CN106896724B - Tracking system and tracking method for sun tracker - Google Patents
Tracking system and tracking method for sun tracker Download PDFInfo
- Publication number
- CN106896724B CN106896724B CN201710228320.6A CN201710228320A CN106896724B CN 106896724 B CN106896724 B CN 106896724B CN 201710228320 A CN201710228320 A CN 201710228320A CN 106896724 B CN106896724 B CN 106896724B
- Authority
- CN
- China
- Prior art keywords
- network
- fitness
- optimal
- tracking
- population
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The invention provides a tracking system and a tracking method for a sun tracker, wherein an individual optimization is carried out according to sample data through a GA sample data preprocessing module, an individual with the maximum fitness is obtained and is transmitted to a BP network adjusting module to be used as an optimal initial value for training; on the basis of the optimal initial value, reversely transmitting through a BP network adjusting module, adjusting weights and thresholds of a hidden layer and an output layer to enable an obtained error evaluation function to be smaller than a set error threshold, and obtaining adjusted P, I, D parameters; the BP network adjusting module transmits the adjusted P, I, D parameter to the PID controller module, and the PID controller module controls according to the two degrees of freedom, so that the tracking performance and the anti-interference performance are respectively optimal. The invention combines the genetic algorithm and the neural network, fully utilizes the advantages of the genetic algorithm and the neural network, and ensures that the control system not only has the learning function, robustness and generalization capability of the neural network, but also has the global search optimization capability of the genetic algorithm.
Description
Technical Field
The invention relates to the technical field of sun trackers, in particular to a tracking system and a tracking method for a sun tracker.
Background
In order to accurately track the sun, a sun tracker of Solar-imaging flux (SOF) for Solar power generation and mobile vehicle-mounted measurement has been developed. A solar tracker for solar power generation is used for fixed-point monitoring, a double-shaft solar tracker is mostly adopted at present, the real-time azimuth angle and altitude angle of the sun can be calculated according to longitude and latitude information, and the output of the double-shaft solar tracker makes a control signal for adjusting the states of a pitching axis and an azimuth axis so as to realize solar tracking. Currently, a PSD is used as a feedback element, and the position of a main reflecting mirror is adjusted by using the position of a light spot on the PSD as a feedback signal, so that a spectrometer at the rear end obtains the maximum light intensity. And on the control algorithm, different adjusting step lengths are set according to the size of the PSD output signal.
Wherein x is the adjustment step length, a, b are constants, and a>b, e is error signal of actual position and central point of sun output by PSD, e0Is a set threshold constant. The current control algorithm has low tracking speed, poor accuracy and poor robustness, and when the vehicle bumps or has interference signals caused by other factors, the tracking effect is poor or even the tracking cannot be performed. In order to improve the performance of the mobile sun tracker, two-degree-of-freedom PID control of a BP network is optimized by adopting a GA algorithm.
Disclosure of Invention
The invention aims to solve the defects that a control algorithm of a sun tracker in the prior art is low in tracking speed, poor in precision and poor in robustness, and when a vehicle bumps or interference signals caused by other factors occur, the tracking effect is poor or even tracking cannot be performed.
In order to achieve the purpose, the technical scheme of the invention is as follows:
the tracking system for the sun tracker comprises a GA sample data preprocessing module, a BP network adjusting module and a PID controller module;
the GA sample data preprocessing module carries out individual optimization according to sample data to obtain an individual with the maximum fitness and transmits the individual to the BP network adjusting module to be used as an optimal initial value for training;
on the basis of the optimal initial value, reversely transmitting through a BP network adjusting module, adjusting weights and thresholds of a hidden layer and an output layer to enable an obtained error evaluation function to be smaller than a set error threshold, and obtaining adjusted P, I, D parameters;
the BP network adjusting module transmits the adjusted P, I, D parameter to the PID controller module, and the PID controller module controls the controlled object according to the two degrees of freedom, so that the tracking performance and the anti-interference performance of the PID controller can reach the best.
Preferably, the tracking system further comprises a feedback system module; the feedback system obtains the output signal of the controlled object and the input information of the PID controller, and the feedback system obtains an error signal according to the given input signal and the output signal of the controlled object and adjusts according to the error signal to form closed-loop control.
Preferably, the GA sample data preprocessing module includes a selection, crossover, and mutation operator operation unit; the selection, crossing and mutation operator operation units have a crossing rate pcAnd the rate of variation pmAnd carrying out self-adaptive adjustment to obtain the individual with the optimal fitness in the population, and directly entering the individual with the optimal fitness into the next generation of population, namely a new population, without any crossing and variation operation.
Preferably, the GA sample data preprocessing module further includes a catastrophe determination unit; and the catastrophe judgment unit calculates the fitness of the new population, and if catastrophe stops or reaches the set catastrophe times, GA algorithm preprocessing is completed, and the obtained optimal initial weight and threshold are transmitted to the BP network.
Preferably, the BP network adjusting module comprises a BP network training unit, the BP network training unit obtains the optimized weight and threshold as initial values, the initial weight and threshold are adjusted according to the momentum-adaptive learning rate, and an L evenberg-Marquardt algorithm optimization algorithm is adopted to find the optimal PID parameter value.
