A variable speed pumped storage unit starting method and system

By constructing the mechanical torque-speed starting model of the pumped storage unit, and using the neural network algorithm to generate reference data, the problem of starting instability when historical starting data is missing is solved, and the smooth starting of the unit under any load is achieved.

CN118487514BActive Publication Date: 2024-11-12DONGFANG ELECTRIC CHENGDU INTELLIGENT TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410942293.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-12
Estimated Expiration
2044-07-15

Smart Images

  • Figure CN118487514B_ABST
    Figure CN118487514B_ABST
Patent Text Reader

Abstract

The present invention discloses a variable speed pumped storage unit starting method and system, which relates to the technical field of pumped storage unit starting, and comprises the following steps: obtaining the original starting data of the variable speed pumped storage unit, and preprocessing the original starting data; configuring a Bp neural network, and inputting the preprocessed original starting data into the configured Bp neural network for training, generating a unit starting model between the speed, mechanical torque and starting load; using the unit starting model, obtaining the starting reference target speed of the variable speed pumped storage unit under the target starting load; starting the variable speed pumped storage unit, when the real-time speed of the variable speed pumped storage unit deviates from the starting reference target speed, the starting speed of the unit is corrected in real time through a PID control loop. The present invention can obtain the starting reference data of the unit under any load condition, and realize the smooth starting of the unit.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of pumped storage unit starting, and more particularly to a variable speed pumped storage unit starting method and system. Background Art

[0002] Pumped-storage power stations are flexible and reliable in operation, with rapid operating conditions and little impact on the environment. They are currently recognized as the most mature, economical and largest energy storage method. With the large-scale grid connection of new energy sources such as wind power generation and photovoltaic power generation, pumped-storage power stations have become one of the indispensable ways to suppress fluctuations in wind and solar power and maintain the stability of grid voltage and frequency.

[0003] The pumped storage unit can be operated in a water pumping mode to pump water from the lower reservoir to the upper reservoir for temporary storage, or it can be operated in a power generation mode to use the water energy stored in the upper reservoir to generate electricity and supply the power grid, thus realizing energy storage and release. According to the changes in the working conditions, the unit needs to be started frequently. The commonly used starting method at present is the variable frequency starting method, which uses a static variable frequency starting device (SFC) to drag the unit from zero speed to rated speed, and the unit is connected to the grid synchronously and then the SFC is stopped to realize the smooth start of the unit.

[0004] In the prior art, when a pumped storage unit is started with load, in order to ensure the stability of the unit's operating state at each start, the unit's starting speed is controlled near the target speed by referring to the unit's historical starting data under the load, including data such as speed and mechanical torque. When the unit's historical starting data under a certain load is missing, the unit starts without corresponding reference data, which will affect the unit's starting state. Summary of the invention

[0005] In order to overcome the defects existing in the above-mentioned prior art, the present invention discloses a variable speed pumped storage unit starting method and system to solve the problem in the prior art that it is necessary to refer to historical starting data. When the historical starting data of the unit under a certain load is missing, the unit starts without corresponding reference data, which affects the starting state of the unit. The present invention is based on the existing historical data of the unit's smooth start under multiple load conditions, and forms a unit starting model suitable for the mechanical torque-speed when the unit starts under any load. The unit starting model is used to obtain reference data so that when the unit is started frequently in the future, the starting speed can be corrected in time according to the reference data.

[0006] In order to achieve the above objectives, the technical solution adopted by the present invention is:

[0007] A variable speed pumped storage unit starting method comprises the following steps:

[0008] 1. Raw data acquisition and preprocessing

[0009] S100, obtaining original starting data of the variable speed pumped storage unit, and preprocessing the original starting data;

[0010] Preferably, in step S100, the original starting data includes the speed n of the variable speed pumped storage unit during the process from zero speed to rated speed. i , Mechanical torque M i And the corresponding starting load P i At the same time, the original starting data is the historical data of the unit's existing smooth starting under multiple load conditions.

