[0045] The preferred embodiments are described in detail below in conjunction with the drawings. It should be emphasized that the following description is only exemplary, and is not intended to limit the scope and application of the present invention.
[0046] figure 1 This is the hardware structure diagram of the embodiment of the present invention. Among them, the main parameters of the small permanent magnet direct drive wind turbine 2 are: the diameter of the wind wheel is 1.2m, the rated power is 300W, the rated voltage is 24V, the rated speed is 800r/min, and it starts Wind speed 1m/s, rated wind speed 10m/s, safe wind speed 25m/s; the main parameters of the maximum power tracking device are: MPPT controller adopts dsPIC33FJ06GS101 single-chip microcomputer, DC-DC converter adopts Boost circuit, drive module chooses MCP14E3, voltage sensor 5. Choose LV28-P, current sensor 8 chooses LA25-NP, speed sensor 3 adopts voltage zero-crossing detection frequency meter, first capacitor C1=10uF, second capacitor C2=100uF.
[0047] The maximum power tracking control method of the small permanent magnet direct drive wind power generation system includes the following steps:
[0048] Step 1: Establish the firefly support vector machine regression prediction model:
[0049] Step 101: Put n Two wind speed sensors are respectively installed on the front side of the fan impeller, which is coaxial with the impeller circular plane, parallel to the center of the circle and the distance from the center of the circular plane of equal size 1/(n-1) Radius, 2/(n-1) At the radius,..., at 1 time radius, the wind speed vector can be measured V =[ V 1 , V 2 ,..., V n ] T; In the embodiment, n=4, so the wind speed vector can be measured V =[ V 1 , V 2 ,..., V 4 ] T;
[0050] Step 102: Remember the wind speed vector in a certain wind speed environment as V(i) = [ V 1 (i) , V 2 (i) ,..., V n (i) ] T , The DC voltage at the output of the rectifier in the wind power generation system is V oc (i) , The DC current is I oc (i) , The fan speed is ω oc (i) , Calculate the output power at the moment P oc (i) = V oc (i)· I oc (i) , It can form a set of samples ( V(i) , P oc (i) , ω oc (i) ); By collecting sample groups under different wind speed environments, a sample set is formed {( V(i) , P oc (i) , ω oc (i) )}; In the embodiment, a small experimental wind tunnel is formed by using a blower, a frequency converter and a straight pipe section, and the output of the blower is controlled by adjusting the frequency setting value of the frequency converter to create a different wind speed environment. Set the frequency of the inverter from 10Hz to 60Hz at intervals of 2Hz, which can provide 31 wind speed environments, namely { V(i) = [ V 1 (i) , V 2 (i) , V 3 (i) , V 4 (i) ] T , i =1, …, 31}. In each wind speed environment, by adjusting the duty ratio of the PWM signal of the DC-DC converter, the fan speed is controlled to start from 300r/min and increase to 1500r/min at 30r/min intervals. A total of 41 speeds are recorded. The DC voltage at the output of the corresponding rectifier at a speed V oc With DC current I oc , And calculate the corresponding output power P oc = V oc · I oc , Then 1271 groups of samples can be accumulated.
[0051] Step 103: Use the sample set {( V(i) , P oc (i) , ω oc (i) )} Train the firefly support vector machine regression prediction model, where the model input is { , ω oc (i) )}, the model output is V(i) In the embodiment, 900 groups of 1271 groups of samples are randomly selected as training samples of the model, and the remaining 371 groups are used as test samples of the model to test the accuracy of the model.
[0052] Support vector machine regression SVR maps the input space to a high-dimensional feature space through nonlinear mapping, and uses a linear function set for regression estimation. The firefly support vector machine of the present invention is a multi-output support vector machine, and a given sample set ( Is the input vector, Is the corresponding output vector, N is the number of samples, n is the dimension of the input vector, and m is the dimension of the output vector), the linear regression function used by the multi-output support vector machine SVR is:
[0053] (3)
[0054] among them:
[0055] y i Output vector for linear regression function;
[0056] Is the nonlinear mapping matrix from the input space to the high-dimensional feature space;
[0057] x i Is the input vector;
[0058] W Is the weight matrix;
[0059] b Is the bias vector.
[0060] Weight matrix W And bias vector b Calculate by minimizing the following formula:
[0061] (4)
[0062] Where:
[0063] C Is the penalty factor;
[0064] Is the introduced slack variable vector;
[0065] According to (4), the Lagrangian equation is established, and the linear regression function can be solved as:
[0066] (5)
[0067] Where:
[0068] K ( x i , x ) Is the kernel function,;
[0069] Is the Lagrangian coefficient, non-zero Corresponding vector x i It is called support vector.
[0070] Choosing different forms of kernel functions can generate different support vector machines. Commonly used kernel functions are: polynomial function, Gaussian function, Sigmoid function, etc. The present invention selects the Gaussian function as the kernel function, namely:
[0071] (6)
[0072] among them, Is the width parameter of the Gaussian kernel function.
