Bayesian deep learning-based fan power prediction method and system

Abstract

The invention discloses a Bayesian deep learning-based fan power prediction method and system. The method comprises the steps of screening mixed sample features related to fan power prediction under the condition of considering icing according to a physical model; taking the mixed sample features as input, and constructing an adaptive wind power prediction model based on a Bayesian multilayer neural network; generating a fan power output expert experience sample database according to the physical model, taking the fan power output expert experience sample database as a self-adaptive wind power prediction model constructed by priori pre-training, and collecting historical operation data of a wind power plant as posteriori; based on a random variational derivation method, regularly retraining the adaptive wind power prediction model to obtain a fan power prediction model; and inputting the fan operation real-time data into the fan power prediction model, and outputting to obtain a fan power prediction result. And expert knowledge and historical operation data are comprehensively utilized to greatly improve the accuracy of wind power prediction under the fan icing condition.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power prediction, and in particular relates to a wind turbine power prediction method and system based on Bayesian deep learning. Background Art

[0002] When wind turbines are operating in cold or high-altitude areas, ice is easily formed on the blade surface, which leads to a decrease in aerodynamic performance, which in turn causes a significant decrease in power generation, an increase in structural load, and even safety accidents. Wind turbine icing not only affects its own operating performance, but may also have an adverse effect on the stability of the power grid. Therefore, accurately predicting the power output of wind turbines under icing conditions is of great significance to ensure the safe and stable operation of wind farms and grid dispatching.

[0003] At present, wind turbine power prediction methods mainly include physical modeling, statistical methods and intelligent algorithms based on machine learning. The physical modeling method is based on the aerodynamic principle of the wind turbine and considers factors such as wind speed and wind direction to calculate power. However, under icing conditions, the performance of the blades changes and the model is difficult to describe accurately. The statistical method uses historical data for modeling, but when there are few icing samples, the reliability of the model is reduced. The intelligent algorithm based on machine learning can improve the prediction accuracy to a certain extent by learning a large amount of historical data, but when the icing data is scarce, the model is prone to overfitting and cannot quantify the uncertainty of the prediction results.

[0004] Traditional machine learning methods often ignore the uncertainty of model prediction results when dealing with wind turbine power prediction problems under icing conditions, which may lead to scheduling decision errors. In addition, wind turbine icing mainly occurs in winter, and there are fewer data samples. Especially for newly built wind farms, directly applying traditional deep learning models may have the risk of overfitting, insufficient model generalization ability, and cannot provide credibility assessment of prediction results. Summary of the invention

[0005] The present invention provides a method and system for predicting wind turbine power based on Bayesian deep learning, which is used to solve the technical problem that blade icing causes changes in aerodynamic performance, and traditional physical modeling methods are difficult to accurately describe the characteristics of blades after icing, resulting in a decrease in power prediction accuracy.

[0006] In a first aspect, the present invention provides a wind turbine power prediction method based on Bayesian deep learning, comprising: Construct a physical model of how icing affects wind turbine output; Determine sample characteristics of the measurement system according to the real-time measurement data, and then select mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model; Taking the mixed sample features as input, constructing an adaptive wind power prediction model based on a Bayesian multi-layer neural network; Generate a wind turbine power output expert experience sample database according to the physical model, use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model; Real-time operation data of the wind turbine is acquired, and the real-time operation data of the wind turbine is input into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

[0007] In a second aspect, the present invention provides a wind turbine power prediction system based on Bayesian deep learning, comprising: A first building module is configured to build a physical model of the impact of icing process on wind turbine output; A screening module is configured to determine sample characteristics of the measurement system according to the real-time measurement data, and then screen mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model; A second building module is configured to take the mixed sample features as input to build an adaptive wind power prediction model based on a Bayesian multi-layer neural network; A training module is configured to generate a wind turbine power output expert experience sample database according to the physical model, and use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model; The output module is configured to obtain real-time data of wind turbine operation and input the real-time data of wind turbine operation into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

[0008] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the wind turbine power prediction method based on Bayesian deep learning of any embodiment of the present invention.

[0009] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the program instructions are executed by a processor, the processor executes the steps of the wind turbine power prediction method based on Bayesian deep learning of any embodiment of the present invention.

