CFD and big data combined wind power intelligent prediction method and system
Through the intelligent wind power prediction method combined with CFD and big data, problems such as large amount of calculation and insufficient data in wind power prediction in wind farms are solved, and high-precision wind power prediction in wind farms are achieved, supporting real-time operation optimization of wind farms.
Patent Information
- Application Number
- CN202510215884.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-26
- Publication Date
- 2025-06-13
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has problems such as large amount of calculation, difficulty in real-time guidance of decision-making, insufficient data, and low modeling accuracy, and data not sharing of different operators in wind power prediction.
Using the intelligent wind power prediction method combined with CFD and big data, wind power prediction is carried out by building a basic cloud platform database, using BiT-LSTM wind speed prediction model and PINN wind power prediction model, combined with SCADA data, CFD micro-scale simulation data and WRF mesoscale simulation data, wind power prediction data are carried out.
It improves the accuracy and reliability of wind power prediction in wind farms, can guide the operation, maintenance and power generation plan of wind farms in real time, and optimize the energy structure.
Smart Images
Figure CN120146613A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of efficient utilization of wind energy. Specifically, it relates to a wind power intelligent prediction method and system combining CFD and big data. Background Art
[0002] With the intensification of the energy crisis, wind energy, as a clean energy, has received increasing attention. The inherent characteristics of wind energy such as intermittency, volatility, and randomness have brought many uncontrollable factors to the collection and rational utilization of wind energy. Therefore, improving the utilization rate of wind energy resources requires the support of high-precision and reliable wind power prediction results of wind farms.
[0003] Accurate wind power prediction, especially for offshore wind power and wind power prediction in complex terrains, helps to promote the transformation of the energy structure and achieve green and low-carbon development of energy. However, wind energy resources, especially offshore wind energy and wind energy in complex terrain areas, have complex and changeable characteristics, and different factors such as wind speed, wind direction, temperature, and humidity will all affect the final wind energy prediction. Summary of the Invention
[0004] In view of the above problems, the purpose of the present invention is to provide a wind power intelligent prediction method and system combining CFD and big data to solve the problems such as large computational amount when considering CFD and WRF, which is difficult to guide decision-making in real time, insufficient data for newly built wind farms, low modeling accuracy, and non-sharing of data among different operators.
[0005] The technical solution of the present invention is as follows: The wind power intelligent prediction method combining CFD and big data of the present invention has the following operating steps:
[0006] Step (1): Construct a basic database of the cloud platform, including basic meteorological element data at microscale, mesoscale, and stations.
[0007] Step (2): Considering N s wind farms in the research area, due to the competition in the operation and management of different wind farms, BiT-LSTM wind speed prediction models are respectively constructed for different wind farms. The hyperparameter set in the BiT-LSTM model is optimized by the ivy algorithm lvy, and the minimum composite loss function is used as the search direction for the hyperparameter set.
[0008] Step (3): Construct a general PINN wind power prediction model, dynamically correct the wind energy utilization factor according to the actual operation data of the wind farm collected by SCADA, and at the same time consider the uncertainty principle, and use the 3-sigma rule to correct the prediction error of converting the BiT-LSTM wind speed into wind power to obtain the predicted power interval.
[0009] Step (4): Carry out operation and maintenance of the wind farm, report the power generation plan, and then optimize the energy structure according to the predicted power interval.
[0010] Furthermore, in step (1), the microscale meteorological element basic database includes the microscale wind field distributions formed under different types of weather conditions, specifically the wind speed, temperature, air pressure, and wind direction at 0 - 10m, which are generated by simulation using the CFD model;
[0011] The mesoscale meteorological element basic database includes the wind speed, temperature, and air pressure at the kilometer level, which are generated by WRF simulation;
[0012] The station meteorological element basic database includes the wind speed, temperature, humidity, air pressure, and wind direction monitored by the meteorological station;
[0013] Among them, both the CFD simulation and the WRF simulation are carried out offline.