[0073] The accuracy of the support vector machine regression prediction model mainly depends on two main parameters C , , Penalty factor C Used to control the compromise between model complexity and approximation error; kernel parameters Reflects the distribution or range characteristics of the training sample data. In order to obtain a high-precision regression prediction model, the present invention uses the firefly algorithm to optimize these two parameters.
[0074] The Firefly algorithm is a new type of swarm intelligence optimization algorithm proposed by Indian scholars Krishnanand and Ghose in 2005. The firefly algorithm simulates the behavior of fireflies in nature attracting their companions for courtship or foraging by luminescence. It has a good global search capability and has been successfully applied to issues such as function optimization and sensor configuration.
[0075] In the GSO algorithm, each firefly is distributed in the defined space of the objective function, and the brightness of their own luciferin is related to the fitness value of the objective function at their location. The brighter the firefly represents the better its location . Hypothesis x i (t) Means t Moment i The location of the firefly, J(x) For the fitness evaluation function, l i (t) Means t Moment i The fluorescein concentration of only fireflies, then
[0076] (7)
[0077] In formula (7): Is the fluorescein volatilization coefficient; Is the fluorescein enhancement factor, that is, the fitness extraction ratio.
[0078] Fireflies are looking for neighbors to gather within the regional decision-making range. The brighter the fireflies have the higher the attraction to attract the surrounding fireflies to move in this direction. The size of the regional decision-making range will be affected by the number of neighbors. When the neighbor density is low, the firefly's decision-making radius will increase to facilitate finding more neighbors; otherwise, the decision-making radius will decrease. The update formula of the decision domain is
[0079] (8)
[0080] In formula (8): for t+ 1 time i The decision-making range of a firefly; r s Is the range of perception; Is the neighborhood change rate; n t Is the neighbor threshold; N i ( t )for t Moment i Only the neighbors of fireflies gather.
[0081] (9)
[0082] In formula (9), Means x The norm. During the movement of fireflies, the direction of their movement is determined according to the concentration of fluorescein of each firefly in its neighbor set. P ij ( t ) Means t Moment i Only fireflies gather to their neighbors j The probability that only fireflies move, then
[0083] (10)
[0084] Use the roulette method to choose the i Only the direction in which the fireflies move i Only fireflies in t The position at time +1 is
[0085] (11)
[0086] In formula (11), Is the movement step length of the firefly.
[0087] The specific steps to optimize the support vector machine using the Firefly algorithm are:
[0088] 1) Initialize the parameters of the Firefly algorithm. Initialize the number of fireflies n , Fluorescein volatilization coefficient , Fluorescein enhancement factor γ , The range of perception r s , The neighborhood change rate β , Neighbor threshold n t , Move step s , Initial fluorescein l 0 And the maximum number of iterations, perception range r s For the parameters to be optimized in the support vector machine regression prediction model (penalty factor C And nuclear parameters ) The range of change. In the embodiment, set n= 20, =0.4, β =0.08, n t =5, s =0.03, initial fluorescein l 0 =5, the maximum number of iterations is 1000, and the penalty factor C The range of values is [10 -1 ,10 4 ], nuclear parameters The range of values is [10 -2 ,10 4 ];
[0089] 2) Parameters to be optimized C versus Within the value range of, randomly generate a set of values as the initial position of the individual firefly. The fitness evaluation function uses the location of the firefly (a certain C versus ) The negative value of the average relative error of the trained support vector machine regression prediction model on the test sample. Average relative error The expression is:
[0090] (12)
[0091] among them, N 1 Is the number of test samples, Is the true value of the wind speed vector, The estimated value of the wind speed vector output by the regression model. It can be seen that the smaller the average relative error of the model on the test sample, the higher the fitness;
[0092] 3) According to formulas (7)-(11), calculate the fluorescein concentration of each firefly, the decision domain range and the neighbor's fluorescein concentration, and use the neighbor's fluorescein concentration to determine the moving direction of each firefly and move forward;
[0093] 4) Judge whether the Firefly algorithm reaches the termination condition. If it reaches, select the firefly position with the highest fitness as the parameter of the support vector machine model, and record the support vector in the corresponding regression model for use; otherwise, go to 3). In this embodiment, the set termination condition is the average relative error of the support vector machine regression prediction model on the test sample 3% or the number of iterations exceeds the maximum number of iterations. After 128 iterations, the optimization process ends, and the parameters of the support vector machine model with the highest adaptability are C =486.2, =69.54, the model includes a total of 89 support vectors.
[0094] Using the obtained support vector machine regression prediction model, the known power and speed can be estimated { P oc , ω oc } Wind speed vector under working conditions, such as figure 2 Shown.