[0011] Preferably, in step S100, the pretreatment includes the following steps:

[0012] S110, eliminating abnormal data in the original startup data;

[0013] S120, forming an original data matrix according to the data category of the original starting data ;in, is the original data matrix, is the rotation speed, is the mechanical torque, is the starting load, and is the input variable, is the output variable;

[0014] S130, the original data matrix Perform normalization to obtain the normalized data matrix ;in, is the normalized data matrix, is the normalized speed, is the normalized mechanical torque, is the normalized starting load.

[0015] In the above steps, the original startup data of the pumped storage unit is obtained, and data preprocessing is performed to remove abnormal data. In order to facilitate subsequent calculations and improve computational efficiency, normalization processing is required to map the data to between [0 1].

[0016] Preferably, in step S130, the normalization processing includes: using a conversion function to perform a linear transformation on the original data matrix so that the original data matrix is ​​mapped to between [0 1] to obtain a normalized data matrix.

[0017] Preferably, in step S130, the conversion function is:

[0018] ;

[0019] in, is the normalized value. is the value to be converted. It is the maximum value of the data in the same category of the original starting data. It is the minimum value of the same category in the original starting data.

[0020] 2. Neural Network Configuration and Training

[0021] S200, configuring a Bp neural network, and inputting the preprocessed original starting data into the configured Bp neural network for training, to generate a unit starting model between the speed, mechanical torque and starting load;

[0022] 2.1. Configure Bp neural network

[0023] Preferably, in step S200, configuring the Bp neural network includes configuring the number of infrastructure layers, activation function, error function, number of training times and mean square error threshold of the Bp neural network.

[0024] In the above steps, Bp represents the error back propagation algorithm, which determines the number of layers, activation function, error function, number of training times, mean square error threshold, etc. of the Bp neural network algorithm infrastructure for subsequent use.

[0025] Preferably, in step S200, the basic architecture of the Bp neural network includes an input layer, an output layer and a hidden layer, and the number of nodes in the input layer is n. I , the number of nodes in the output layer is n 0 , the number of neurons in the hidden layer is M L .

[0026] In the present invention, according to the original starting data, the basic architecture of the Bp neural network is confirmed, including one input layer, one output layer, and L hidden layers; the number of nodes in the input layer is n I , the number of nodes in the output layer is n 0 , the number of neurons in each hidden layer is M L .

[0027] 2.2 Neural Network Training

[0028] Preferably, in step S200, the unit startup model is:

[0029] ;

[0030] in, is the rotation speed, is the mechanical torque, is the starting load, It is the unit start function.

[0031] Preferably, step S200 includes the following steps:

[0032] S210, inputting the pre-processed original starting data into the Bp neural network, setting the input layer, output layer and activation function of the Bp neural network, and initializing the Bp neural network;

[0033] Preferably, in step S210, the input layer, output layer and activation function of the Bp neural network are:

[0034] ;

[0035] ;

[0036] ;

[0037] in, is the input layer, is the output layer, is the normalized mechanical torque, is the normalized starting load, is the normalized speed, For transposition, is the activation function, is the independent variable of the function.

[0038] In the above steps, define the input layer X and output layer Y, represented by vectors. Define the activation function in the neural network algorithm, and use the Sigmoid function.

[0039] S220, setting the hidden layer of the Bp neural network, and combining the input layer, output layer and activation function, as well as the weight matrix and bias matrix, to obtain the output of each neuron in the hidden layer and the output layer;

[0040] Preferably, step S220 includes: defining the weight matrix and bias matrix from the input layer to the hidden layer as W and B respectively, the output matrix of the hidden layer as V, the weight matrix and bias matrix from the hidden layer to the output layer as H and C respectively, wherein the initial values ​​of the W, B, H, and C matrices are random values, then:

[0041] ;

[0042] in, represents the input layer, Indicates the calculation results of the unit starting model, Represents the activation function.

