Summary of the invention
Technical problem to be solved by this invention optimizes the power of Wind turbines, provides a kind of method and system of Wind turbines power optimization.
The technological scheme that the present invention solves the problems of the technologies described above is as follows: a kind of method of Wind turbines power optimization, comprises the steps,
Step 1, gathers running of wind generating set data;
The described data collected are processed, namely choose the data of level time resolution second by step 2, reject fault data;
Step 3, according to data construct generated output model P (t) after process and vibration amplitude model A (t);
Step 4, the generated output model that Select Error mean square value is minimum and vibration amplitude model are optimal power generation power module and optimum vibration amplitude model;
Step 5, according to optimal power generation power module and optimum vibration amplitude model buildings power optimization function and vibration majorized function.
On the basis of technique scheme, the present invention can also do following improvement.
Further, the data gathered in step 1 comprise the following operational parameter data collection of Wind turbines, wind speed v (t), driftage y (t), and wind direction d (t) becomes vane angle p (t), temperature T (t).
Further, the fault data in step 2 comprises data because Wind turbines fault collection arrives and obviously abnormal data.
Further, P (t) in step 3=f (v (t), y (t), d (t), p (t), T (t))+error, A (t)=g (v (t), y (t), d (t), p (t), T (t))+error, wherein f (), g () the regression class algorithm that is data mining, t is the working time of Wind turbines, and error is the error amount of described generated output model and vibration amplitude model.
Further, also comprise in step 4 and determine described optimal power generation power module and wind speed v (t) corresponding to described optimum vibration amplitude model, driftage y (t), wind direction d (t), become the parameter value of vane angle p (t) and temperature T (t).
Further, in step 4, treated data are divided into two-part at random, wherein 70% for the training data of generated output model and vibration amplitude model training, and other 30% for the predictive data set of MSE error mean square.
Further, before carry out step 5, resampling can be adopted to the non-equilibrium data of training data part, make the quantity in parameter wind speed v (t) in described generated output model P (t) and vibration amplitude model A (t), driftage y (t), wind direction d (t), change vane angle p (t) and temperature T (t) each interval in five dimension state spaces be no less than predefined value n1, n1 represents interval censored data quantity predefined value.
Further, in step 5 before building majorized function, need parameter wind speed v (t), driftage y (t), wind direction d (t), become vane angle p (t) and the controlled optimization of temperature T (t).
Further, in step 5, power optimization function is Tp=max{P (t) }, Tp refers to the performance number after power optimization, vibration majorized function is TA=max{A (t) }, wherein TA refers to the vibration after optimization, C=max (Tp|TA<=C), C represents vibration limit value, namely vibration can power under acceptable conditions best.
The another kind of technological scheme that the present invention solves the problems of the technologies described above is as follows: a kind of system of Wind turbines power optimization, comprises data acquisition module, data processing module, model construction module, optimal models builds module and majorized function builds module;
Described data acquisition module is for gathering running of wind generating set data;
Described data processing module is used for the described data collected to process, and namely chooses the data of level time resolution second, rejects fault data;
Described model construction module is used for data construct generated output model P (t) after according to process and vibration amplitude model A (t);
Described optimal models structure module is used for the minimum generated output model of Select Error mean square value and vibration amplitude model is optimal power generation power module and optimum vibration amplitude model;
Described majorized function builds module for according to optimal power generation power module and optimum vibration amplitude model buildings power optimization function and vibration majorized function.
The invention has the beneficial effects as follows: from the power optimization of the angle research Wind turbines of Controlling model, consider the optimization of Transmitted chains and the vibration of tower cylinder simultaneously.
Embodiment
Be described principle of the present invention and feature below in conjunction with accompanying drawing, example, only for explaining the present invention, is not intended to limit scope of the present invention.
The present invention, from the power optimization of the angle research Wind turbines of Controlling model, considers the optimization of Transmitted chains and the vibration of tower cylinder simultaneously in optimization aim.The present invention has inquired into data digging method to optimize the basic ideas framework of blower fan production capacity and vibration, and achieves by change oar control power optimization with an Engineering Projects.