[0010] The wind turbine power prediction method and system based on Bayesian deep learning of the present application takes the physical model of wind turbine icing, natural ice melting of blades and wind turbine output as prior knowledge, and the historical operation data of wind turbines collected by wind power plants as a posteriori, and uses a Bayesian multi-layer neural network to construct a probability mapping relationship between wind turbine operation data and wind power under icing conditions. Finally, the neural network parameters are trained by combining the Bayesian statistical inference theory and the random variational derivation method, thereby comprehensively utilizing expert knowledge and historical operation data to greatly improve the accuracy of wind power prediction under wind turbine icing conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A flow chart of a wind turbine power prediction method based on Bayesian deep learning provided by an embodiment of the present invention; Figure 2 A structural block diagram of a wind turbine power prediction system based on Bayesian deep learning provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] See also Figure 1 , which shows a flow chart of a wind turbine power prediction method based on Bayesian deep learning in the present application.

[0015] like Figure 1 As shown, the wind turbine power prediction method based on Bayesian deep learning specifically includes the following steps: Step S101, constructing a physical model of the impact of icing process on wind turbine output.

[0016] In this step, a wind turbine power output model based on wind energy conversion theory, a wind turbine blade ice accumulation model based on FENSAP software, and a wind turbine blade natural deicing model based on convection radiation effect and blade rotation ice throwing are constructed, wherein the expression of the wind turbine power output model is: , , , , , , In the formula, is the fan power output value, is the mechanical loss coefficient, is the power loss coefficient, is the rotation coefficient, is the yaw error, is the air density, is the area of ​​wind sweep, is the inflow wind speed, is the distance from the blade unit to the hub center, is the fan blade profile, For the leaf tip, is the wake effect, is the tip speed ratio, is the blade tip shape factor, is the lift-to-drag ratio, is the lift per unit length of the airfoil along the span, is the drag per unit length of the airfoil along the span, is the inflow angle of the local wind direction blade unit, represents the local inflow angle of the blade, express Number of blades , Tip Speed ​​Ratio and the tip shape factor related, is the rotor radius, is the number of leaves.

[0017] The expression of airflow in the fan blade ice accumulation model is: , In the formula, is the laminar kinematic viscosity, is the dimensionless laminar viscosity, , are all truncation functions, is the vorticity, is the Reynolds number at infinity, , , as well as are the closure coefficients, is a closed function, is the distance from the wall, is the Stefan-Boltzmann constant, is the Hamiltonian operator, is the speed change.

[0018] The Euler model is used for impact calculations, where the continuity and momentum equations associated with the droplet are expressed as:

[0019] , In the formula, is the droplet volume fraction, is the air flow rate, is the droplet relative velocity, is the droplet drag coefficient, is the droplet Reynolds number, is the droplet inertia parameter, is the air density, is the density of water, is the Froude number, is the acceleration due to gravity; According to the droplet collection efficiency Solving the freezing process, the expressions of the partial differential equations for mass conservation and energy conservation are obtained as follows: , In the formula, is the water film density, is the height, is the symbol of divergence, is the flow rate, is the reference speed, is the liquid water content, is the evaporation rate of liquid water, is the freezing rate of liquid water; , In the formula, is the specific heat capacity of the water film, is the latent heat of evaporation, is the latent heat of sublimation, is the latent heat of melting, is the specific heat capacity of ice, is the solid heat dissipation coefficient, is the reference temperature, is the reference droplet temperature in Kelvin, is the reference droplet temperature in degrees Celsius, is the water-ice interface temperature in degrees Celsius, is the heat transferred by convection, is the water-ice interface temperature in Kelvin; , In the formula, is the normal vector of the blade surface; The expression of the ice melting rate of the natural deicing model of the fan blades is: , In the formula, is the diameter of the frozen object, is the convection exchange coefficient, , are the air temperature and ice surface temperature, is the radiation constant, is the specific heat capacity of ice, is the latent heat of ice, is the density of ice; The expression for ice thickness when ice falls off is: , In the formula, is the thickness of ice when it falls off, is the viscous shear stress, is the wind turbine diameter, is the blade rotation rate.

[0020] Step S102: determining sample characteristics of the measurement system according to the real-time measurement data, and then screening mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model.

[0021] In this step, the expression of mixed sample features is: , In the formula, is the air flow rate, is the blade rotation speed, is the tip speed ratio, is the pitch angle, is the number of leaves, is the rotor radius, is the yaw error, is the air density, is the inflow wind speed, is the liquid water content, is the droplet volume fraction, is the droplet relative velocity, is the water-ice interface temperature in degrees Celsius, is the water-ice interface temperature in Kelvin, is the convection exchange coefficient, is the air surface, is the ice surface temperature, is the density of ice.

[0022] Step S103: Taking the mixed sample features as input, an adaptive wind power prediction model based on a Bayesian multi-layer neural network is constructed.