The invention also provides a tracking method applied to any one of the tracking systems for the sun tracker, which comprises the following specific steps of firstly carrying out PID parameter adjustment by taking the best tracking performance as a control target and then carrying out parameter adjustment by taking the best anti-interference performance as the control target:
1) sample data preprocessing
Firstly, carrying out individual optimization on sample data through a GA algorithm to obtain an individual with the maximum fitness and transmitting the individual to a BP network as an optimal initial value for training;
2) BP network training
On the basis of the optimal initial value, through reverse transmission of a BP network, the weights and thresholds of a hidden layer and an output layer are adjusted, so that the obtained error evaluation function is smaller than the set error threshold, and the adjusted P, I, D parameter is obtained;
3) PID control
The BP network adjusting module transmits the adjusted P, I, D parameter to the PID controller module, and the PID controller module controls the controlled object according to the two degrees of freedom, so that the tracking performance and the anti-interference performance of the PID controller can reach the best.
Preferably, the tracking method further comprises a signal feedback adjustment process; the method specifically comprises the following steps:
the feedback system obtains the output signal of the controlled object and the input information of the PID controller, and the feedback system obtains an error signal according to the input signal of the PID controller and the output signal of the controlled object and adjusts according to the error signal to form closed-loop control.
Preferably, in the step 1), the crossing rate p is specifically determined by selecting, crossing and mutating operator operation unitscAnd the rate of variation pmCarrying out self-adaptive adjustment to obtain individuals with optimal fitness in the population, and directly entering the individuals with optimal fitness into the next generation of population, namely a new population, without any crossing and variation operation; adaptive cross rate p usedcAnd changeSpecific index pmThe calculation formula of (2) is as follows:
wherein, α1，α2Two constants greater than 0, pc1,pc2,pm1,pm2The constants are empirically obtained and are 0.85, 0.65, 0.1 and 0.001 respectively; f. ofavg、fmaxAnd f' is the average fitness, the total fitness and the individual fitness of the population respectively.
Preferably, after the new population is obtained, performing catastrophe judgment on the fitness of the new population, if catastrophe stops or a set catastrophe frequency is reached, finishing preprocessing by the GA algorithm, and transmitting the obtained optimal initial weight and the threshold to the BP network.
Preferably, in step 2), the BP network training unit obtains the optimized weight and threshold as initial values, adjusts the initial weight and threshold according to the momentum-adaptive learning rate, and finds an optimal PID parameter value by using an L evenberg-Marquardt algorithm optimization algorithm, wherein the momentum-adaptive learning rate adjustment calculation formula is as follows:
β(k+1)＝τ*3λ*β(k)
wherein β is a learning rate factor, λ is a gradient direction, τ is a learning error coefficient, and w is a weight function;
the calculation formula of the L evenberg-Marquardt algorithm is:
Δw＝(JTJ+μI)-1JTe
wherein e is an error amount, J is a Jacobian matrix of the network error degree to the weight, I is a unit matrix, mu is a proportionality coefficient, and delta w is a weight increment.
Compared with the prior art, the invention has the following beneficial effects:
the two-degree-of-freedom ID respectively adopts the best tracking performance and the best anti-interference capability as control parameters, and PID parameters are respectively adjusted. The genetic algorithm and the neural network are combined, and the advantages of the genetic algorithm and the neural network are fully utilized, so that the control system has the learning function, robustness and generalization capability of the neural network and the global search optimization capability of the genetic algorithm.
In the patent, a GA algorithm is adopted to optimize a BP network to respectively complete the setting of PID parameters according to control parameters, so that the most accurate P, I, D parameters are obtained in the shortest time, more accurate control signals are input, and for a sun tracker, the real-time accurate tracking of the sun is realized.
By adopting a method of combining momentum items in heuristic learning rules with a self-adaptive learning rate and a digital optimization learning method, each improved method is complementary, the defects that the convergence speed is slow and the convergence speed falls into a minimum value in a BP network are effectively improved, and the control precision and speed in the later period are greatly improved.
The algorithm is used for controlling a closed-loop sun tracking system, and has the characteristics of high tracking precision, short adjustment time and strong anti-interference capability
Drawings
Fig. 1 is a block diagram of a module configuration in embodiment 1 of the present invention;
FIG. 2 is a schematic block diagram of a system according to embodiment 1 of the present invention;
fig. 3 is a 4:7:3BP network topology structure diagram in embodiment 2 of the present invention;
FIG. 4 is a flowchart of the algorithm of embodiment 2 of the present invention;
FIG. 5 is a functional block diagram of an algorithm according to embodiment 2 of the present invention;
fig. 6 is a control circuit diagram of the sun tracker in embodiment 3 of the present invention.
Detailed Description
So that the manner in which the above recited features of the present invention can be understood and readily understood, a more particular description of the invention, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings, wherein:
description of the concept: the sample in the following is a combination of all individuals, and the population is a combination of a certain number of individuals, and the number is determined by the set population size. And the individual is a neural network initial value or randomly generated training data.