As shown in Figure 1, a kind of method of Wind turbines power optimization, comprises the steps:
Step 1, gathers running of wind generating set data, with the running of wind generating set time for variable, gather the following operational parameter data collection of Wind turbines, wind speed v (t), driftage y (t), wind direction d (t), becomes vane angle p (t), temperature T (t);
Step 2, data processing, selects the data of right times resolution, according to engineering experience, generally chooses the data of second rank, rejects fault data, comprises the data because Wind turbines fault collection arrives and obviously abnormal data.The process of this data processing is conducive to the degree of accuracy improving data, ensures the validity of subsequent calculations.
Step 3, according to data construct generated output model P (t) after process and vibration amplitude model A (t), wherein P (t)=f (v (t), y (t), d (t), p (t), T (t))+error, A (t)=g (v (t), y (t), d (t), p (t), T (t))+error, wherein f (.), the regression class algorithm that g (.) is data mining, what can realize above-mentioned effect in data mining algorithm has neural network, support vector machine, k nearest neighbor and random forest method etc., error is the error amount of model.
Step 4, the generated output model that Select Error mean square value is minimum and vibration amplitude model are optimal power generation power module and optimum vibration amplitude model, and determine this optimal power generation power module and wind speed v (t) corresponding to optimum vibration amplitude model, driftage y (t), wind direction d (t), become the parameter value of vane angle p (t) and temperature T (t).In this process, treated data are divided into two-part at random, wherein 70% for the training data of model training, and other 30% for the predictive data set of MSE (i.e. error mean square), and wherein MSE adopts predictive data set to be to prevent overfitting.
In order to reduce calculated load, speed-up computation, resampling is adopted to the non-equilibrium data of training data part, make parameter wind speed v (t) in above-mentioned generated output model P (t) and vibration amplitude model A (t), driftage y (t), wind direction d (t), become vane angle p (t) with temperature T (t) in n (corresponding number of parameters, namely n is 5) quantity in each interval is no less than predefined value n1 (n1 depends on total training sample amount in dimension state space, under interval number determination condition, training sample is larger, n1 is larger, data volume after resampling can not exceed former training data), even if the quantity that parameter is tieed up in state space at certain n is no less than n1.So both remain original training data and can cover all possible point, additionally reduce the volume of data.
Step 5, according to optimal power generation power module and optimum vibration amplitude model buildings power optimization function and vibration majorized function; Before building majorized function, parameter should adopt controlled optimization, to guarantee that the result optimized has certain actual physical meaning; Be that to become the power optimization of vane angle be example with parameter, optimization aim can be decomposed into, Tp=max{P (t) }, Tp refers to the performance number after power optimization, wherein becoming vane angle change should controlled optimization within the specific limits, p_min<=p (t) <=p_max, its scope is currency+-someValue (giving tacit consent to 10 °).
Equally, the majorized function that vibration is optimized can be built, TA=max{A (t) }, wherein becoming vane angle change should in the controlled optimization of certain limit, p_min<=p (t) <=p_max, its scope is currency+-someValue (giving tacit consent to 10 °).
The simultaneously majorized function of optimizing power and vibration, variable element is for becoming propeller angle, and C=max (Tp|TA<=C), C represent vibration limit value, namely vibration can power under acceptable conditions best.
As shown in Figure 2, a kind of system of Wind turbines power optimization, comprise data acquisition module, data processing module, model construction module, optimal models build module, training data resampling module and majorized function build module;
It is variable that data acquisition module was used for time, gathers the following operational parameter data collection of Wind turbines, wind speed v (t), driftage y (t), and wind direction d (t) becomes vane angle p (t), temperature T (t);
Data processing module, for selecting the data of right times resolution, according to engineering experience, generally chooses the data of second rank, rejects fault data, comprises the data because Wind turbines fault collection arrives and obviously abnormal data.The process of this data processing is conducive to the degree of accuracy improving data, ensures the validity of subsequent calculations.
Model construction module is used for data construct generated output model P (t) after according to process and vibration amplitude model A (t), wherein P (t)=f (v (t), y (t), d (t), p (t), T (t))+error, A (t)=g (v (t), y (t), d (t), p (t), T (t))+error, wherein f (), the regression class algorithm that g () is data mining, what can realize above-mentioned effect in data mining algorithm has neural network, support vector machine, k nearest neighbor and random forest method etc., error is the error amount of model.
