on the results of the model test, and the governor manufacturer automatically coordinates through the program. However, due to the deviation of the model test and the errors in installation and component processing, the relationship obtained is often different from the actual machine operation, making it difficult for the guide blades of the unit to operate at the optimal opening. Therefore, when the existing units are put into use, the coordinated relationship is often optimized according to the actual situation of the unit. The establishment of the relationship between the guide blades of the unit is often achieved through the construction of the unit model. Nowadays, most physical and mathematical models or new algorithms are used to build the unit model. However, there are internal problems such as the volute, water pipe pressure, and friction coefficient at the turbine test site that fluctuate with the operation of the hydropower unit. In terms of external problems, it is difficult for the model to fully simulate the actual climate and geographical environment factors in various places, so it is difficult to ensure the accuracy and universal applicability of the model; that is, in a highly nonlinear and complex system such as a turbine, the performance of the model built by the existing method is often limited; and it is difficult for the unit to maintain the same working conditions over time. When the built model faces constantly changing working conditions, the construction and maintenance of the rule base becomes extremely difficult. The adjustment and optimization of the model requires a lot of professional knowledge, resulting in its lack of adaptability and flexibility in practical applications. Summary of the invention

[0003] The purpose of the present invention is to solve the deficiencies of the above-mentioned background technology and to provide a blade-guide vane coordinated control method, system and medium for an axial-flow propeller turbine unit, which can adapt to the actual situation of each unit and respond quickly.

[0004] The technical solution adopted by the present invention is: a blade-guide vane coordinated control method for an axial flow propeller turbine unit, comprising the following steps:

[0005] A training set is formed based on real-time operating data of the axial-flow propeller turbine unit under normal operating conditions to update the model parameters of the trained feedforward neural network; a single sample in the training set includes the head height and unit power as model inputs, and the guide vane opening and the blade opening as training labels;

[0006] Generate a guide blade opening relationship table through a feedforward neural network; the guide blade opening relationship table is used to characterize the mapping relationship between the head height and unit power of the axial flow propeller turbine unit and the guide blade opening and the blade opening;

[0007] When the axial-flow propeller turbine unit is subjected to frequency disturbance, the guide vane opening and the blade opening target values ​​are obtained based on the current head height of the axial-flow propeller turbine unit and the relationship table between the target unit power and the guide vane opening;

[0008] The blades and guide vanes of the axial flow propeller turbine unit are controlled by the guide vane opening and the blade opening target values.

[0009] The beneficial effects of the present invention are as follows: when the grid frequency is disturbed, the present invention can directly call the opening table to quickly calculate the optimal guide vane and blade opening, timely adjust the unit output, stabilize the grid frequency, and improve the unit's response speed to grid frequency changes. The present invention can continuously obtain the corresponding unit guide and blade opening values ​​under multiple working conditions as the unit is running, and complete the guide-blade relationship data of the unit, thereby training the actual machine operation data of the axial flow propeller turbine unit, and through the real-time updated feedforward neural network model (Feedforward Neural Network, FNN), it can accurately predict the optimal opening of the guide vane and blade, and realize accurate control of the turbine unit. The present invention coordinates the opening of the blade and guide vane, optimizes the operating efficiency of the turbine unit, reduces energy loss, and enables the unit to have better frequency regulation performance and enhances adaptability to frequency disturbances. The present invention reduces the volatility of the unit operation and improves the stability and reliability of the power grid through precise control strategies.

[0010] Furthermore, the present invention eliminates noise and outliers in the data by denoising historical and real-time data, thereby improving the quality of training data. Denoising makes the feedforward neural network more stable during training and updating, avoiding the interference of noise on the model weights. The model parameters are updated using the denoised real-time data, so that the model can adapt to changes in working conditions in a timely manner and maintain high-precision prediction.

[0011] Furthermore, the present invention adopts a deep neural network structure with 10 hidden layers, which enhances the model's ability to fit complex nonlinear relationships and adapt to the complex operating characteristics of the turbine unit. The present invention uses the train function for training, which ensures the effectiveness and convergence of the model training and improves the prediction accuracy. The deep network structure adopted by the present invention helps the model learn key features from a large amount of data and improves the prediction ability of unknown data.