[0043] S230, using the error function, calculating the deviation between the neural network output value and the actual output value;

[0044] Preferably, step S230 includes: defining an error function E to measure the deviation between the unit startup model calculation result and the theoretical data result, and the error function is:

[0045] ;

[0046] in, is the error function, is the actual output value, Output value for the neural network.

[0047] In the above steps, the error function E is defined to measure the deviation between the model calculation results and the theoretical data results in the form of mean square error, and then determine whether the algorithm stopping condition is met.

[0048] S240, determining whether the deviation is less than or equal to a threshold value, if so, proceeding to step S250, if not, proceeding to step S270;

[0049] The above threshold is the mean square error threshold.

[0050] S250, determining whether the Bp neural network training has reached the maximum number of training times, if so, proceeding to step S260, if not, proceeding to step S270;

[0051] S260, generating and outputting a unit starting model;

[0052] ;

[0053] S270, calculate the gradient of the error function with respect to each parameter in the weight matrix and the bias matrix based on the gradient descent method, then use the learning rate and the gradient descent method to update the weight matrix and the bias matrix, and then return to step S220.

[0054] 3. Obtaining the starting reference target speed

[0055] S300, using the unit starting model to obtain a starting reference target speed of the variable speed pumped storage unit under a target starting load;

[0056] Preferably, in step S300, for the unit starting model, when the actual target starting load is constant, as the mechanical torque changes, the starting reference target speed of the variable speed pumped storage unit is:

[0057] ;

[0058] in, For starting reference target speed, is the mechanical torque, is the actual target starting load, It is the unit start function.

[0059] In the above steps, for the unit starting model, when the actual target starting load is constant, as the mechanical torque changes, the starting reference target speed of the variable speed pumped storage unit is the mechanical torque-speed unit starting model, that is, the mathematical relationship between the speed and the mechanical torque. This mathematical relationship can be used to obtain the reference data of the speed and mechanical torque when the unit is started, and the reference data can be used to correct the starting speed in time.

[0060] 4. Correct the unit speed according to the starting reference target speed

[0061] S400, starting the variable speed pumped storage unit. When the real-time speed of the variable speed pumped storage unit deviates from the starting reference target speed, the starting speed of the unit is corrected in real time through the PID control loop.

[0062] In the above steps, according to the starting reference target speed When the real-time speed of the unit deviates from the reference speed, the PID control loop is used to achieve precise control of the unit speed.

[0063] Preferably, the real-time correction of the unit starting speed through the PID control loop includes:

[0064] The speed governor adjusts the guide vane opening by obtaining the deviation between the pump-turbine speed and the speed command, adjusts the water inlet flow, controls the output of the pump-turbine mechanical torque, and adjusts the output power of the generator; if the rotor speed needs to be controlled to be stable, that is, the speed acceleration is equal to 0, then the guide vane opening is adjusted to adjust the water flow so that the mechanical torque is equal to the electromagnetic torque.

[0065] Based on the above variable speed pumped storage unit starting method, the present invention also provides a variable speed pumped storage unit starting system, including a data acquisition and preprocessing module, a system configuration module, a data calculation and output module, and a control module;

[0066] The data acquisition and preprocessing module is used to read the original starting data of the variable speed pumped storage unit and preprocess the original starting data, including removing abnormal data in the original starting data and normalizing the original starting data;

[0067] The system configuration module is used to configure the Bp neural network, including configuring the number of infrastructure layers, activation function, error function, number of training times and mean square error threshold of the Bp neural network;

[0068] The data calculation and output module is used to perform BP neural network training on the pre-processed original starting data, generate a unit starting model, and then determine the starting reference target speed of the variable speed pumped storage unit;

[0069] The control module is used to correct the starting speed of the unit in real time according to the starting reference target speed.