Example 1
As shown in fig. 1 and 2, the tracking system for the sun tracker includes a GA sample data preprocessing module, a BP network adjusting module, a PID controller module, and a feedback system module.
The GA sample data preprocessing module carries out individual optimization according to the sample data to obtain an individual with the maximum fitness and transmits the individual to the BP network adjusting module to be used as an optimal initial value for training;
on the basis of the optimal initial value, reversely transmitting through a BP network adjusting module, adjusting weights and thresholds of a hidden layer and an output layer to enable an obtained error evaluation function to be smaller than a set error threshold, and obtaining adjusted P, I, D parameters;
the BP network adjusting module transmits the adjusted P, I, D parameter to the PID controller module, and the PID controller module controls according to the two degrees of freedom, so that the tracking performance and the anti-interference performance are respectively optimal.
The output signal of the controlled object is transmitted to the input end of the PID controller through the feedback system module, and the feedback system obtains an error signal according to the input signal of the PID controller and the output signal of the controlled object and adjusts according to the error signal to form closed-loop control.
The GA sample data preprocessing module comprises an iteration number and parameter initial value setting unit, a coding unit, a fitness calculating unit, a selection/intersection/mutation operator operating unit and a catastrophe judging unit; the BP network adjusting module comprises a network structure determining unit, an initialization network unit, an initial value obtaining unit, a BP network training unit and a precision judging unit.
Firstly, initializing iteration times n and crossing rate p of GA algorithm by setting iteration times and parameter initial value unitcThe rate of variation pmThe population size m; then go toThe network structure determining unit determines the topological structure of the BP network, and the total number of the network is three layers: an input layer, a hidden layer and an output layer; wherein the input layer has 4 neurons for respectively designating input signals rin(k) The actual output signal yout(k) Error amount e (k) and constant a; three neurons of the output layer respectively correspond to three parameters k controlled by PIDp,ki,kd(ii) a The number of the neurons of the hidden layer is based on an empirical formula
On the basis, through continuous simulation and debugging, the number of the hidden layer neurons is determined to be 7, namely the topological structure of the BP network is 4:7:3, as shown in FIG. 3. Initializing a weight value and a threshold initial value of a network unit to the BP network, an error evaluation function and an error threshold, and calculating whether an error signal is smaller than the error threshold under the condition of the initial value, if so, directly entering a test stage without training the BP network, otherwise, entering a training stage.
If the training stage is started, real number coding is carried out on a population containing a BP network initial value through a coding unit, a fitness function is defined as the difference between expected data and actual output data, then fitness calculation is carried out through a fitness calculation unit, and specifically GA operation comprises a selection operator, a cross operator and a mutation operator; wherein the crossing rate pcAnd the rate of variation pmObtaining an individual with the optimal fitness in the population by adopting a self-adaptive method, and then directly entering the individual with the optimal fitness into a next generation population, namely a new population without any crossing and variation operation, wherein the probability of other individuals entering the next population is in direct proportion to the relative fitness of the other individuals in the whole population; adaptive cross rate p usedcAnd the rate of variation pmThe calculation formula of (2) is as follows:
wherein, α1、α2Two constants greater than 0, pc1,pc2,pm1,pm2The constants are empirically obtained and are 0.85, 0.65, 0.1 and 0.001 respectively; f. ofavg、fmaxF' respectively calculating the average fitness, the total fitness and the fitness of the individual of the population, finally calculating the total fitness of the generated new population through a catastrophe judgment unit, if more excellent individuals do not appear in successive n generations, indicating that the GA algorithm has an early maturing thread and falls into a local minimum value, carrying out catastrophe at the moment, killing all excellent individuals of the current population, entering the next generation, if the fitness value is the same as that before catastrophe after a plurality of times of catastrophes, stopping the catastrophe, if the catastrophe stops or reaches the set catastrophe times, finishing the preprocessing of the GA algorithm, transmitting the obtained optimal initial weight and threshold to an initial value obtaining unit, and transmitting the obtained initial value to a BP network training unit by the BP network training unit, adjusting the initial weight and threshold according to the self-adaptive learning rate, and searching for the optimal PID parameter value by adopting a Mar L evenberg-quardt algorithm;
the momentum-adaptive learning rate adjustment calculation formula is as follows:
β(k+1)＝τ*3λ*β(k)
wherein β is a learning rate factor, λ is a gradient direction, τ is a learning error coefficient, and w is a weight function;
the calculation formula of the L evenberg-Marquardt algorithm is:
Δw＝(JTJ+μI)-1JTe
wherein e is an error amount, J is a Jacobian matrix of the network error degree to the weight, I is a unit matrix, mu is a proportionality coefficient, and delta w is a weight increment; when μ is large, it is the Gauss-Newton algorithm, and when μ is small, it approaches the gradient descent method. And the precision judging unit carries out final evaluation, when the error evaluation function of the BP network is smaller than the error threshold, the whole learning process of the BP network is finished, and then the learning effect and the generalization capability of the BP network are tested by using the test data.