[0014] Furthermore, in step (2), there are a total of N s wind farms in the study area. The calculation formula for constructing the BiT - LSTM wind speed prediction model for different wind farms is:
[0015]
[0016] Q f = BiLSTM(Q′ m , Para 2 )
[0017] o = FC(Q f )
[0018] In the formula, represents the input features of the j - th wind farm, including the historical wind speed and the time series of meteorological features associated with the wind speed change at the next moment; α represents the positional encoding; MHA, BiTCN, Layernorm, BiLSTM, and FC respectively refer to the multi - head attention layer, bidirectional temporal convolutional network layer, normalization layer, bidirectional long short - term memory network layer, and fully - connected layer; Para 2 represents the hyperparameters of BiLSTM, and Para 1 represents the hyperparameters of BiTCN.
[0019] Furthermore, in step (2), the BiT - LSTM wind speed prediction model specifically uses the historical wind speed and its associated meteorological feature time series as the input for predicting the wind speed change at the next moment. The BiTCN model is used as the basic model, and BiTCN is combined with BiLSTM and the dynamic attention mechanism.
[0020] Furthermore, in step (2), the hyperparameter set [Para 1 , Para 2 of the BiT - LSTM model is optimized by the ivy algorithm lvy, and its calculation formula is:
[0021]
[0022] Wherein, P i represents the i-th set of hyperparameter sets; β = 1 + rand / 2 represents the control parameter update probability; D represents the parameter dimension in the set of hyperparameters to be optimized; N pop represents the total number of hyperparameter sets; % represents the Hadamard division; Gv represents the optimization rate; r represents a random number between [0, 1]; f(·) represents the composite loss function; its update optimization rate is as follows:
[0023]
[0024] Furthermore, in step (2), the composite loss function comprehensively considers the predicted wind speed error L data , the physical mechanism error L ph , and the deviation L d between the predicted wind speed distribution and the actual wind speed distribution; the wind speed and its associated meteorological elements in the cloud platform basic database are used as the input for improving the BiT-LSTM wind speed prediction model, and each sub BiT-LSTM model is stacked. Its specific calculation formula is:
[0025] f(S t , Para, BiT-LSTM) = ω 1 L data + ω 2 L ph + ω 3 L d
[0026]
[0027] L ph = L PDE + L IC + L BC
[0028] L d = Wa(Di B , Di s )
[0029] Wherein, ω 1 ~ω 3 represent dynamic factors, and the proportion of each part is adjusted according to the actual situation; δ represents the threshold for controlling the smoothing range of the predicted data error; L PDE , L IC , L BC respectively represent the differential equation, initial condition, and boundary condition errors; Di B , Di srespectively represent the wind speed distribution predicted by BiT-LSTM and Di s the wind speed distribution of the monitoring station, and Wa(·) represents the Wasserstein distance.
[0030] Furthermore, in step (3), the calculation formula of the general PINN wind power prediction model is:
[0031]
[0032] P fore =[P wind -3σ, P wind +3σ]
[0033] wherein, P fore represents the PINN predicted wind power interval, P wind represents the wind power obtained from the predicted wind speed, σ is the standard deviation of P wind , λ i represents the importance factor of the jth wind farm, and its value is determined according to the energy entropy of the prediction error sequence of the BiT-LSTM wind speed prediction model; is the predicted wind speed; C represents the wind energy utilization factor, which is dynamically corrected according to the actual operation data of the wind farm collected by SCADA; ρ represents the density of air; A represents the wind turbine area.
[0034] Furthermore, in step (3), the actual operation data of the wind farm collected by SCADA is stored in the basic database of the cloud platform.
[0035] Furthermore, a computer-readable storage medium, characterized in that a wind power prediction program is stored thereon, and when the wind power prediction program is executed by a processor, the steps of the above method are implemented.