[0070] Beneficial effects of the present invention:

[0071] Based on the historical start-up data of the pumped storage unit, the present invention uses a neural network algorithm to train and obtain the unit start-up model (mathematical expression) between the corresponding parameters, and generalizes the relationship between the physical quantities in the start-up process of the full-power pumped storage unit. Based on the above unit start-up model, the unit start-up reference data under any load condition can be obtained to achieve a smooth start of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic diagram of the starting method of the variable speed pumped storage unit of the present invention;

[0073] Figure 2 This is a schematic diagram of neural network training in step S2 of the present invention;

[0074] Figure 3 It is the PID control block diagram of the present invention;

[0075] Figure 4 It is a schematic diagram of the variable speed pumped storage unit starting system of the present invention. DETAILED DESCRIPTION

[0076] The concept, specific structure and technical effects of the present invention will be clearly and completely described below in conjunction with the embodiments and drawings to fully understand the purpose, characteristics and effects of the present invention.

[0077] Example 1

[0078] A variable speed pumped storage unit starting method, such as Figure 1 As shown, the following steps are included:

[0079] S100, obtaining original starting data of the variable speed pumped storage unit, and preprocessing the original starting data;

[0080] S200, configuring a Bp neural network, and inputting the preprocessed original starting data into the configured Bp neural network for training, to generate a unit starting model between the speed, mechanical torque and starting load;

[0081] S300, using the unit starting model to obtain a starting reference target speed of the variable speed pumped storage unit under a target starting load;

[0082] S400, starting the variable speed pumped storage unit. When the real-time speed of the variable speed pumped storage unit deviates from the starting reference target speed, the starting speed of the unit is corrected in real time through the PID control loop.

[0083] Example 2

[0084] Based on Example 1, this embodiment further elaborates on step S100. In step S100, the original starting data of the variable speed pumped storage unit is collected and summarized, and the original starting data of the pumped storage unit is preprocessed. The original starting data includes the speed n of the variable speed pumped storage unit from zero speed to rated speed. i , Mechanical torque M i And the corresponding starting load P i .

[0085] Specifically, the preprocessing includes the following steps:

[0086] S110, eliminating abnormal data in the original startup data;

[0087] S120, forming an original data matrix according to the data category of the original starting data ;in, is the original data matrix, is the rotation speed, is the mechanical torque, is the starting load, and is the input variable, is the output variable;

[0088] S130, the original data matrix Perform normalization to obtain the normalized data matrix ;in, is the normalized data matrix, is the normalized speed, is the normalized mechanical torque, is the normalized starting load.

[0089] In the above steps, the original startup data of the pumped storage unit is obtained, and data preprocessing is performed to remove abnormal data; in order to facilitate subsequent calculations, normalization processing is required. Normalization processing includes: using a conversion function to perform a linear transformation on the original data matrix so that the original data matrix is ​​mapped to [0 1].

[0090] The conversion function is:

[0091] ;

[0092] in, is the normalized value. is the value to be converted. It is the maximum value of the data in the same category of the original starting data. It is the minimum value of the same category in the original starting data.

[0093] In this embodiment, according to the above conversion function, the original data matrix Normalized to Taking the speed n as an example, , the normalized mechanical torque can be obtained in the same way , normalized starting load , and then get the normalized data matrix .

[0094] In the above steps, outliers in the original starting data are removed. In order to facilitate the subsequent Bp neural network calculation, the original starting data is linearly transformed so that the data result value is mapped to the range [0 1]. Normalize to a normalized data matrix .

[0095] Example 3

[0096] This embodiment further elaborates step S200 on the basis of embodiment 2. In step S200, it is necessary to clarify the number of layers, activation function, error function, number of training times, mean square error threshold, etc. of the basic architecture of the Bp neural network algorithm for use in subsequent steps.

[0097] In this embodiment, the Bp neural network infrastructure includes one input layer, one output layer, and L hidden layers. In this embodiment, the hidden layer is 1, the number of input layer nodes is 2, the number of output layer nodes is 1, and the number of neurons in each hidden layer is 3.

[0098] Example 4

[0099] This embodiment further describes step S200 based on embodiment 3. Figure 2 As shown, step S200 includes the following steps:

[0100] 1. Input original data, set network related parameters and initialize the network

[0101] S210, inputting the pre-processed original starting data into the Bp neural network, setting the input layer, output layer and activation function of the Bp neural network, and initializing the Bp neural network;

[0102] In step S210, the input layer, output layer and activation function of the Bp neural network are:

[0103] ;

[0104] ;

[0105] ;

[0106] in, is the input layer, is the output layer, is the normalized mechanical torque, is the normalized starting load, is the normalized speed, For transposition, is the activation function, is the independent variable of the function.

[0107] In the above steps, define the input layer X and output layer Y, represented by vectors. Define the activation function in the neural network algorithm, and use the Sigmoid function.

[0108] 2. Find the output of each neuron in the hidden layer and output layer

[0109] S220, setting the hidden layer of the Bp neural network, and combining the input layer, output layer and activation function, as well as the weight matrix and bias matrix, to obtain the output of each neuron in the hidden layer and the output layer;

[0110] Step S220 includes: defining the weight matrix and bias matrix from the input layer to the hidden layer as W and B respectively, the output matrix of the hidden layer as V, the weight matrix and bias matrix from the hidden layer to the output layer as H and C respectively, wherein the initial values ​​of the W, B, H, and C matrices are random values, then:

[0111] ;

[0112] in, represents the input layer, It represents the calculation result of the unit starting model. Represents the activation function.

[0113] 3. Find the deviation between expected output and actual output

[0114] S230, using the error function, calculating the deviation between the neural network output value and the actual output value;

[0115] Step S230 includes: defining an error function E to measure the deviation between the unit startup model calculation result and the theoretical data result, and the error function is:

[0116] ;

[0117] in, is the error function, is the actual output value, Output value for the neural network.

[0118] In the above steps, the error function E is defined to measure the deviation between the model calculation results and the theoretical data results in the form of mean square error, and then determine whether the algorithm stopping condition is met.

[0119] 4. Threshold determination

[0120] S240, determining whether the deviation is less than or equal to a threshold value, if so, proceeding to step S250, if not, proceeding to step S270;

[0121] The above threshold is the mean square error threshold.

[0122] 5. Determine the number of training times

[0123] S250, determining whether the Bp neural network training has reached the maximum number of training times, if so, proceeding to step S260, if not, proceeding to step S270;

[0124] 6. Output data results

[0125] S260, generating and outputting a unit starting model;

[0126] ;

[0127] in, is the rotation speed, is the mechanical torque, is the starting load, It is the unit start function.

[0128] 7. Find the error gradient and update the weight and bias matrices

[0129] S270, based on the gradient descent method, calculate the gradient of the error function for each parameter in the weight matrix and the bias matrix, then use the learning rate and gradient descent method to update the weight matrices W, H and the bias matrices B, C, and then return to step S220.

[0130] In step S240 and step S250, first determine whether the value output by the error function is less than the set threshold. If so, further determine whether the maximum number of training times has been reached. If so, output the data result. If any process does not meet the set conditions, then as shown in step S270, calculate the gradient of the error function with respect to each parameter in the weight matrix and the bias matrix based on the gradient descent method, and update the weight and bias matrices.

[0131] After updating the weight and bias matrices, return to step S220 and repeat until the set conditions are met, and then output the data results.

[0132] Example 5

[0133] This embodiment further elaborates on step S300 based on embodiment 4. In step S300, for the unit starting model, when the actual target starting load is constant, as the mechanical torque changes, the starting reference target speed of the variable speed pumped storage unit is:

[0134] ;

[0135] in, For starting reference target speed, is the mechanical torque, is the actual target starting load, It is the unit start function.

[0136] In this embodiment, for the unit starting model, when the actual target starting load is constant, as the mechanical torque changes, the starting reference target speed of the variable speed pumped storage unit is the mechanical torque-speed unit starting model, that is, the mathematical relationship between the speed and the mechanical torque. This mathematical relationship can be used to obtain reference data of the speed and mechanical torque when the unit is started, and the reference data can be used to timely correct the starting speed.

[0137] Example 6

[0138] This embodiment further describes step S400 based on embodiment 5. In step S400, according to the starting reference target speed, When the real-time speed of the unit deviates from the reference speed, the PID control loop is used to achieve precise control of the unit speed.

[0139] like Figure 3 As shown, It is the mechanical torque-speed unit starting model; It is the reference target speed for the unit to start; It is the actual speed when the unit starts; PID is the controller.

[0140] like Figure 3 As shown, The mathematical model of the guide vane opening control device in the speed governor is represented. It represents the time constant of the servomotor in the hydraulic follow-up system of the speed governor. represents the independent variable of the complex frequency-domain function.

[0141] like Figure 3 As shown, Represents the mathematical model of the turbine and water diversion system. Here, the rigid water hammer model is used. It represents the time constant reflecting the water hammer effect of the water diversion system. represents the independent variable of the complex frequency-domain function.

[0142] like Figure 3As shown, Represents the mathematical model of the unit motion equation, Indicates the rotor moment of inertia of the unit ( ).

[0143] like Figure 3 As shown, Indicates the guide vane opening of the pump-turbine; It represents the mechanical torque of the pump turbine; represents the electromagnetic torque of the pump turbine; The subscript T represents a turbine.

[0144] like Figure 3 As shown, the control method for starting the variable speed pumped storage unit is as follows: the speed governor adjusts the size of the guide vane opening by obtaining the deviation between the speed of the pump turbine and the speed command, adjusts the water inlet flow, and then controls the output of the mechanical torque of the pump turbine, thereby further adjusting the output power of the generator. If the rotor speed is to be controlled to be stable, that is, the speed acceleration is equal to 0, it is necessary to adjust the guide vane opening to adjust the water flow so that the mechanical torque is equal to the electromagnetic torque.

[0145] Example 7

[0146] A variable speed pumped storage unit starting system, such as Figure 4 As shown, it includes a data acquisition and preprocessing module, a system configuration module, a data calculation and output module, and a control module;

[0147] The data acquisition and preprocessing module is used to read the original starting data of the variable speed pumped storage unit and preprocess the original starting data, including removing abnormal data in the original starting data and normalizing the original starting data;

[0148] The system configuration module is used to configure the Bp neural network, including configuring the number of infrastructure layers, activation function, error function, number of training times and mean square error threshold of the Bp neural network;

[0149] The data calculation and output module is used to perform BP neural network training on the pre-processed original starting data, generate a unit starting model, and then determine the starting reference target speed of the variable speed pumped storage unit;

[0150] The control module is used to correct the starting speed of the unit in real time according to the starting reference target speed.

[0151] The above is a specific description of the implementation mode of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention, and these equivalents or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A variable speed pumped storage unit starting method, characterized in that: The following steps are involved: S100, obtaining original starting data of the variable speed pumped storage unit, and preprocessing the original starting data; S200, configuring a Bp neural network, and inputting the preprocessed original starting data into the configured Bp neural network for training, to generate a unit starting model between the speed, mechanical torque and starting load; S300, using the unit starting model to obtain a starting reference target speed of the variable speed pumped storage unit under a target starting load; S400, starting the variable speed pumped storage unit, when the real-time speed of the variable speed pumped storage unit deviates from the starting reference target speed, correcting the starting speed of the unit in real time through the PID control loop; In step S100, the pre-processing includes the following steps: S110, eliminating abnormal data in the original startup data; S120, forming an original data matrix according to the data category of the original starting data ;in, is the original data matrix, is the rotation speed, is the mechanical torque, is the starting load, and is the input variable, is the output variable; S130, the original data matrix Perform normalization to obtain the normalized data matrix ;in, is the normalized data matrix, is the normalized speed, is the normalized mechanical torque, is the normalized starting load; In step S200, the unit startup model is: ; in, is the rotation speed, is the mechanical torque, is the starting load, is the unit start function; In step S300, for the unit starting model, when the actual target starting load is constant, as the mechanical torque changes, the starting reference target speed of the variable speed pumped storage unit is: ; in, For starting reference target speed, is the mechanical torque, is the actual target starting load, is the unit start function; Step S200 includes the following steps: S210, inputting the pre-processed original starting data into the Bp neural network, setting the input layer, output layer and activation function of the Bp neural network, and initializing the Bp neural network; S220, setting the hidden layer of the Bp neural network, and combining the input layer, output layer and activation function, as well as the weight matrix and bias matrix, to obtain the output of each neuron in the hidden layer and the output layer; S230, using the error function, calculating the deviation between the neural network output value and the actual output value; S240, determining whether the deviation is less than or equal to a threshold value, if so, proceeding to step S250, if not, proceeding to step S270; S250, determining whether the Bp neural network training has reached the maximum number of training times, if so, proceeding to step S260, if not, proceeding to step S270; S260, generating and outputting a unit starting model; S270, calculate the gradient of the error function with respect to each parameter in the weight matrix and the bias matrix based on the gradient descent method, then use the learning rate and the gradient descent method to update the weight matrix and the bias matrix, and then return to step S220.

2. A variable speed pumped storage unit starting method according to claim 1, characterized in that: In step S130, the normalization process includes: using a conversion function to perform a linear transformation on the original data matrix so that the original data matrix is ​​mapped to between [0 1] to obtain a normalized data matrix; The conversion function is: ; in, is the normalized value. is the value to be converted. It is the maximum value of the data in the same category of the original starting data. It is the minimum value of the same category in the original starting data.

3. A variable speed pumped storage unit starting method as claimed in claim 1, characterized in that: In step S200, the configuration of the Bp neural network includes configuring the number of infrastructure layers, activation function, error function, number of training times and mean square error threshold of the Bp neural network; Among them, the basic architecture of the Bp neural network includes an input layer, an output layer, and a hidden layer. The number of nodes in the input layer is n I , the number of output layer nodes is n0, and the number of hidden layer neurons is M L .

4. A variable speed pumped storage unit starting method as claimed in claim 1, characterized in that: In step S210, the input layer, output layer and activation function of the Bp neural network are: ; ; ; in, is the input layer, is the output layer, is the normalized mechanical torque, is the normalized starting load, is the normalized speed, For transposition, is the activation function, is the independent variable of the function.

5. A variable speed pumped storage unit starting method as claimed in claim 4, characterized in that: Step S220 includes: defining the weight matrix and bias matrix from the input layer to the hidden layer as W and B respectively, the output matrix of the hidden layer as V, the weight matrix and bias matrix from the hidden layer to the output layer as H and C respectively, wherein the initial values ​​of the W, B, H, and C matrices are random values, then: ; in, represents the input layer, represents the calculation results of the group start model, Represents the activation function.

6. A variable speed pumped storage unit starting method as claimed in claim 5, characterized in that: Step S230 includes: defining an error function E to measure the deviation between the group start model calculation result and the theoretical data result, and the error function is: ; in, is the error function, is the actual output value, Output value for the neural network.

7. A system for starting a variable speed pumped storage unit according to any one of claims 1 to 6, characterized in that: It includes data acquisition and preprocessing module, system configuration module, data calculation and output module, and control module; The data acquisition and preprocessing module is used to read the original starting data of the variable speed pumped storage unit and preprocess the original starting data, including removing abnormal data in the original starting data and normalizing the original starting data; The system configuration module is used to configure the Bp neural network, including configuring the number of infrastructure layers, activation function, error function, number of training times and mean square error threshold of the Bp neural network; The data calculation and output module is used to perform BP neural network training on the pre-processed original starting data, generate a unit starting model, and then determine the starting reference target speed of the variable speed pumped storage unit; The control module is used to correct the starting speed of the unit in real time according to the starting reference target speed.