Example 2
As shown in fig. 4 and 5, the tracking method for the sun tracker is applied to the tracking system provided in embodiment 1. The algorithm firstly takes the best tracking performance as a control target to carry out PID parameter adjustment, and then takes the best anti-interference performance as the control target to carry out parameter adjustment, and the specific adjustment steps are as follows:
step 1, randomly generating 2000 groups of data, wherein 1500 groups of data are used for training a BP network, and the other 500 groups of data are used for testing the BP network, and carrying out normalization processing on the data;
First, a total of three layers of the network are determined: an input layer, a hidden layer and an output layer; wherein the input layer has 4 neurons for respectively designating input signals rin(k) The actual output signal yout(k) Error amount e (k) and constant a; three neurons of the output layer respectively correspond to three parameters k controlled by PIDp,ki,kd(ii) a The number of the neurons of the hidden layer is based on an empirical formula
On the basis, through continuous simulation and debugging, the number of the hidden layer neurons is determined to be 7, namely the topological structure of the BP network is 4:7:3, as shown in FIG. 3.
Step 3, initializing BP network
Setting a weight value and a threshold initial value of the BP network, an error evaluation function and an error threshold, and calculating whether an error signal is smaller than the error threshold under the condition of the initial value, if so, directly entering a test stage without training the BP network, otherwise, entering a step 4); wherein the initial values are weight values and initial threshold values;
step 4, initializing evolution times n and crossing rate p of GA algorithmcThe rate of variation pmThe population size m;
step 5, carrying out real number coding on the population containing the initial value of the BP network, and defining a fitness function as the difference between expected data and actual output data;
step 6, executing GA operation
The GA operation comprises a selection operator, a crossover operator and a mutation operator; wherein the crossing rate pcAnd the rate of variation pmObtaining an individual with the optimal fitness in the population by adopting a self-adaptive method, and then directly entering the individual with the optimal fitness into a next generation population, namely a new population without any crossing and variation operation, wherein the probability of other individuals entering the next population is in direct proportion to the relative fitness of the other individuals in the whole population; adaptive cross rate p usedcAnd the rate of variation pmThe calculation formula of (2) is as follows:
wherein, α1，α2Two constants greater than 0, pc1,pc2,pm1,pm2The constants are empirically obtained and are 0.85, 0.65, 0.1 and 0.001 respectively; f. ofavg、fmaxF' is the average fitness, the total fitness and the individual fitness of the population respectively;
and 7, carrying out catastrophe judgment, calculating the total fitness of the generated new population, and if no more excellent individual appears in the continuous n generations, indicating that the GA algorithm has an early maturing thread and falls into a local minimum value, carrying out catastrophe at the moment, killing all excellent individuals of the current population, and entering the next generation. If the fitness value is the same as that before the catastrophe after a plurality of catastrophes, the catastrophe is stopped; if the catastrophe stops or the set catastrophe times are reached, the GA algorithm is optimized and completed, and the obtained optimal initial weight and the threshold are transmitted to the BP network;
step 8.BP network training
Adjusting the initial weight and the threshold according to the momentum-adaptive learning rate and searching for an optimal PID parameter value by adopting an L evenberg-Marquardt algorithm optimization algorithm, wherein the weight and the threshold after the received GA algorithm optimization are used as initial values;
the momentum-adaptive learning rate adjustment calculation formula is as follows:
β(k+1)＝τ*3λ*β(k)
wherein β is a learning rate factor, λ is a gradient direction, τ is a learning error coefficient, and w is a weight function;
the calculation formula of the L evenberg-Marquardt algorithm is:
Δw＝(JTJ+μI)-1JTe
wherein e is an error amount, J is a Jacobian matrix of the network error degree to the weight, I is a unit matrix, mu is a proportionality coefficient, and delta w is a weight increment; when μ is large, it is the Gauss-Newton algorithm, and when μ is small, it approaches the gradient descent method.
And 9, when the error evaluation function of the BP network is smaller than the error threshold, finishing the learning process of the whole BP network, and then testing the learning effect and generalization capability of the BP network by using the test data.
Example 3
This embodiment will describe the evolution process of the algorithm provided in the prior art to embodiment 2.
Two degrees of freedom
The control indexes of the PID control system mainly comprise: the external disturbance suppression effect and the target tracking characteristic show opposite change trends when one-degree-of-freedom control is adopted, and the optimal performance is achieved simultaneously. Two-degree-of-freedom PID (Two degree of freedom) control is to respectively adjust PID parameters with optimal target tracking characteristics and optimal external disturbance rejection characteristics, so that the performance of the whole control system is optimal.
When the 2DOF PID control is adopted, the requirements of understandability, simple structure, better combination with the traditional technology and capability of inheriting the technical result are required to be met. In the control loop for the sun tracker, a target filter type 2DOFPID control is employed. The control loop is shown in fig. 6.
The target value filter h(s) can be expressed as:
wherein α is a two-degree-of-freedom coefficient of proportional gain (generally 0. ltoreq. α. ltoreq.1), β is a two-degree-of-freedom coefficient of integral time (generally 0. ltoreq. β. ltoreq.1), γ is a two-degree-of-freedom coefficient of differential time (generally 0. ltoreq. γ. ltoreq.2), and 1/η is differential gain (generally 0.1. ltoreq. γ. ltoreq.1).