[0095] Step 2: Use the firefly support vector machine regression prediction model to estimate wind speed: P oc (i) = V oc (i)· I oc (i)
[0096] At a certain moment, the voltage sensor and current sensor are used to collect the DC output voltage after passing through the rectifier in the wind power generation system. V oc And output current I oc , Use the speed sensor to collect the fan speed ω oc , And calculate the output power at the moment P oc = V oc · I oc;will{ P oc , ω oc }As the input of the Firefly support vector machine regression prediction model, the model output is the estimated value of the current wind speed vector , And get the estimated value of effective wind speed according to the following formula V m :
[0097]
[0098] In this embodiment, the DC output voltage after the rectifier collected at a certain moment V oc 15V, output current I oc 12A, the output power at the moment is 180W, and the speed information collected by the speed sensor ω oc Is 600r/min, using {180W, 600r/min} as the input of the firefly support vector machine regression prediction model, and the estimated value of the output wind speed vector is V =[8.92, 9.14, 8.94, 9.02] T , The estimated value of effective wind speed V m =9.0m/s.
[0099] Step 3: According to the best tip speed ratio method, get the wind speed V m Corresponding prediction value of optimal fan speed ω ref :
[0100]
[0101] among them, Is the tip speed ratio; R Is the radius of the impeller; the tip speed ratio is the ratio of the tip tangential speed to the wind speed in front of the wind wheel. According to the formula of wind energy utilization rate, there is a fixed tip speed ratio at different wind speeds so that the wind energy utilization rate is fixed at Maximum value, such as image 3 Shown. In the embodiment, take =6.7, when the effective wind speed is 9m/s, R =0.6m, the best predicted value of fan speed is 961r/min.
[0102] Step 4: Adjust the fan speed through the proportional integral control method to achieve the best fan speed prediction value ω ref;and ω ref As the initial value, the disturbance observation method is used to track the maximum power of the fan with the set disturbance step Δω; in the embodiment, the disturbance step Δω=5r/min.
[0103] Figure 4 It is the flow chart of the Perturbation Observation Method (P&O). The principle is to periodically perturb the fan speed ( ω+ Δ ω ), then compare the power changes before and after the disturbance. If the output power increases, it means that the disturbance direction is correct and continue to the same direction ( + Δ ω ) Disturbance; if the output power decreases, the opposite direction (-Δ ω ) Direction disturbance. In the embodiment, since the AD module of dsPIC33FJ06GS101 only needs 0.5 to complete one sampling u s. When measuring the DC output voltage and DC output current of the fan, in order to eliminate the signal jitter caused by high-frequency chopping in the DC-DC converter, the AD module is made to convert 10 times and take the average value as the measured value.
[0104] Step 5: When the power difference before and after the disturbance obtained by the disturbance observation method is greater than or equal to the set threshold T r When the wind speed has a sudden change, repeat steps 2 to 5; otherwise, continue to use the disturbance observation method to track the maximum power of the fan with the set disturbance step length, such as Figure 5 Shown. In the embodiment, take T r =15W.
[0105] Pass the above maximum power tracking (MPPT) method C The language program is written into the control chip dsPIC33FJ06GS101, and the PWM signal is output to drive the DC-DC converter to realize the maximum power tracking function.
[0106] In order to verify the correctness of the proposed method, it was compared with conventional disturbance observation (P&O) on the hardware platform of the present invention. Specifically: under the same wind speed environment (set the frequency of the inverter in the wind tunnel to 40Hz), control the initial speed of the fan to ω 0 =600r/min, compare the tracking speed of the two methods and the average power of the steady state process respectively.
[0107] The tracking process of the conventional disturbance observation P&O method (adjustment period T=5s, step length Δω=20r/min) is: after 17 continuous beats of the positive direction disturbance, start at the maximum power point MPP (ω opt =950r/min) oscillate around, that is, enter the steady state process, and the total required time is about 85 seconds.
[0108] The tracking process of the method of the present invention is as follows: first, according to the power at the moment of 180W and the rotational speed of 600r/min, the wind speed vector is estimated by the SVR model as V =[8.92, 9.14, 8.94, 9.02] T , The estimated value of effective wind speed V m =9.0m/s. According to the best tip speed ratio method, the predicted value of the maximum power point speed ω is obtained ref =961r/min, directly adjust the working speed ω to 961r/min through the PI controller, and then start to observe P&O (adjustment period T=5s, step Δω=5r/min), because of ω ref It is already very close to the maximum power point MPP (ω opt =950r/min), after 3 consecutive beats of negative direction disturbance, it enters the steady state process, and the total time required is about 15 seconds. It can be seen that the tracking speed of the method proposed in the present invention is significantly higher than that of the conventional disturbance observation method.
[0109] After entering the steady-state process, the average power of the 60-second steady-state process is calculated respectively, and the conventional disturbance observation (P&O) method is 209 watts, while the method of the present invention is 222 watts, which shows that the method of the present invention can be effective To reduce power loss.
[0110] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited to this. Any person skilled in the art can infer changes within the core technology disclosed in the present invention. Or replacements should be covered within the protection scope of the present invention.