[0036] The beneficial effects of the present invention are as follows: 1. Taking SCADA monitoring and collected data, CFD micro-scale simulation data, and WRF meso-scale simulation data as input or reference for correction, making full use of data complementarity; 2. Combining numerical simulations of CFD and WRF with big data from site monitoring to drive artificial intelligence models, solving the problems of insufficient information in traditional site monitoring data alone and difficulty in considering the physical mechanisms of the influence of terrain, turbulence, and wake effects on wind field changes, comprehensively enhancing the physical mechanism and data reliability of artificial intelligence wind power prediction; 3. Constructing a BiT-LSTM model that integrates the advantages of multiple models, designing a composite loss function that comprehensively considers prediction error, physical mechanism error, and wind speed distribution error, comprehensively enhancing the accuracy and effectiveness of the model; 4. Improving the BiT-LSTM model with the high-precision wind field prediction results of CFD, WRF simulation results, and site monitoring information as input, fully considering the spatio-temporal heterogeneity of wind field wind speed, and integrating multi-scale spatio-temporal meteorological data of stations, fields, and regions, which can effectively extract spatio-temporal features and improve the power prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is the overall operation block diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0038] The following further elaborates on the specific technical solutions of the present invention in combination with specific examples.
[0039] As shown in the figure, the intelligent wind power prediction method combining CFD and big data according to the present invention includes the following steps:
[0040] Step (1): Construct a cloud platform basic database, including basic meteorological element data of micro-scale, meso-scale, and sites.
[0041] Step (2): Considering N s wind farms in the research area, there is commercial competition in the operation and management of different wind farms. Therefore, a BiT-LSTM wind speed prediction model is constructed for each different wind farm. The hyperparameter set of the BiT-LSTM model is optimized using the ivy algorithm lvy, and the minimum of the composite loss function is used as the search direction for the hyperparameter set.
[0042] Step (3): Construct a general PINN wind power prediction model, dynamically correct the wind energy utilization factor according to the actual operation data of the wind farm collected by SCADA, and at the same time consider the uncertainty principle, and use the 3σ criterion to correct the prediction error of converting the BiT-LSTM wind speed into wind power to obtain the predicted power interval.
[0043] Step (4): Carry out operation and maintenance of the wind farm, report the power generation plan, and then optimize the energy structure according to the predicted power interval.
[0044] Further, in step (1), the microscale meteorological element basic database contains the microscale wind field distributions formed under different types of weather conditions, specifically the wind speed, temperature, air pressure, and wind direction at 0 - 10m, which are generated by CFD model simulation;
[0045] The mesoscale meteorological element basic database contains the wind speed, temperature, and air pressure at the kilometer level, which are generated by WRF simulation;
[0046] The station meteorological element basic database contains the wind speed, temperature, humidity, air pressure, and wind direction monitored by meteorological stations; among them, both the CFD simulation and the WRF simulation are performed offline.
[0047] In step (2), there are a total of N s wind farms in the research area. The calculation formula for constructing the BiT - LSTM wind speed prediction model for different wind farms is:
[0048]
[0049] Q f = BiLSTM(Q′ m , Para 2 )
[0050]
[0051] In the formula, represents the input features of the j - th wind farm, including the historical wind speed and the time series of meteorological features associated with the wind speed change at the next moment; α represents the position encoding; MHA, BiTCN, Layernorm, BiLSTM, and FC refer to the multi - head attention layer, bidirectional temporal convolutional network layer, normalization layer, bidirectional long short - term memory network layer, and fully - connected layer respectively; Para 2 represents the hyperparameters of BiLSTM, and Para 1 represents the hyperparameters of BiTCN.
[0052] In step (2), the BiT - LSTM wind speed prediction model uses the historical wind speed and its associated meteorological feature time series as the input for predicting the wind speed change at the next moment. The BiTCN model is used as the basic model to extract the dynamic association relationship between the historical wind speed and the associated meteorological features. BiTCN overcomes the problem that traditional TCN only considers historical association information and ignores the impact of predicted meteorological information on the wind speed. BiTCN can not only learn the evolution law of historical wind speed but also consider the positive feedback of historical meteorological features and predicted meteorological features on the wind speed change; BiTCN is combined with BiLSTM and the dynamic attention mechanism to improve the prediction performance.