PID control of the solar tracker is carried out by adopting a method with variable two-degree-of-freedom coefficients, and the adjusting steps are as follows: adding external disturbance and adjusting Kp,Ki,KdThe method comprises the steps of obtaining α gamma values preliminarily according to a Chian-Hrone-Reswick (CHR) adjusting method, taking β as 0.15 according to engineering experience, and finely adjusting two-degree-of-freedom coefficients α and gamma near a set value according to the variation of a target value to enable the target tracking characteristic to be optimal.
The two degrees of freedom can improve the control characteristic and is easy to adjust, and the control performance can be optimal under the condition of ensuring simple structure only by adding a target value filter.
BP network
As the most widely applied artificial neural network model, the error Back Propagation Neural Network (BPNN) has the capability of realizing any complex nonlinear mapping and can approach any nonlinear continuous function with any precision; a parallel distributed processing mode; the self-learning capability is provided; the method has certain popularization, generalization and self-adaption capabilities, namely the capability of applying the set of weights to general situations; the capability of data fusion can process quantitative and qualitative information simultaneously; the method can be used for multivariable systems and online learning.
The basic idea of the BP algorithm is a least square method, which adopts a gradient search technology to minimize the mean square error value of the actual output value and the expected output value of the network. The learning process of the algorithm consists of forward propagation of information and backward propagation of errors. The input information is transmitted to the output layer after being processed layer by layer from the input layer through the hidden layer, and the state of each layer of neuron nodes only influences the state of the next layer of neurons. When the output layer can not obtain the expected output, the reverse propagation is carried out, the error signal is returned along the original connecting path, the connecting weight and the closed value of each layer of neuron are modified, the error function is reduced along the negative gradient direction, and finally the error between the actual output value and the expected output value is minimized.
The BP neural network has an input layer, a hidden layer and an output layer, each layer has different numbers of neurons. The topological structure of the neural network is different according to the structure and complexity of the research object. In the system, a topological structure comprising an input layer, a hidden layer and an output layer and the number of neurons being 3:5:3 is adopted.
The three input signals are respectively designated input signals rin(k) The actual output signal yout(k) Error amount e (k). Three output signals respectively correspond to three parameters k of PID controlp,ki,kd。
The learning process of the BP network consists of two parts of forward propagation and backward propagation. Firstly, the forward propagation process of signals is carried out, information of an input layer is propagated to nodes of a hidden layer forwards, the information of the hidden layer is transmitted to the nodes of an output layer through the activation function operation of each unit, and the information of the hidden layer is output through the activation function operation of each node on the output layer. In the forward propagation process, the neuron state of each layer only affects the next layer of neuron network. If the error between the actual output and the expected output value is larger than the set error function threshold value, the method turns to a back propagation process, the error signals are propagated in the back direction, the weight values of all layers of neurons are modified successively, the corrected output is obtained through forward propagation, and the two processes are repeatedly applied to minimize the error signals. When the actual error signal is less than the set error threshold, the learning process of the network ends.
In the forward propagation process, the input and output of the hidden layer node are:
the input and the output of the output layer are as follows:
wij,wjkthe connection weight between the input layer and the hidden layer, and between the hidden layer and the output layer.
The output values of the hidden layer and the output layer, respectively. f, (k), g (k) are activation functions which are used for adding nonlinear factors so as to make up for the defect of insufficient expression force of a linear model, the selection of the activation functions needs to meet the characteristics of monotonous increasing, bounded and first-order differentiable, and after simulation, the bipolar sigmoid function of the activation function of the hidden layer is shown as follows:
the error evaluation function is expressed as:
and adjusting the weight of each node through a back propagation process according to the error evaluation function, wherein the correction quantity of the weight is in direct proportion to the negative gradient direction of the error, namely:
order to
Then
The weight adjustment quantity between the hidden layer and the input layer is as follows:
order to
Has a delta wij＝ηjxi
η is the learning step length, i.e. the learning rate, the weight between any node of the hidden layer and the output layer at the next iteration, and the weights between the input layer and the hidden layer are respectively:
wjk(k)＝Δwjk+wjk(k)
wij(k)＝Δwij+wij(k)
however, the standard BP neural network is actually a gradient descent search method, and has the disadvantages of slow convergence rate and easy falling into local minima, and the disadvantages can be corrected by using an improved BP algorithm, which can be divided into two categories, namely, a heuristic learning rule is adopted, such as adding an additional momentum term, adopting an adaptive learning rate and the like, a learning method based on numerical optimization, such as a conjugate gradient method, a quasi-newton method and an L evenberg-Marquardt method (L-M method for short),
1. adding momentum items
In order to accelerate the convergence speed, the last correction weight coefficient is added to the correction quantity of the weight value, and the weight value is taken as one of the bases of the current correction, namely:
Δwij(k+1)＝ηjxi+αΔwij(k)
wherein α is the momentum factor.
2. Momentum-adaptive learning rate adjustment algorithm
When adaptive adjustment of the learning rate is performed, the basic idea is to increase η to shorten the learning time when learning converges, and to decrease η in time until convergence when η is too large to converge.