[0023] In this step, the adaptive wind power prediction model includes: A variational Bayesian gated recurrent unit, wherein the variational Bayesian gated recurrent unit is used to process time series observation history data in SCADA, wherein the time series observation history data in SCADA contains mixed sample features; Variational Bayesian neuron, the basic unit of the variational Bayesian neuron is an artificial neuron, and the expression is: , In the formula, is the deviation, is the output of the variational Bayesian neuron, is the activation function, is the weight assigned to the network connection, is the input of the variational Bayes neuron, is the input quantity.

[0024] GRU unit, the expression of the GRU unit is: , , , , , In the formula, To reset the gate, To update the gate, , are the hidden states of the GRU unit at the t-th iteration and the hidden states of the GRU unit at the t-1-th iteration, respectively. For Hadamard, is the sigmoid function, is a candidate hidden state, is the output of the GRU unit, is the input of the GRU unit at time t, is the weight of the reset gate output occupied by the previous hidden state, The weight of the reset gate output is the weight of the current input. is the weight of the update gate output occupied by the previous hidden state, is the weight of the current input in updating the gate output, is the weight of the candidate hidden state occupied by the previous hidden state and the reset gate Hadamard product, is the weight of the candidate hidden state occupied by the current input, is the hyperbolic tangent function.

[0025] Step S104, generating a wind turbine power output expert experience sample database based on the physical model, and using the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, and collecting historical operation data of the wind farm as a posteriori, and regularly retraining the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model.

[0026] In this step, the wind turbine geographic data, weather forecast data and historical operation data observed by SCADA are input into the physical model to calculate the wind turbine power, and the wind turbine power output expert experience sample database is obtained; the geographic information data includes longitude, latitude and altitude data. The weather forecast data includes: time, temperature, atmospheric pressure, relative humidity, weather type, wind speed, wind direction and air density at different heights of 10m, 50m and 100m.

[0027] The constructed adaptive wind power prediction model is pre-trained based on the database expert experience sample database, where: Inputs to the time-adaptive wind power prediction model for: , In the formula, for Historical operation data at all times, is the wind turbine geographic data, for Weather forecast data at all times; The output of the time-adaptive wind power prediction model is Power output of the fan at the moment , the expression is: , In the formula, for The wind power output of the wth wind turbine at time, for Moment Wind power output of a typhoon turbine.

[0028] It should be noted that, based on the Bayesian statistical inference theory, the model retraining, the wind turbine power prediction model based on the Bayesian multi-layer neural network can be recorded as: , In the formula, is a set of random parameters, is the probability density function of the corresponding distribution, is the variational parameter set, is a sample data set.

[0029] According to Bayesian theory, the posterior inference problem can be expressed as: , SVI introduces a series of distributions , using prior knowledge to determine the detailed type and initial parameters of the distribution , by optimizing , find the closest distribution.

[0030] Step S105, acquiring real-time operation data of the wind turbine, and inputting the real-time operation data of the wind turbine into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

[0031] In summary, the method of the present application takes the physical model of wind turbine icing, natural ice melting of blades and wind turbine output as prior knowledge, and the historical operation data of wind turbines collected by wind power plants as a posteriori, and uses a Bayesian multi-layer neural network to construct a probability mapping relationship between wind turbine operation data and wind power under icing conditions. Finally, the Bayesian statistical inference theory and the random variational derivation method are combined to train the neural network parameters, thereby comprehensively utilizing expert knowledge and historical operation data to greatly improve the accuracy of wind power prediction under wind turbine icing conditions.

[0032] See also Figure 2 , which shows a structural block diagram of a wind turbine power prediction system based on Bayesian deep learning in the present application.

[0033] like Figure 2 As shown, the wind turbine power prediction system 200 includes a first construction module 210 , a screening module 220 , a second construction module 230 , a training module 240 and an output module 250 .

[0034] Among them, the first construction module 210 is configured to construct a physical model of the impact of the icing process on the wind turbine output; the screening module 220 is configured to determine the sample characteristics of the measurement system based on real-time measurement data, and then screen the mixed sample characteristics related to the wind turbine power prediction under icing conditions according to the physical model; the second construction module 230 is configured to use the mixed sample characteristics as input to construct an adaptive wind power prediction model based on a Bayesian multi-layer neural network; the training module 240 is configured to generate a wind turbine power output expert experience sample database according to the physical model, and use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, and collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on the random variational derivation method to obtain a wind turbine power prediction model; the output module 250 is configured to obtain real-time operation data of the wind turbine, and input the real-time operation data of the wind turbine into the wind turbine power prediction model, and the wind turbine power prediction model outputs the wind turbine power prediction result.

[0035] It should be understood that Figure 2 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects are also applicable to Figure 2 The modules in it will not be described in detail here.

[0036] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor is caused to execute the wind turbine power prediction method based on Bayesian deep learning in any of the above method embodiments; As an implementation mode, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows: Construct a physical model of how icing affects wind turbine output; Determine sample characteristics of the measurement system according to the real-time measurement data, and then select mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model; Taking the mixed sample features as input, constructing an adaptive wind power prediction model based on a Bayesian multi-layer neural network; Generate a wind turbine power output expert experience sample database according to the physical model, use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model; Real-time operation data of the wind turbine is acquired, and the real-time operation data of the wind turbine is input into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

[0037] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the wind turbine power prediction system based on Bayesian deep learning, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the wind turbine power prediction system based on Bayesian deep learning via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0038] Figure 3 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 3 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 3 In the example, the connection through the bus is taken as an example. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the wind turbine power prediction method based on Bayesian deep learning in the above-mentioned method embodiment is realized. The input device 330 can receive input digital or character information, and generate key signal input related to user settings and function control of the wind turbine power prediction system based on Bayesian deep learning. The output device 340 may include display devices such as display screens.

[0039] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0040] As an implementation mode, the electronic device is applied to a wind turbine power prediction system based on Bayesian deep learning, and is used for a client, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: Construct a physical model of how icing affects wind turbine output; Determine sample characteristics of the measurement system according to the real-time measurement data, and then select mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model; Taking the mixed sample features as input, constructing an adaptive wind power prediction model based on a Bayesian multi-layer neural network; Generate a wind turbine power output expert experience sample database according to the physical model, use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model; Real-time operation data of the wind turbine is acquired, and the real-time operation data of the wind turbine is input into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

[0041] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wind turbine power prediction method based on Bayesian deep learning, characterized in that: include: Construct a physical model of how icing affects wind turbine output; Determine sample characteristics of the measurement system according to the real-time measurement data, and then select mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model; Taking the mixed sample features as input, constructing an adaptive wind power prediction model based on a Bayesian multi-layer neural network; Generate a wind turbine power output expert experience sample database according to the physical model, use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model; Real-time operation data of the wind turbine is acquired, and the real-time operation data of the wind turbine is input into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

2. The wind turbine power prediction method based on Bayesian deep learning according to claim 1 is characterized in that: The physical model of the influence of icing process on wind turbine output is constructed as follows: A wind turbine power output model based on wind energy conversion theory, a wind turbine blade ice accumulation model based on FENSAP software, and a wind turbine blade natural deicing model based on convection radiation effect and blade rotation ice throwing were constructed. The expression of the wind turbine power output model is: ， ， ， ， ， ， In the formula, is the fan power output value, is the mechanical loss coefficient, is the power loss coefficient, is the rotation coefficient, is the yaw error, is the air density, is the area of ​​wind sweep, is the inflow wind speed, is the distance from the blade unit to the hub center, is the fan blade profile, For the leaf tip, is the wake effect, is the tip speed ratio, is the blade tip shape factor, is the lift-to-drag ratio, is the lift per unit length of the airfoil along the span, is the drag per unit length of the airfoil along the span, is the inflow angle of the blade unit in the local wind direction, represents the local inflow angle of the blade, express Number of blades , Tip Speed ​​Ratio and the tip shape factor related, is the rotor radius, is the number of leaves; The expression of airflow in the fan blade ice accumulation model is: ， In the formula, is the laminar kinematic viscosity, is the dimensionless laminar viscosity, , are all truncation functions, is the vorticity, is the Reynolds number at infinity, , , as well as are the closure coefficients, is a closed function, is the distance from the wall, is the Stefan-Boltzmann constant, is the Hamiltonian operator, is the speed change; The Euler model is used for impact calculations, where the continuity and momentum equations associated with the droplet are expressed as: ， ， In the formula, is the droplet volume fraction, is the air flow rate, is the droplet relative velocity, is the droplet drag coefficient, is the droplet Reynolds number, is the droplet inertia parameter, is the air density, is the density of water, is the Froude number, is the acceleration due to gravity; According to the droplet collection efficiency Solving the freezing process, the expressions of the partial differential equations for mass conservation and energy conservation are obtained as follows: ， In the formula, is the water film density, is the height, is the sign of the divergence, is the flow rate, is the reference speed, is the liquid water content, is the evaporation rate of liquid water, is the freezing rate of liquid water; ， In the formula, is the specific heat capacity of the water film, is the latent heat of evaporation, is the latent heat of sublimation, is the latent heat of melting, is the specific heat capacity of ice, is the solid heat dissipation coefficient, is the reference temperature, is the reference droplet temperature in Kelvin, is the reference droplet temperature in degrees Celsius, is the water-ice interface temperature in degrees Celsius, is the heat transferred by convection, is the water-ice interface temperature in Kelvin; ， In the formula, is the normal vector of the blade surface; The expression of the ice melting rate of the natural deicing model of the fan blades is: ， In the formula, is the diameter of the frozen object, is the convection exchange coefficient, , are the air temperature and ice surface temperature, is the radiation constant, is the specific heat capacity of ice, is the latent heat of ice, is the density of ice; The expression for ice thickness when ice falls off is: ， In the formula, is the thickness of ice when it falls off, is the viscous shear stress, is the wind turbine diameter, is the blade rotation rate.