Optimal models builds module and is used for the minimum generated output model of Select Error mean square value and vibration amplitude model is optimal power generation power module and optimum vibration amplitude model, and determines this optimal power generation power module and wind speed v (t) corresponding to optimum vibration amplitude model, driftage y (t), wind direction d (t), becomes the parameter value of vane angle p (t) and temperature T (t).In this process, treated data are divided into two-part at random, wherein 70% for the training data of model training, and other 30% for the predictive data set of MSE (i.e. error mean square), and wherein MSE adopts predictive data set to be to prevent overfitting.
Training data resampling module is used for adopting resampling to the non-equilibrium data of training data part, in order to reduce calculated load, speed-up computation, resampling is adopted to the non-equilibrium data of training data part, make parameter wind speed v (t) in above-mentioned generated output model P (t) and vibration amplitude model A (t), driftage y (t), wind direction d (t), become vane angle p (t) with temperature T (t) in n (corresponding number of parameters, namely n is 5) quantity in each interval is no less than predefined value n1 (n1 depends on total training sample amount in dimension state space, under interval number determination condition, training sample is larger, n1 is larger, data volume after resampling can not exceed former training data), even if the quantity that parameter is tieed up in state space at certain n is no less than n1.So both remain original training data and can cover all possible point, additionally reduce the volume of data.
Majorized function builds module for building majorized function, and before building majorized function, parameter should adopt controlled optimization, to guarantee that the result optimized has certain actual physical meaning; Be that to become the power optimization of vane angle be example with parameter, optimization aim can be decomposed into,
Tp=max{P (t) }, Tp refers to the performance number after power optimization, wherein becoming vane angle change should controlled optimization within the specific limits, p_min<=p (t) <=p_max, its scope is currency+-someValue (giving tacit consent to 10 °).
Equally, the majorized function that vibration is optimized can be built, TA=max{A (t) }, wherein becoming vane angle change should in the controlled optimization of certain limit, p_min<=p (t) <=p_max, its scope is currency+-someValue (giving tacit consent to 10 °).
The simultaneously majorized function of optimizing power and vibration, variable element is for becoming propeller angle, and max (Tp|TA<=C), C represent vibration limit value, namely vibration can power under acceptable conditions best.
The machine type data of an existing GW1500, we will set up simple power optimization function, and variable element is for becoming vane angle.
Always have 45 groups of data, in modal analysis, we are research wherein 14 groups of variablees only: WROT.PtAngVal.Bl1, WROT.PtAngVal.Bl2, WROT.PtAngVal.Bl3, WMAN.State, DateTime, WTUR.PwrAt.InstMag.f, WNAC.ExlTmp.instMag.f, WNAC.TopBoxTmp, WNAC.IntTmp.instMag.f, WROT.PtCptTmp.Bl1, WROT.PtCptTmp.Bl2, WROT.PtCptTmp.Bl3, WROT.PtCnvTmp.Bl1, WROT.PtCnvTmp.Bl2, WROT.PtCnvTmp.Bl3.
First data scrubbing is carried out.Power curve after cleaning, change oar curve and speed curves are respectively as shown in Fig. 3, Fig. 4 and Fig. 5.
Wind speed ' NacWS', driftage ' YawPos', wind direction ' Wdir', the data become in oar ' PtAngBl3', temperature ' ExlTmp' are normalized, and ' PwrAct' data are also normalized, use neuron network (setting of attention parameters), Modling model, training set data is the initial data of random 70%, if Fig. 6 is error time sequence chart.As can be seen, error time sequence at most can not more than 15%, and overall MSE is 0.0001, and model can describe the relation of each variable and power very well.
To become vane angle ' PtAngBl3' as variable element, be optimized process to power, be optimized result, becomes propeller angle optimization and adopt controlled optimization, can adjust by the rule of the fixed percentage change of currency or currency+-fixed value change.Here is one and becomes 95% confidecne curve of oar curve according to limit, carrys out the example of control and optimize.If Fig. 7 is standardized change oar curve confidence interval figure, the curve 1 marked in Fig. 7 and curve 3 are the upper and lower of confidence interval, curve 2 represents P50, Fig. 8 is the change propeller angle figure before and after optimizing, change propeller angle in Fig. 8 before curve 1 representing optimized, change propeller angle after curve 2 representing optimized, Fig. 9 is that after optimizing, power deducts the power diagram do not optimized.
Calculate the result after optimization, result before and after optimization is analyzed, and result shows, the productivity ratio after optimization improve 0.6% originally, the physical change rule that the result before and after the optimization of change propeller angle is more realistic.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.