β(k+1)＝τ*3λ*β(k)
Wherein β is a learning rate factor, λ is a gradient direction, τ is a learning error coefficient, and w is a weight function.
3. Adopting L evenberg-Marquardt algorithm
L-M is the combination of gradient descent method and Newton method in nature, and has very high convergence rate when the network weight is less L-M optimization algorithm has the weight adjustment rate:
Δw＝(JTJ+μI)-1JTe
wherein e is error amount, J is Jacobian matrix of network error degree to weight, I is unit matrix, mu is proportionality coefficient, and Δ w is weight increment. When μ is large, it is the Gauss-Newton algorithm, and when μ is small, it approaches the gradient descent method.
The convergence rate is improved by adding the momentum term, but the selection of the learning rate is difficult, and the requirement on the initial value is high; the self-adaptive learning rate greatly improves the defect of easy falling into local minimum values, but the convergence speed is slow.
L evenberg-Marquardt algorithm greatly improves the convergence rate, but the problem of easy falling into local minimum value cannot be improved, a method of combining momentum items in heuristic learning rules, adaptive learning rate and a digital optimization learning method is adopted, each improved method is complementary, the defects that the convergence rate is slow and the fall into the minimum value exists in a BP network are effectively improved, and the control precision and speed in the later period are greatly improved.
The improved BP algorithm reduces the probability of entering local minimum value, but does not fundamentally realize rapid global search, and the GA algorithm GA is a search technology simulating a natural biological evolution mechanism and has global search capability, so that the network weight can be better adjusted by adopting a mode of combining the GA algorithm and the BP network, and rapid and high-precision PID control is realized.
And (3) GA algorithm:
the main idea of the GA-BPNN algorithm is as follows: firstly, the GA algorithm is used for searching the vicinity of the optimal solution quickly, a better search space is positioned in the solution space, a better initial value is provided for the BP network, and then the BP algorithm is used for searching the optimal solution in the small search space. The switching between GA algorithm and BP algorithm can be done by the size of the error, with the GA algorithm being used if the error is larger than a certain value and the BP algorithm being used when smaller than this value until a limited accuracy or maximum number of steps is reached.
The simple GA algorithm uses a selection operator, a crossover operator and a mutation operator, and the GA process is as follows:
1. determining a coding mode: the conversion from the solution space to the search space is realized, and the coding form is as follows:
wherein the content of the first and second substances,
the connection weight between the input layer and the hidden layer, and between the hidden layer and the output layer.
2. Determining parameters: group size, crossover, mutation probability, and number of termination iterations;
3. initialization: randomly generating n individuals, and providing an initial population P (0);
4. evaluation: decoding to a solution space, calculating the candidate solution, the fitness of the solution set and the average fitness.
GA operation: comprises a selection operator, a crossover operator and a mutation operator, and operates the population P (t) to generate the next generation
P(t+1)。
6. And repeating the steps 4 and 5 until the parameters converge or reach the preset index.
The simple GA algorithm is prone to premature, i.e. premature, convergence of the solution intended for local optimization, and therefore adaptive adjustment of the crossover rate and the variance rate is required. Improved adaptive crossover rate pcAnd the rate of variation pmThe calculation formula of (2) is as follows:
α1，α2two constants greater than 0, pc1,pc2,pm1,pm2The constants are empirically found to be 0.9, 0.6, 0.1, 0.001, respectively.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are merely illustrative of the principles of the invention, but that various changes and modifications may be made without departing from the spirit and scope of the invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (9)
1. A tracking system for a sun tracker, characterized by: the system comprises a GA sample data preprocessing module, a BP network adjusting module and a PID controller module;
the GA sample data preprocessing module carries out individual optimization according to sample data to obtain an individual with the maximum fitness and transmits the individual to the BP network adjusting module to be used as an optimal initial value for training;
specifically, the crossing rate p is determined by selecting, crossing and mutating operator operation unitscAnd the rate of variation pmCarrying out self-adaptive adjustment to obtain individuals with optimal fitness in the population, and directly entering the individuals with optimal fitness into the next generation of population, namely a new population, without any crossing and variation operation; adaptive cross rate p usedcAnd the rate of variation pmThe calculation formula of (2) is as follows:
wherein, α1，α2Two constants greater than 0, pc1,pc2,pm1,pm2The constants are empirically obtained and are 0.85, 0.65, 0.1 and 0.001 respectively; f. ofavg、fmaxF' is the average fitness, the total fitness and the individual fitness of the population respectively;
on the basis of the optimal initial value, reversely transmitting through a BP network adjusting module, adjusting weights and thresholds of a hidden layer and an output layer to enable an obtained error evaluation function to be smaller than a set error threshold, and obtaining adjusted P, I, D parameters;
the BP network adjusting module transmits the adjusted P, I, D parameter to the PID controller module, and the PID controller module controls the controlled object according to the two degrees of freedom, so that the tracking performance and the anti-interference performance of the PID controller can reach the best.
2. The tracking system for a sun tracker according to claim 1, wherein: the tracking system further comprises a feedback system module; and the feedback system module acquires an output signal of the controlled object and input information of the PID controller, obtains an error signal and adjusts according to the error signal to form closed-loop control.