[0012] Furthermore, in the process of generating the guide blade opening relationship table, the present invention ensures the accuracy of the data in the relationship table by selecting a working range with high training accuracy; fills the gaps between the data through interpolation processing, so that the relationship table can cover a wider range of working conditions and provide a more comprehensive control basis. The present invention writes the relationship table data into an Excel table, which facilitates the viewing, management and updating of the data and improves the maintainability of the system.

[0013] Furthermore, the present invention updates the models and relationship tables under different powers respectively, optimizes each power segment, and improves the control accuracy under each working condition; directly calls the corresponding relationship table according to the current target power, reduces the calculation time, and speeds up the control response speed; can adapt to changes in different power conditions, and improves the adaptability and flexibility of the system.

[0014] Furthermore, the present invention strictly evaluates the model accuracy by comparing the target value and the actual value, ensuring that the model meets the expected accuracy requirements at each power; retraining the model that does not meet the accuracy requirements reflects the dynamic optimization of the model and maintains the high performance of the model; the relationship table is only generated after the model accuracy meets the requirements, ensuring the reliability of the control data and reducing the risk of system operation.

[0015] Furthermore, the present invention dynamically calculates the target unit power according to the frequency disturbance situation, so that the unit can respond to the changes in the grid frequency in a timely manner; by adjusting the unit power output, it participates in the frequency regulation of the grid, thereby enhancing the stability and safety of the grid; and comprehensively considers the head height and rated frequency to ensure that the adjusted target power is within the optimal operating range of the unit to avoid overload or inefficient operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the implementation and operation of the method proposed by the present invention;

[0017] Figure 2 is a flow chart of data preprocessing in a specific embodiment;

[0018] Figure 3 is a schematic diagram of feedforward neural network training unit data of a specific embodiment;

[0019] Figure 4It is a schematic diagram showing the relationship between the guide blades under different powers of multiple heads generated in a specific embodiment;

[0020] Figure 5 It is a comparison chart of the predicted results output by the FNN model of a specific embodiment and the actual values. DETAILED DESCRIPTION

[0021] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation on the present invention.

[0022] like Figure 1 As shown, the present invention provides a blade-guide vane coordinated control method for an axial flow propeller turbine unit, comprising the following steps:

[0023] A training set is formed based on real-time operating data of the axial-flow propeller turbine unit under normal operating conditions to update the model parameters of the trained feedforward neural network; a single sample in the training set includes the head height and unit power as model inputs, and the guide vane opening and the blade opening as training labels;

[0024] A guide blade opening relationship table is generated through a feedforward neural network; the guide blade opening relationship table is used to characterize the mapping relationship between the head height and unit power of the axial flow propeller turbine unit and the guide vane opening and the blade opening, that is, (Y1, Y2) = f(h, P), where h is the head height, P is the unit power, and Y1 and Y2 are the guide and blade openings.

[0025] When the axial-flow propeller turbine unit is subjected to frequency disturbance, the guide vane opening and the blade opening target values ​​are obtained based on the current head height of the axial-flow propeller turbine unit and the relationship table between the target unit power and the guide vane opening;

[0026] The blades and guide vanes of the axial flow propeller turbine unit are controlled by the guide vane opening and the blade opening target values.

[0027] Specifically, the power, water head, guide vane and blade data of the unit under different operating conditions are first recorded and preprocessed to avoid errors and blank data affecting the training effect of the feedforward neural network model; the processed training data are Import a feedforward neural network, where h i and P i As the input of the model, As the output of the feedforward neural network, i represents the sample number. Then, the trained feedforward neural network model outputs the guide and blade opening table corresponding to different powers under multiple water heads, and introduces an autonomous data collection mechanism to continuously supplement new data.

[0028] Specifically, the historical operating data of the axial-flow propeller turbine unit under normal operating conditions collected are denoised and used for training the feedforward neural network; the real-time operating data of the axial-flow propeller turbine unit under normal operating conditions collected are denoised and used for updating the feedforward neural network.

[0029] Preferably, in order to avoid the influence of error data and blank data on model training, the recorded data table should be preprocessed before using the actual machine data. For the processing of error data, consider using the Raida filter criterion to denoise it. The Raida filter criterion is a probability rule commonly used in statistics, which describes the distribution of data under normal distribution. For a random variable that obeys a normal distribution, about 99.73% of the data is within the range of the mean value μ±3σ. The data in the unit data table is recorded as in The mean μ of h and P h and μ P Calculate using formula (1);

[0030]

[0031] The standard deviation σ of h and P h and σ P Calculated by formula (2).