3. The wind turbine power prediction method based on Bayesian deep learning according to claim 1 is characterized in that: The expression of the mixed sample feature is: ， In the formula, is the air flow rate, is the blade rotation speed, is the tip speed ratio, is the pitch angle, is the number of leaves, is the rotor radius, is the yaw error, is the air density, is the inflow wind speed, is the liquid water content, is the droplet volume fraction, is the droplet relative velocity, is the water-ice interface temperature in degrees Celsius, is the water-ice interface temperature in Kelvin, is the convection exchange coefficient, is the air surface, is the ice surface temperature, is the density of ice.

4. The wind turbine power prediction method based on Bayesian deep learning according to claim 1 is characterized in that: The adaptive wind power prediction model includes: A variational Bayesian gated recurrent unit, wherein the variational Bayesian gated recurrent unit is used to process time series observation history data in SCADA, wherein the time series observation history data in SCADA contains mixed sample features; Variational Bayesian neuron, the basic unit of the variational Bayesian neuron is an artificial neuron, and the expression is: ， In the formula, is the deviation, is the output of the variational Bayesian neuron, is the activation function, is the weight assigned to the network connection, is the input of the variational Bayes neuron, is the input quantity; GRU unit, the expression of the GRU unit is: ， ， ， ， ， In the formula, To reset the gate, To update the gate, , are the hidden states of the GRU unit at the t-th iteration and the hidden states of the GRU unit at the t-1-th iteration, respectively. For Hadamard, is the sigmoid function, is a candidate hidden state, is the output of the GRU unit, is the input of the GRU unit at time t, is the weight of the reset gate output occupied by the previous hidden state, The weight of the reset gate output is the weight of the current input. is the weight of the update gate output occupied by the previous hidden state, is the weight of the current input in updating the gate output, is the weight of the candidate hidden state occupied by the previous hidden state and the reset gate Hadamard product, is the weight of the candidate hidden state occupied by the current input, is the hyperbolic tangent function.

5. The wind turbine power prediction method based on Bayesian deep learning according to claim 1 is characterized in that: The step of generating a wind turbine power output expert experience sample database according to the physical model and using the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model includes: Inputting wind turbine geographic data, weather forecast data, and historical operation data observed by SCADA into the physical model to calculate wind turbine power, and obtaining a wind turbine power output expert experience sample database; The constructed adaptive wind power prediction model is pre-trained based on the database expert experience sample database, where: Inputs to the time-adaptive wind power prediction model for: ， In the formula, for Historical operation data at all times, is the wind turbine geographic data, for Weather forecast data at all times; The output of the time-adaptive wind power prediction model is Power output of the fan at the moment , the expression is: ， In the formula, for The wind power output of the wth wind turbine at time, for Moment Wind power output of a typhoon turbine.

6. A wind turbine power prediction system based on Bayesian deep learning, characterized in that: include: A first building module is configured to build a physical model of the impact of icing process on wind turbine output; A screening module is configured to determine sample characteristics of the measurement system according to the real-time measurement data, and then screen mixed sample characteristics related to wind turbine power prediction under icing conditions according to the physical model; A second building module is configured to take the mixed sample features as input to build an adaptive wind power prediction model based on a Bayesian multi-layer neural network; A training module is configured to generate a wind turbine power output expert experience sample database according to the physical model, and use the wind turbine power output expert experience sample database as a priori pre-training to construct an adaptive wind power prediction model, collect historical operation data of the wind farm as a posteriori, and regularly retrain the adaptive wind power prediction model based on a random variational derivation method to obtain a wind turbine power prediction model; The output module is configured to obtain real-time data of wind turbine operation and input the real-time data of wind turbine operation into the wind turbine power prediction model, and the wind turbine power prediction model outputs a wind turbine power prediction result.

7. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images