3. The tracking system for a sun tracker according to claim 1, wherein: the GA sample data preprocessing module comprises a selection operator operation unit, a crossover operator operation unit and a mutation operator operation unit; the selection, crossing and mutation operator operation units have a crossing rate pcAnd the rate of variation pmAnd carrying out self-adaptive adjustment to obtain the individual with the optimal fitness in the population, and directly entering the individual with the optimal fitness into the next generation of population, namely a new population, without any crossing and variation operation.
4. The tracking system for a sun tracker according to claim 3, wherein: the GA sample data preprocessing module also comprises a catastrophe judging unit; and the catastrophe judgment unit calculates the fitness of the new population, and if catastrophe stops or reaches the set catastrophe times, GA algorithm preprocessing is completed, and the obtained optimal initial weight and threshold are transmitted to the BP network.
5. The tracking system for the solar tracker according to claim 4, wherein the BP network adjusting module comprises a BP network training unit, the BP network training unit obtains the optimized weight and threshold as initial values, adjusts the initial weight and threshold according to the momentum-adaptive learning rate, and finds the optimal PID parameter value by adopting L evenberg-Marquardt algorithm optimization algorithm.
6. A tracking method applied to the tracking system for a sun tracker according to any one of claims 1 to 5, characterized in that: the method comprises the following steps of firstly carrying out PID parameter adjustment by taking the best tracking performance as a control target, and then carrying out parameter adjustment by taking the best anti-interference performance as the control target, wherein the specific adjustment steps are as follows:
1) sample data preprocessing
Firstly, carrying out individual optimization on sample data through a GA algorithm to obtain an individual with the maximum fitness and transmitting the individual to a BP network as an optimal initial value for training;
in the step 1), the crossing rate p is specifically determined by selecting, crossing and mutating operator operation unitscAnd the rate of variation pmCarrying out self-adaptive adjustment to obtain individuals with optimal fitness in the population, and directly entering the individuals with optimal fitness into the next generation of population, namely a new population, without any crossing and variation operation; adaptive cross rate p usedcAnd the rate of variation pmThe calculation formula of (2) is as follows:
wherein, α1，α2Two constants greater than 0, pc1,pc2,pm1,pm2The constants are empirically obtained and are 0.85, 0.65, 0.1 and 0.001 respectively; f. ofavg、fmaxF' is the average fitness, the total fitness and the individual fitness of the population respectively;
2) BP network training
On the basis of the optimal initial value, through reverse transmission of a BP network, the weights and thresholds of a hidden layer and an output layer are adjusted, so that the obtained error evaluation function is smaller than the set error threshold, and the adjusted P, I, D parameter is obtained;
3) PID control
The BP network adjusting module transmits the adjusted P, I, D parameter to the PID controller module, and the PID controller module controls the controlled object according to the two degrees of freedom, so that the tracking performance and the anti-interference performance of the PID controller can reach the best.
7. The tracking method for a tracking system of a sun tracker according to claim 6, characterized in that: the tracking method further comprises a signal feedback adjustment process; the method specifically comprises the following steps:
the feedback system obtains the output signal of the controlled object and the input information of the PID controller, and the feedback system obtains an error signal according to the given input signal and the output signal of the controlled object and adjusts according to the error signal to form closed-loop control.
8. The tracking method for a tracking system of a sun tracker according to claim 6, characterized in that: and after the new population is obtained, carrying out catastrophe judgment on the fitness of the new population, finishing the preprocessing of the GA algorithm if catastrophe stops or reaches a set catastrophe frequency, and transmitting the obtained optimal initial weight and threshold to the BP network.