[0053] In step (2), the hyperparameter set of the BiT - LSTM model [Para1 , Para 2 Optimized by the Ivy algorithm, and its calculation formula is:
[0054]
[0055] In the formula, P i represents the i-th set of hyperparameter sets; β = 1 + rand / 2 represents the control parameter update probability; D represents the dimension of the parameters in the set of hyperparameters to be optimized; N pop represents the total number of hyperparameter sets; % represents the Hadamard division; Gv represents the optimization rate; r represents a random number between [0, 1]; f(·) represents the composite loss function; its updated optimization rate is as follows:
[0056]
[0057] In step (2), the composite objective function comprehensively considers the prediction wind speed error L data , the physical mechanism error L ph , and the deviation L d between the predicted wind speed distribution and the actual wind speed distribution. The wind speed and its associated meteorological elements in the cloud platform basic database are used as the input for improving the BiT-LSTM wind speed prediction model, and each sub-BiT-LSTM model is stacked for refined prediction. Its specific calculation formula is:
[0058] f(S t , Para, BiT-LSTM) = ω 1 L data + ω 2 L ph + ω 3 L d
[0059]
[0060] L ph = L PDE + L IC + L BC
[0061] L d = Wa(Di B , Di s )
[0062] In the formula, ω 1 ~ω 3 represent dynamic factors, and the proportion of each part is adjusted according to the actual situation; δ represents the threshold for controlling the smoothing range of the prediction data error; L PDE , L IC , L BCrepresent the differential equation, initial condition, and boundary condition errors; Di B , Di s represent the predicted wind speed distribution by BiT-LSTM and Di s the wind speed distribution of the monitoring station, respectively. Wa(·) represents the Wasserstein distance.
[0063] In step (3), the calculation formula of the general PINN wind power prediction model is:
[0064]
[0065] P fore =[P wind -3σ, P wind +3σ]
[0066] In the formula, P fore represents the PINN predicted wind power interval, P wind represents the wind power obtained from the predicted wind speed, σ is the standard deviation of P wind , and λ i represents the importance factor of the j-th wind farm, and its value is determined according to the energy entropy of the prediction error sequence of the BiT-LSTM wind speed prediction model; is the predicted wind speed; C represents the wind energy utilization factor, which is dynamically corrected according to the actual operation data of the wind farm collected by SCADA; ρ represents the density of air; A represents the wind turbine area.
[0067] Furthermore, in step (3), the actual operation data of the wind farm collected by SCADA is stored in the basic database of the cloud platform.
[0068] Furthermore, a computer-readable storage medium stores a wind power prediction program, and when the wind power prediction program is executed by a processor, the steps of the method are implemented.
Claims
1. The wind power intelligent prediction method combining CFD and big data is characterized by: The operation steps are as follows: Step (1): Build a cloud platform basic database, including basic data of meteorological elements at microscale, mesoscale and site levels; Step (2): Consider N s There are competitions in the operation and management of different wind farms. Therefore, BiT-LSTM wind speed prediction models are constructed for different wind farms. The hyperparameter set in the BiT-LSTM model is optimized using the Ivy algorithm, and the search direction of the hyperparameter set is to minimize the composite loss function. Step (3): Construct a general PINN wind power prediction model, dynamically correct the wind energy utilization factor based on the actual operation data of the wind farm collected by SCADA, and consider the uncertainty principle. Use the Laida criterion to correct the prediction error of BiT-LSTM wind speed conversion to wind power, and obtain the predicted power range; Step (4): Perform wind farm operation and maintenance and report power generation plans based on the predicted power range to optimize the energy structure.
2. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (1), the micro-scale meteorological element basic database includes the micro-scale wind field distribution formed under different types of weather conditions, specifically the wind speed, temperature, air pressure and wind direction of 0-10m, which is generated by CFD model simulation; The basic database of mesoscale meteorological elements includes kilometer-level wind speed, temperature, and air pressure, which are generated by WRF simulation; The site meteorological element basic database includes wind speed, temperature and humidity, air pressure and wind direction monitored by the meteorological station.
3. The wind power intelligent prediction method combining CFD and big data according to claim 2 is characterized in that: The CFD simulation and WRF simulation are both performed offline.
4. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (2), the total number of s Wind farms, the calculation formula for constructing the BiT-LSTM wind speed prediction model for different wind farms is: Q f =BiLSTM(Q′ m ,,Para2) o=FC(Q f ) In the formula, represents the input features of the j-th wind farm, including the historical wind speed and the meteorological characteristic time series associated with the wind speed change at the next moment; α represents the position encoding; MHA, BiTCN, Layernorm, BiLSTM and FC refer to the multi-head attention layer, bidirectional temporal convolutional network layer, normalization layer, bidirectional long short-term memory network layer and fully connected layer respectively; Para2 represents the hyperparameters of BiLSTM, and Para1 represents the hyperparameters of BiTCN.
5. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (2), the BiT-LSTM wind speed prediction model specifically uses the historical wind speed and its associated meteorological characteristic time series as the input for predicting the wind speed change at the next moment, uses the BiTCN model as the basic model, and combines BiTCN with BiLSTM and the dynamic attention mechanism.
6. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (2), the hyperparameter set [Para1, Para2] of the BiT-LSTM model is optimized by the Ivy algorithm lvy, and its calculation formula is: Where P i represents the i-th set of hyperparameters; β = 1 + rand / 2 represents the probability of controlling parameter updates; D represents the parameter dimension in the hyperparameter set to be optimized; N pop represents the total number of hyperparameter sets; % represents Hadamard division; Gv represents the optimization rate; r represents a random number between [0,1]; f(·) represents the composite loss function; its update optimization rate is as follows:
7. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (2), the composite loss function comprehensively considers the predicted wind speed error L involved in the modeling process. data , physical mechanism error L ph and the deviation L between the predicted wind speed distribution and the actual wind speed distribution d ; The wind speed and its associated meteorological elements in the cloud platform basic database are used as the input of the improved BiT-LSTM wind speed prediction model, and each sub-BiT-LSTM model is stacked. The specific calculation formula is: f(S t ,Para,BiT-LSTM)=ω1L data +ω2L ph +ω3L d L ph =L PDE +L IC +L BC L d =Wa(Di B ,Di s ) In the formula, ω1~ω3 represent dynamic factors, and the proportion of each part is adjusted according to the actual situation; δ represents the threshold value for controlling the smoothing range of the prediction data error; L PDE , L IC , L BC Respectively represent the differential equation, initial condition and boundary condition errors; Di B 、Di s Represents BiT-LSTM predicted wind speed distribution and Di s Wind speed distribution at the monitoring station. Wa(·) represents the Wasserstein distance.
8. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (3), the calculation formula of the general PINN wind power prediction model is: P foe =[P wind -3σ,P wind +3σ] Where P fore represents the wind power range predicted by PINN, P wind represents the wind power obtained from the predicted wind speed, σ is P wind The standard deviation of i It represents the importance factor of the j-th wind farm, and its value is determined according to the energy entropy of the forecast error sequence of the BiT-LSTM wind speed forecast model; represents the predicted wind speed; C represents the wind energy utilization factor, which is dynamically corrected according to the actual operation data of the wind farm collected by SCADA; ρ represents the density of air; A represents the area of the wind rotor.
9. The wind power intelligent prediction method combining CFD and big data according to claim 1 is characterized in that: In step (3), the SCADA collects actual operation data of the wind farm and stores it in the cloud platform basic database.
10. A computer-readable storage medium, characterized in that: A wind power prediction program is stored thereon, and when the wind power prediction program is executed by a processor, the steps of the method as described in any one of claims 1 to 9 are implemented.