[0032]

[0033] for If there is any hydraulic head h i Satisfy h i <μ h -3σ h or i >μ h +3σ h And there exists any power P i Satisfy P i <μ P -3σ P or P i >μ P +3σ P . Then mark h i or P i For the abnormal value h marked in the original data table i and P i As well as blank and duplicate data, the processed training data table is obtained. Figure 2 shown.

[0034] Specifically, after the training data is preprocessed, it is input into the feedforward neural network for training. Feedforward Neural Network (FNN). The basic structure of FNN includes an input layer, one or more hidden layers and an output layer, each layer consisting of several neurons. The input signal is transmitted from the input layer to the output layer layer by layer through the hidden layer, without a feedback loop in the middle, and the information only flows forward. The input layer receives external data and passes it to the next layer. Each input neuron corresponds to a feature value; the hidden layer is located between the input layer and the output layer, and there can be one or more. The neurons in each hidden layer receive input from the previous layer and pass the results to the next layer after processing through the activation function. The existence of hidden layers enables the network to capture complex nonlinear relationships. The neurons in the output layer receive input from the last hidden layer and generate the final output results. The number of neurons in the output layer depends on the specific task requirements, such as the number of categories in classification tasks or the predicted values ​​in regression tasks.

[0035] FNN builds a complex model by stacking layers. The calculation of each layer is essentially a linear transformation, so the selection of activation functions between hidden layers of FNN is particularly important. Without activation functions, the entire network is equivalent to a stack of linear transformations. The introduction of activation functions enables neural networks to represent complex nonlinear relationships, thereby enhancing the network's expressive power. The relationship between a network layer and its next layer in a neural network is shown in formula (3).

[0036] y=α(ωx+b) (3)

[0037] Where y and x are the output and input values ​​of the neural network layer, α is the activation function, ω is the weight matrix of each layer, and b is a bias vector added to each layer. The selection of α, ω, and b can determine the training effect of the model.

[0038] Due to the complex working conditions of the turbine, the Adam adaptive learning rate optimization algorithm is introduced to optimize the parameters ω and b in formula (3). The Adam algorithm dynamically adjusts the learning rate so that the update step size of each weight and bias adapts to the current gradient size, thereby optimizing the convergence speed and stability of the training process. This is particularly important in FNN training, especially when dealing with complex data or large-scale data, which can significantly improve the training efficiency and prediction accuracy of the model. First, initialize the parameters, set the first-order moment and the second-order moment to 0, and the time step to 0, that is, m t =0 (initial first-order moment); v t = 0 (initial second-order moment); t = 0 (time step). For each time step t, each parameter (including the weight matrix and ω bias vector b) are updated as follows:

[0039]

[0040] where g t Represents the gradient calculated at the current time t, that is, the loss function About parameters The partial derivative of θ Indicates the calculation of the partial derivative of the loss function J(θ) with respect to the parameter θ (i.e. the gradient of the current parameter). t Represents the moving average of the first-order moment of the gradient, used to update the first-order moment estimate, representing the average of past gradients. t Represents the moving average of the second-order moment of the gradient, used to update the second-order moment estimate, representing the average of the square of the past gradient. β1 and β2 represent the decay rate. is the first-order moment of bias correction, The second-order moment for bias correction is used to reduce the bias of the first-order moment and the second-order moment estimation in the initial stage. is the update parameter, where

[0041] α1 is the learning rate, which controls the step size of each parameter update and is a key hyperparameter that affects the training speed and model convergence. It is usually set to α = 0.001, but the specific value needs to be adjusted according to the complexity of the problem and the data set. ε is a very small value used to prevent division by zero in the parameter update formula; ε is usually set to 10 -8 .

[0042] Activation function Consider introducing the hyperbolic tangent activation function (tansig) to process the data of axial flow propeller turbine. The expression of tansig function is shown in (5).

[0043]

[0044] The output value range of the tansig function is (-1,1). It is a bounded nonlinear function that can map any input to a limited interval. It can provide strong nonlinear transformation capabilities, so that the feedforward neural network can better adapt to the complex nonlinear relationship between different operating conditions of the unit to improve the accuracy of prediction and control.

[0045] In order to deal with the complex nonlinear relationship of the turbine, the feedforward neural network used in this embodiment includes an input layer, 10 hidden layers and an output layer to improve the accuracy of the trained model; finally, the train function is called to train the model output by FNN, and the training logic of the train function is: 1. Initialize the network: define a neural network object with initial weights and biases; 2. Error calculation: use the current weights and biases to calculate the network output, compare it with the target output, and get the error; 3. Weight update: calculate the gradient according to the error, and adjust the weights and biases to reduce the error. This usually uses some kind of optimization algorithm (such as gradient descent); 4. Repeat training: repeat the error calculation and weight update until the stopping condition is met (such as the maximum number of iterations, target error, performance target, etc.). Finally, the corresponding Y1 and Y2 opening relationship model of the trained axial flow propeller turbine under different working conditions is obtained. The above process is as follows Figure 3 shown.

[0046] Specifically, the process of generating a guide blade opening relationship table through a feedforward neural network includes: setting a working range based on the properties of an axial flow propeller turbine unit; the working range includes a unit power range and a head height range; interpolating the feedforward neural network according to the set working range and the set guide blade opening relationship table accuracy, obtaining a one-to-one correspondence between the head height and the unit power and the guide vane opening and the blade opening and writing them into an Excel table to generate a guide blade opening relationship table.

[0047] Preferably, the model under the working condition where the training accuracy reaches the set standard is interpolated to generate a Y1 and Y2 relationship table.

[0048] In this embodiment, considering the influence of the dead zone of the unit, the water head h precision of the guide blade opening relationship table is 0.1, and the power precision is 2MW. The working range is set according to the actual working condition of the turbine. For example, the working condition range is set as: the power range is [P~P+2m](MW), and the water head range is [h~h+0.1n](m), where m and n are integers. The guide blade opening relationship table is generated based on the above working range.

[0049] For example, if the operating head range of a 100MW turbine unit is between 10m and 30m, the unit power P can be selected from the interval of 2MW between [10,100]MW, and the head height h can be selected from the interval of 0.1m between [10,30].

[0050] Fill in the corresponding values ​​in the newly created Excel table, that is, the unit power P is selected as [P, P+2, P+4,…, P+2m]; h is selected as [h, h+0.1, h+0.2,…, h+0.1n], and then read the values ​​in the table through the FNN model and output the corresponding Y1 and Y2. The output Y1 and Y2 values ​​are the predicted guide and propeller blade openings corresponding to different powers under multiple heads, and then form a guide blade opening relationship table. The schematic diagram of the guide blade opening relationship table is as follows Figure 4 shown.

[0051] Specifically, based on the training set, the corresponding feedforward neural network and guide blade opening relationship table are updated for different unit powers respectively; according to the current target unit power of the axial flow propeller turbine unit, the corresponding guide blade opening relationship is called to generate the guide vane opening and blade opening target values.

[0052] The process of updating the feedforward neural network and the guide blade opening relationship table includes: for any unit power, comparing the target value of the guide vane opening and the blade opening at the power with the actual value, and evaluating the prediction accuracy of the feedforward neural network at the power; if the prediction accuracy of the feedforward neural network at the power does not meet the standard, retraining the feedforward neural network at the power; when the prediction accuracy reaches the set standard, generating the corresponding guide blade opening relationship table based on the feedforward neural network at the power.

[0053] Preferably, the autonomous data collection mechanism provided by the present invention will automatically collect the operating data of the turbine unit in real time for training and updating the FNN model to generate the Y1 and Y2 opening tables under different working conditions (i.e., different unit powers: P, P+2, P+4, ..., P+2m). Since it is necessary to consider the actual working conditions of the unit, the mechanism can autonomously determine the model prediction accuracy under different working conditions. The data under the working conditions that do not meet the standards in the model are supplemented, and the FNN is re-driven to train the model under the working conditions.

[0054] The trained single FNN model in this embodiment can be used to predict the guide vane opening and blade opening target values ​​under multiple working conditions. When it is necessary to retrain a specific working condition, you can choose to "fine-tune" the model, that is, based on the FNN model, only further train the data of the specific working condition; or use selective training (selective retraining), that is, under the existing model framework, only train the specific data of a certain working condition, and keep the data of other working conditions unchanged. The core of this method is to use "frozen" layers (freeze the weights of existing layers) and only train specific layers to adapt to the characteristics of the working condition; or if the working conditions are very different or fine-tuning cannot meet the requirements, you can also choose to completely "retrain" the working condition based on the original model. This means that part of the data in the multi-condition training will be ignored, and only the data under a single condition will be fully trained.

[0055] The evaluation criteria for prediction effect are R 2 and mean square error (MSE), R 2 The closer the value is to 1, the smaller the error between the average predicted value (target value) and the actual value.

[0056] In this embodiment, when the feedforward neural network is trained, the historical operating data of the axial flow propeller turbine unit under normal working conditions is used to form a training set and a verification set, and the model training accuracy is verified by the verification set. After the feedforward neural network training is completed and put into use, a prediction verification set is formed by the real-time operating data of the axial flow propeller turbine unit under normal working conditions, and the model prediction accuracy is verified by the prediction verification set, and a new training set is formed based on the newly collected real-time operating data to optimize the model accuracy and the guide blade opening relationship table. Among them, the standards of training accuracy and prediction accuracy are the same.

[0057] It should be emphasized that when the axial flow propeller turbine unit is working normally, the feedforward neural network continuously collects real-time operation data to calculate and verify the guide blade opening target value to optimize its own model and the guide blade opening relationship table; however, the traditional method is still used to control the guide blade opening, and the guide blade opening relationship table is not used to control the guide blade opening. Only when the turbine unit is subjected to frequency disturbance, the guide blade opening relationship table is called to control the guide blade opening, and the unit guide and blade openings are adjusted separately to speed up the response process.

[0058] Specifically, the guide vane opening relationship table is filtered based on the Mahalanobis distance and used to generate the guide vane opening and blade opening target values. The Mahalanobis distance is the covariance distance of the data. Unlike the Euclidean distance, it considers the connection between various characteristics and effectively measures the similarity of two unknown sample sets. The Mahalanobis distance is scale-independent, that is, it is independent of the measurement scale and can eliminate the interference of the correlation between variables. It is widely used in classification and cluster analysis. When constructing the relationship table of the actual unit operation data, each set of coordinates (h, P, Y1, Y2) h, P is used as the independent variable, and Y1, Y2 is used as the dependent variable for calculation and solution. Suppose the error data set of n operating points is U:

[0059]

[0060] in In the formula, Y′1, Y′2, are predicted values, and Y1, Y2 are actual values.

[0061] The Mahalanobis distance (M) from each error data point to the mean vector μ D )for:

[0062]

[0063] In formula (6), Σ is the covariance matrix. The filtering algorithm measures the Mahalanobis distance between the fitting node and the sampled data, retaining M D Less than the threshold H D The data is used as effective measurement data to construct the turbine operation characteristic surface. When the random error of the measurement data causes the data point to deviate from the average characteristic in the support domain, the point is determined to be an error point and is eliminated.

[0064] Specifically, when the axial-flow propeller-type water turbine unit is subjected to frequency disturbance, the sum of the power increment and the active output of the axial-flow propeller-type water turbine unit at the current operating head height and rated frequency is taken as the target unit power.

[0065] Preferably, the relationship between the frequency disturbance and the unit power is explained by the mathematical expression of formula (8).

[0066]

[0067] This formula converts the difference Δf between the unit frequency and the grid reference frequency into the power difference ΔP. f is the frequency dead zone of the speed regulation system (Hz), e p The power increment ΔP is the permanent difference coefficient. The turbine active output P at the current operating head rated frequency is 0之 And the corresponding target power P can be obtained ref , combined with the current h, the corresponding Y1 and Y2 are obtained by calling and querying the corresponding guide blade opening relationship table.

[0068] This specific embodiment implements and verifies the above-mentioned feedforward neural network-based axial flow propeller turbine blade-guide vane coordinated control method:

[0069] FNN is used to establish the guide-blade relationship table corresponding to the power under multiple heads of the axial-flow propeller turbine unit. A total of 1,048,576 data from the actual operation of the turbine unit for three days were recorded in an Excel table to form a training set and a validation set. They were preprocessed in MATLAB, and after eliminating errors, blanks, and duplicate data, a total of 289,826 data remained. The number of network layers and activation functions of FNN were set in MATLAB, and the preprocessed training set and validation set were used for training and validation. The predicted data of 50 groups of validation samples randomly selected from the trained model were compared with their actual values ​​in the data table, as shown in the figure. Figure 5 As shown. The mean square error (MSE) of the guide vane opening is 0.13457, R 2 is 0.99277; the mean square error (MSE) of the blade opening is 0.097557, R 2 The effect diagram can prove that the (Y1,Y2)=f(h,P) model of the axial flow propeller turbine unit trained by FNN has a good effect.

[0070] Example 2

[0071] The present invention provides a blade-guide vane coordinated control system for an axial-flow propeller turbine set, which is used to implement the blade-guide vane coordinated control method for an axial-flow propeller turbine set described in the above technical solution.

[0072] Example 3

[0073] The present invention provides various types of computer-readable storage media, on which a computer program is stored. When the computer program is executed by a processor, the blade-guide vane coordinated control method of an axial-flow impeller turbine group described in the above technical solution is implemented.

[0074] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0076] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0078] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

[0079] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A method for controlling blades and guide vanes of an axial-flow propeller turbine, characterized in that: The following steps are involved: A training set is formed based on real-time operating data of the axial-flow propeller turbine unit under normal operating conditions to update the model parameters of the trained feedforward neural network; a single sample in the training set includes the head height and unit power as model inputs, and the guide vane opening and the blade opening as training labels; Generate a guide blade opening relationship table through a feedforward neural network; the guide blade opening relationship table is used to characterize the mapping relationship between the head height and unit power of the axial flow propeller turbine unit and the guide blade opening and the blade opening; When the axial-flow propeller turbine unit is subjected to frequency disturbance, the guide vane opening and the blade opening target values ​​are obtained based on the current head height of the axial-flow propeller turbine unit and the relationship table between the target unit power and the guide vane opening; The blades and guide vanes of the axial flow propeller turbine unit are controlled by the guide vane opening and the blade opening target values.

2. A method according to claim 1, characterized in that: The historical operating data of the axial-flow propeller turbine unit under normal working conditions collected are denoised and used for training the feedforward neural network; the real-time operating data of the axial-flow propeller turbine unit under normal working conditions collected are denoised and used for updating the feedforward neural network.

3. A method according to claim 1, characterized in that: The feedforward neural network includes an input layer, 10 hidden layers and an output layer, and is trained using a train function.

4. A method according to claim 1, characterized in that: The guide vane opening relationship table is filtered based on the Mahalanobis distance and used to generate the guide vane opening and blade opening target values.

5. A method according to claim 1, characterized in that: The process of generating a guide blade opening relationship table through a feedforward neural network includes: interpolating the feedforward neural network according to the set unit power range, head height range and guide blade opening relationship table accuracy, obtaining the one-to-one correspondence between head height and unit power and guide vane opening and blade opening and writing them into an Excel table to generate a guide blade opening relationship table.

6. A method according to claim 5, characterized in that: Based on the training set, the corresponding feedforward neural network and guide blade opening relationship table are updated for different unit powers respectively; according to the current target unit power of the axial flow propeller turbine unit, the corresponding guide blade opening relationship is called to generate the guide vane opening and blade opening target values.

7. A method according to claim 6, characterized in that: The process of updating the feedforward neural network and the guide blade opening relationship table includes: for any unit power, comparing the target value of the guide vane opening and the blade opening at the power with the actual value, and evaluating the prediction accuracy of the feedforward neural network at the power; if the prediction accuracy of the feedforward neural network at the power does not meet the standard, retraining the feedforward neural network at the power; when the prediction accuracy reaches the set standard, generating the corresponding guide blade opening relationship table based on the feedforward neural network at the power.

8. A method according to claim 1, characterized in that: When the axial-flow propeller turbine unit is subject to frequency disturbance, the sum of the power increment and the active output of the axial-flow propeller turbine unit at the current operating head height and rated frequency is taken as the target unit power.

9. A blade-guide vane coordinated control system for an axial flow propeller turbine unit, characterized in that: The system is used to implement the blade-guide vane coordinated control method of an axial-flow propeller turbine group according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the blade-guide vane coordinated control method of an axial-flow propeller turbine unit according to any one of claims 1 to 8 is implemented.

Images