9. The tracking method for the tracking system of the solar tracker according to claim 8, wherein in the step 2), the BP network training unit obtains the optimized weight and threshold as initial values, adjusts the initial weight and threshold according to the momentum-adaptive learning rate, and finds the optimal PID parameter value by adopting L evenberg-Marquardt algorithm optimization algorithm, wherein the momentum-adaptive learning rate adjustment calculation formula is as follows:
β(k+1)＝τ*3λ*β(k)
wherein β is a learning rate factor, λ is a gradient direction, τ is a learning error coefficient, and w is a weight function;
the calculation formula of the L evenberg-Marquardt algorithm is:
Δw＝(JTJ+μI)-1JTe
wherein e is an error amount, J is a Jacobian matrix of the network error degree to the weight, I is a unit matrix, mu is a proportionality coefficient, and delta w is a weight increment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710228320.6A CN106896724B (en) | 2017-04-10 | 2017-04-10 | Tracking system and tracking method for sun tracker |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710228320.6A CN106896724B (en) | 2017-04-10 | 2017-04-10 | Tracking system and tracking method for sun tracker |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106896724A CN106896724A (en) | 2017-06-27 |
| CN106896724B true CN106896724B (en) | 2020-08-04 |
Family
ID=59196419
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710228320.6A Active CN106896724B (en) | 2017-04-10 | 2017-04-10 | Tracking system and tracking method for sun tracker |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106896724B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110518625B (en) * | 2019-01-22 | 2023-07-25 | 电子科技大学 | Grid-connected inverter direct-current component suppression method with variable learning rate BP-PID control |
| CN110068759B (en) * | 2019-05-22 | 2021-11-09 | 华雁智能科技(集团)股份有限公司 | Fault type obtaining method and device |
| CN112666830A (en) * | 2020-12-24 | 2021-04-16 | 西南技术物理研究所 | Tracking precision control method based on RBF neural network |
| CN116449685B (en) * | 2023-04-23 | 2024-03-29 | 上海交通大学 | Ship combustion engine control method and device capable of achieving rapid tracking and accurate control |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096373A (en) * | 2010-12-07 | 2011-06-15 | 昆明理工大学 | Microwave drying PID (proportion integration differentiation) control method based on increment improved BP (back propagation) neural network |
| CN104835215A (en) * | 2015-05-11 | 2015-08-12 | 上海航盛实业有限公司 | Safety monitoring system of automobile data recorder |
| CN105303252A (en) * | 2015-10-12 | 2016-02-03 | 国家计算机网络与信息安全管理中心 | Multi-stage nerve network model training method based on genetic algorithm |
| CN106503802A (en) * | 2016-10-20 | 2017-03-15 | 上海电机学院 | A kind of method of utilization genetic algorithm optimization BP neural network system |
-
2017
- 2017-04-10 CN CN201710228320.6A patent/CN106896724B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096373A (en) * | 2010-12-07 | 2011-06-15 | 昆明理工大学 | Microwave drying PID (proportion integration differentiation) control method based on increment improved BP (back propagation) neural network |
| CN104835215A (en) * | 2015-05-11 | 2015-08-12 | 上海航盛实业有限公司 | Safety monitoring system of automobile data recorder |
| CN105303252A (en) * | 2015-10-12 | 2016-02-03 | 国家计算机网络与信息安全管理中心 | Multi-stage nerve network model training method based on genetic algorithm |
| CN106503802A (en) * | 2016-10-20 | 2017-03-15 | 上海电机学院 | A kind of method of utilization genetic algorithm optimization BP neural network system |
Non-Patent Citations (4)
| Title |
|---|
| 基于自适应学习速率的改进型BP算法研究;杨甲沛;《中国优秀硕士学位论文全文数据库》;20090915(第9期) * |
| 基于遗传神经网络的光伏阵列多峰最大功率点追踪;白静晶;《中国优秀硕士学位论文全文数据库》;20120415(第4期) * |
| 基于遗传算法-BP神经网络的PID控制器的设计;曲尔光;《机床与液压》;20070731;第35卷(第7期) * |
| 白静晶.基于遗传神经网络的光伏阵列多峰最大功率点追踪.《中国优秀硕士学位论文全文数据库》.2012,(第4期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106896724A (en) | 2017-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106896724B (en) | Tracking system and tracking method for sun tracker | |
| CN111351488B (en) | Intelligent trajectory reconstruction reentry guidance method for aircraft | |
| CN109117951B (en) | BP neural network-based probability load flow online calculation method | |
| CN107272403A (en) | A kind of PID controller parameter setting algorithm based on improvement particle cluster algorithm | |
| CN110308649B (en) | PID parameter optimization method based on PSO-SOA fusion algorithm and applied to industrial process control | |
| CN108284442B (en) | Mechanical arm flexible joint control method based on fuzzy neural network | |
| CN110488611B (en) | Bionic robot fish motion control method, controller and bionic robot fish | |
| CN110046710A (en) | A kind of the nonlinear function Extremal optimization method and system of neural network | |
| CN102129259B (en) | Neural network proportion integration (PI)-based intelligent temperature control system and method for sand dust environment test wind tunnel | |
| CN111047085A (en) | Hybrid vehicle working condition prediction method based on meta-learning | |
| CN114199248B (en) | AUV co-location method for optimizing ANFIS based on mixed element heuristic algorithm | |
| CN109472105A (en) | Semiconductor product yield Upper bound analysis method | |
| CN111445498A (en) | Target tracking method adopting Bi-L STM neural network | |
| CN113722980A (en) | Ocean wave height prediction method, system, computer equipment, storage medium and terminal | |
| CN116151102A (en) | Intelligent determination method for space target ultra-short arc initial orbit | |
| CN107273971B (en) | Feed-forward neural network structure self-organization method based on neuron significance | |
| Fang et al. | Two-stream fused fuzzy deep neural network for multiagent learning | |
| CN105976029A (en) | Cuckoo-behavior-RNA-GA-based bridge crane neural network modeling method | |
| Chen et al. | Optimization of ADRC Parameters Based on Particle Swarm Optimization Algorithm | |
| CN114779661B (en) | Chemical synthesis robot system based on multi-classification generation confrontation imitation learning algorithm | |
| CN115453880A (en) | Training method of generative model for state prediction based on antagonistic neural network | |
| CN111553562A (en) | Ship collision risk degree estimation method based on improved BP neural network | |
| Li et al. | Improved PID Controller Based on BP Neural Network | |
| CN113065693B (en) | Traffic flow prediction method based on radial basis function neural network | |
| Lin et al. | Algorithm of ship collision risk based on improved BP neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |