HHOPF algorithm based on Harlisia eagle particle filtering
Through the iterative optimization and sine cosine search mechanism of the Harris Eagle particle filtering algorithm, the problems of particle degradation and insufficient diversity in particle filtering are solved, the accuracy and diversity of particle filtering are improved, and the performance of the algorithm is significantly improved.
Patent Information
- Application Number
- CN202411328447.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-24
- Publication Date
- 2024-11-08
AI Technical Summary
Particle filtering algorithms have problems of particle degradation and depletion in nonlinear and non-Gaussian systems, and the calculation amount is large, making it difficult to improve particle accuracy and diversity.
The HHOPF algorithm based on Harris Eagle particle filtering is adopted, and the particle distribution is improved through the Harris Eagle algorithm iteratively seeking optimization, and the sine cosine search mechanism and Brownian motion are introduced to optimize the importance sampling process of particle filtering, and improve the convergence speed and development capabilities.
It improves the accuracy and diversity of particles, reduces the root mean square error, enhances the algorithm development capability, and significantly improves the performance and accuracy of particle filtering.
Smart Images

Figure CN118917350A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of particle filter algorithms and relates to a HHOPF algorithm based on Harris Hawk particle filter. Background Art
[0002] Random dynamic systems are a common type of system in industry. The most basic problem is state estimation, and the important cornerstone of the entire estimation theory is the filtering algorithm. Filtering is a technology that can filter out impurity signals from noise-interfered signals and deduce the estimated target state. As early as 1996, Kalman proposed the Kalman filter algorithm specifically for the estimation problem of linear systems with Gaussian distribution. The algorithm is a time domain estimation method that uses a natural regression recursive method to obtain the minimum mean square error estimate of the system state, which is very suitable for processing linear systems with Gaussian distribution.
[0003] However, real life contains far more than simple Gaussian distributed linear systems, but also complex systems with nonlinear and non-Gaussian characteristics. It was not until the advent of the particle filter algorithm that the particle filter algorithm has completely established its unshakable and solid position in nonlinear non-Gaussian systems.
[0004] Particle filtering is an excellent filtering idea that combines Monte Carlo random sampling and optimal Bayesian filtering. This algorithm is not limited by linearization error and Gaussian noise and is applicable to any nonlinear, non-Gaussian system represented by a state space model. However, it also has some defects such as particle degradation and exhaustion problems and relatively large amount of calculation. Therefore, how to improve the shortcomings of particle filtering and make it more practical and valuable is the current research focus. Summary of the invention
[0005] The problem solved by the present invention is to provide a HHOPF algorithm based on Harris Hawk particle filtering, which improves the accuracy of particles, improves the impoverishment of particles, and increases the diversity of particles.
[0006] The present invention is achieved through the following technical solutions:
[0007] A HHOPF algorithm based on Harris Hawk particle filter includes the following operations:
[0008] 1) Set the number of particles N, the maximum number of iterations Max of Harris Eagle, the initialization of the state equation and the observation equation;
[0009] 2) According to formulas (1) and (2), the state value and observation value of the particle are calculated, and the state value of the particle is used as the initial position of each Harris hawk individual;
[0010] x k =f k (xk-1 ,v k-1 ) (1)
[0011] Z k =h k (x k ,n k ) (2)
[0012] f k (is the relationship between the system state at the previous moment and the current moment, h k is the relationship between the current state and its observed value, x k represents the state of the system k at the current time, x k-1 , represents the state of the system k at the previous moment, Z k represents the observed value of system k at the current moment, v k-1 and n k All noises are Gaussian distributed.
[0013] 3) Using the Harris Eagle algorithm to iteratively optimize and improve the particle distribution, and updating the particle exploration behavior according to the search mechanism of sine and cosine as shown in formula (17) and the Brownian motion as shown in formula (18) to obtain the particle distribution;
[0014]
[0015] Among them, X(t) and X(t+1) are the positions of the particle in the current and next iteration respectively, t is the number of iterations, and X rand (t) is the position of a randomly selected particle, X rabbit (t) is the position of the particle with the best fitness, r1, r2, r6, q are all random numbers between [0, 1], r4 is a random number between [0, 2π], r5 is a random number between [0, 2], and T is the maximum number of iterations;
[0016] 4) Calculate the weights of particles based on the particle filter algorithm;
[0017] 5) Normalize the particle weights based on the particle filter algorithm;
[0018] 6) Calculate the output state of each system based on the particle filter algorithm.
[0019] Furthermore, when the Harris Eagle algorithm is used to iteratively optimize and improve the particle distribution, the progressive rapid dive soft encirclement strategy is used to update the position:
[0020]
[0021] Y(t)=X rabbit (t)-E|J·X rabbit(t)-X(t)| (14)
[0022] Z(t)=Y(t)+S+BM(D) (19)
[0023] Where f(·) is the fitness function, E is the escape energy of the prey, J is a random number between [0, 2], S is a D-dimensional random vector whose elements are random numbers between [0, 1], and BM() is the introduction of Brownian motion in the Harris hawk optimization process, using Brownian motion to replace Levy flight.
[0024] Furthermore, when the Harris Eagle algorithm is used to iteratively optimize and improve the particle distribution, the progressive fast dive hard encirclement strategy is used to update the position:
[0025]
[0026] Y(t)=X rabbit (t)-E|J·X rabbit (t)-X m (t)| (16)
[0027] Z(t)=Y(t)+S+BM(D) (19)
[0028] Where f(·) is the fitness function, E is the escape energy of the prey, J is a random number between [0, 2], and X m (t) is the average position of the individual, S is a D-dimensional random vector whose elements are random numbers between [0, 1], and BM() introduces Brownian motion in the Harris hawk optimization process, using Brownian motion to replace Levy flight.
[0029] Compared with the prior art, the present invention has the following beneficial technical effects:
[0030] The present invention optimizes the importance sampling process of particle filtering through the Harris Hawk algorithm, takes the prior state of particles in the particle filter as the individual position of the initial population of the Harris Hawk, and uses the iterative optimization process to improve the particle distribution; introduces the sine-cosine search mechanism into the Harris Hawk algorithm to increase the convergence speed; in addition, uses Brownian motion to optimize the Harris Hawk optimization process, thereby improving the development capability of the algorithm.
[0031] In the particle iteration process, compared with the PF algorithm and the WOAPF algorithm, the particles of the HHOPF algorithm proposed in the present invention move to the high likelihood region as a whole, and reasonably retain some particles in the low likelihood region, ensuring the particle diversity in the entire filtering process; therefore, HHOPF not only improves the accuracy of particles, but also improves the impoverishment of particles and increases the diversity of particles. The experimental results also further verify that the algorithm proposed in the present invention has significant advantages in coping with the challenges in particle filtering.
[0032] When the number of particles is the same, the prediction curve of the HHOPF algorithm provided by the present invention is significantly better than that of the PF algorithm and the WOAPF algorithm in terms of consistency with the actual state, and the error of the HHOPF algorithm is smaller. Compared with the WOAPF algorithm, although the calculation time of the HHOPF algorithm is slightly longer than that of the WOAPF algorithm, the root mean square error is significantly reduced, which fully demonstrates that the accuracy of the particle filter is improved by increasing the calculation time.
[0033] The HHOPF algorithm provided by the present invention can improve the positioning accuracy of the robot, thereby making mapping more accurate; in robot path planning, it can improve the robot's optimization ability, thereby making navigation more accurate; it can also be used in lithium battery SOC estimation to reduce the estimation error of the lithium battery, thereby improving the stability of the lithium battery, and can improve the remaining life and generalization ability of the ion battery. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a schematic diagram of the process of the present invention;
[0035] Figure 2a-2b Comparative analysis of the results of different algorithms when N=30, where Figure 2a is the particle state estimate; Figure 2b Calculate the error value;
[0036] Figure 3a-Figure 3b Comparative analysis of the results of different algorithms when N=50, where Figure 3a is the particle state estimate, Figure 3b Calculate the error value;
[0037] Figure 4a-4b Comparative analysis of the results of different algorithms when N = 100, where Figure 4a is the particle state estimate; Figure 4b Calculate the error value;
[0038] Figure 5 The particle state distribution when k=30;
[0039] Figure 6 Particle state distribution when k=50. DETAILED DESCRIPTION
[0040] The present invention is further described in detail below in conjunction with the embodiments, which are intended to explain the present invention rather than to limit it.
[0041] The particle filter algorithm is a state estimation method for nonlinear and non-Gaussian systems. However, in the later stages of the algorithm, particle weight degradation and impoverishment problems will occur, affecting the filtering effect. The reason for these problems is that the prior probability density function used in the particle filter algorithm is used as a suggested distribution function, which will lead to a large deviation between the estimated value and the true value; at the same time, due to the influence of the resampling operation, the particle weights are uneven, causing the weights of some particles to become very small; and too many resampling operations will also make the particle state single, the type poor, and the loss of diversity. Therefore, the importance sampling and resampling mechanism of the particle filter algorithm is an important link in causing the degradation and impoverishment of particle importance weights.
[0042] The HHOPF algorithm based on Harris Hawk particle filter proposed in the present invention optimizes the importance sampling process of particle filter through Harris Hawk algorithm, takes the particle prior state in particle filter as the individual position of Harris Hawk initial population, and improves the particle distribution through iterative optimization process. The sine-cosine search mechanism is introduced into Harris Hawk algorithm to improve the convergence speed. In addition, Brownian motion is used to optimize the Harris Hawk optimization process, thereby improving the development capability of the algorithm.
[0043] 1. Particle filter algorithm
[0044] The particle filter algorithm uses the Monte Carlo method to represent probability with the help of particle sets, and can be applied to any form of state space model. The particle filter algorithm approximates its posterior probability distribution by randomly sampling particles. In order to better describe the state change of the system, it is necessary to consider the relationship between the current state and the previous state, and the relationship between the state and the observed value. Therefore, the relationship between the system state at the previous moment and the current moment is shown in Formula 1:
[0045] x k =f k (x k-1 ,v k-1 ) (1)
[0046] The relationship between the current state and its observed value is shown in Formula 2:
[0047] Z k =h k (x k ,n k ) (2)
[0048] Among them, f k (is the relationship between the system state at the previous moment and the current moment, h k is the relationship between the current state and its observed value, x k represents the state of the system k at the current time, x k-1 , represents the state of the system k at the previous moment, Z krepresents the observed value of system k at the current moment, v k-1 and n k All noises obey Gaussian distribution.
[0049] The particle state update equation can be determined by the observation information and particle state information at time k, and the posterior probability density function is as follows:
[0050]
[0051] Assume that the importance function of easy sampling is q(x 0:k |Z 1:k ), it can be expressed by the following formula:
[0052]
[0053] Since it is very difficult to obtain samples directly based on the posterior probability, the importance distribution function can simplify the difficulty of representing the posterior distribution. However, the key to the importance function is to determine the weight of the particle state based on the known information.
[0054]
[0055] Among them, δ is the Dirichlet function; and the system state weight calculation is as follows:
[0056]
[0057] Normalize the weights, that is:
[0058]
[0059] Finally, with the help of the easy-to-sample importance function, the state of the system at time k is:
[0060]
[0061] 2. Harris Hawk Algorithm
[0062] The Harris Hawk Algorithm (HHO Algorithm) simulates the group cooperative behavior of Harris Hawks in nature, especially the sudden attack behavior when lost. The HHO foraging stage can be divided into multiple efficient stages such as exploration behavior, search and development conversion, and development stage, covering different behaviors such as tracking, siege and attack, and capturing food in a collaborative way.
[0063] 2.1 Exploring Behavior Update Strategy
[0064] Harris's Hawks roost randomly in certain places and find their prey using 2 strategies:
[0065]
[0066] Among them, X(t) and X(t+1) are the positions of the individual in the current and next iterations, t is the number of iterations, and X rand (t) is the randomly selected individual position, X rabbit (t) is the prey position, i.e. the position of the individual with the best fitness, r1, r2, r3, r4, q are all random numbers between [0, 1]. lb and ub are the lower and upper limits of the variable, X m (t) is the average position of individuals.
[0067] 2.2 Energy conversion mechanism between search behavior and development behavior
[0068] According to the escape energy of the prey, an ingenious energy conversion mechanism is implemented between the search and various development behaviors. If the escape energy is greater than 1, the search phase is entered; otherwise, the development phase is entered. The escape energy E of the prey in the HHO algorithm is defined as:
[0069]
[0070] Where E0 is a random number between [-1, 1], t is the number of iterations, and T is the maximum number of iterations. When |E|≥1, it enters the search phase, and when |E|<1, it enters the development phase.
[0071] 2.3 Strategy selection method in the development stage
[0072] The update of the predation strategy in the development phase is mainly based on the escape energy of the prey and the strategy selection parameter r. r is a random number between [0, 1], which is used to select among different development strategies. When the escape energy is greater than 0.5, the soft siege strategy is adopted, and when the escape energy is less than 0.5, the hard siege strategy is adopted. r determines whether to adopt the "asymptotic fast dive" strategy. If r is less than 0.5, the above dive strategy is adopted.
[0073] (1) When 0.5≤|E|<1 and r≥0.5, the algorithm will adopt the "soft siege" strategy to update the position, as shown in the following formula:
[0074] X(t+1)=ΔX(t)-E|JX rabbit (t)-X(t)| (11)
[0075] Where ΔX(t)=|X rabbit (t)-X(t)| represents the difference between the prey position and the individual's current position, and J is a random number between [0, 2].
[0076] (2) When |E| < 0.5 and r ≥ 0.5, the “hard siege” strategy is adopted to update the position:
[0077] X(t+1)=X rabbit(t)-E|ΔX(t)| (12)
[0078] (3) When 0.5≤|E|<1 and r<0.5, adopt the “asymptotic fast dive” soft encirclement strategy to update the position:
[0079]
[0080] Y(t)=X rabbit (t)-E|J·X rabbit (t)-X(t)| (14)
[0081] Z(t)=Y(t)+S·LF(2) (15)
[0082] Where f(·) is the fitness function, S is a D-dimensional random vector whose elements are random numbers between [0, 1], LF(2) is the mathematical expression of the Levy flight, and J is a random number between [0, 2].
[0083] (4) When |E| < 0.5 and r < 0.5, the “asymptotic fast dive” hard encirclement strategy is adopted to update the position. The main difference is the state update of formula 14:
[0084] Y(t)=X rabbit (t)-E|J·X rabbit (t)-X m (t)| (16)
[0085] Among them, X m (t) is the average position of individuals.
[0086] 3. Harris Hawk Particle Filter Optimization (HHOPF) Algorithm
[0087] 3.1 Particle importance evaluation based on fitness function
[0088] The larger the particle weight in the particle filter algorithm, the closer it is to the real particle, while the larger the fitness value in the Harris Hawk algorithm, the closer it is to the target prey. Therefore, the fitness fi in the Harris Hawk algorithm represents the particle importance value wi, that is, fi = wi). Through continuous iteration, the four update methods of the HHO algorithm are used to update and improve the distribution of particles, so that the particles are constantly close to the real state, ensuring that the particle set continues to move closer to the high likelihood area, thereby significantly improving the performance of the filtering algorithm.
[0089] 3.2 Exploration strategy update based on sine-cosine search mechanism
[0090] In the exploration phase, when the prey escape energy exceeds 1, the HHO algorithm performs a random search. Although the random search strategy maintains population diversity to a certain extent, it reduces the search efficiency in the invalid area search and may cause a waste of computing resources. Therefore, the present invention introduces a sine-cosine search mechanism to provide more promising and differentiated areas for the HHO algorithm search, thereby enhancing the efficiency and effectiveness of the exploration. The exploration strategy update process based on the sine-cosine search mechanism is as follows:
[0091]
[0092]
[0093] Among them, r1, r2, r6 are all random numbers between [0, 1], r4 is a random number between [0, 2π], r5 is a random number between [0, 2], and t and T are the current number of iterations and the maximum number of iterations respectively.
[0094] 3.3 Development strategy based on Brownian motion
[0095] The development phase of the HHO algorithm covers four different strategies: soft siege, hard siege, progressive soft siege, and progressive hard siege. Both progressive soft siege and progressive hard siege use Levy flight, but the development capability of Levy flight is relatively limited. Since Brownian motion is a classic random process used to describe the endless and irregular movement of tiny particles suspended in liquid or gas, if Brownian motion is used to update the development phase strategy, it will further enhance the development capability of HHO and bring greater potential to the optimization process.
[0096] The adjusted HHO algorithm progressive rapid dive strategy is determined based on Brownian motion. The main difference lies in the update process of Z(t). Therefore, the update formula (15) of the progressive rapid dive soft encirclement strategy and the progressive rapid dive hard encirclement strategy should be adjusted to formula (19). The specific adjustment is as follows:
[0097] Z(t)=Y(t)+S+BM(D) (19)
[0098] Where BM() is the Brownian motion introduced in the Harris Hawk optimization process. When the "asymptotic fast dive" is adopted, the particle update process is changed by changing Z(t) and replacing the Levy flight with Brownian motion.
[0099] See also Figure 1 , the HHOPF algorithm steps of the present invention are as follows:
[0100] A HHOPF algorithm based on Harris Hawk particle filter includes the following operations:
[0101] 1) Set the number of particles N, the maximum number of iterations Max of Harris Eagle, the initialization of the state equation and the observation equation;
[0102] 2) According to formulas (1) and (2), the state value and observation value of the particle are calculated, and the state value of the particle is used as the initial position of each Harris hawk individual;
[0103] x k =f k (x k-1 ,v k-1 ) (1)
[0104] Z k =h k (x k ,n k ) (2)
[0105] f k (is the relationship between the system state at the previous moment and the current moment, h k is the relationship between the current state and its observed value, x k represents the state of the system k at the current moment, Z k represents the observed value of system k at the current moment, v k-1 and n k All noises are Gaussian distributed.
[0106] 3) Using the Harris Eagle algorithm to iteratively optimize and improve the particle distribution, and updating the particle exploration behavior according to the search mechanism of sine and cosine as shown in formula (17) and the Brownian motion as shown in formula (18) to obtain the particle distribution;
[0107]
[0108] Among them, X(t) and X(t+1) are the positions of the particle in the current and next iteration respectively, t is the number of iterations, and X rand (t) is the position of a randomly selected particle, X rabbit (t) is the position of the particle with the best fitness, r1, r2, r6, q are all random numbers between [0, 1], r4 is a random number between [0, 2π], r5 is a random number between [0, 2], and T is the maximum number of iterations;
[0109] 4) Calculate the weights of particles based on the particle filter algorithm;
[0110] 5) Normalize the particle weights based on the particle filter algorithm;
[0111] 6) Calculate the output state of each system based on the particle filter algorithm.
[0112] Specifically, the weight of the particle is calculated according to formula (6);
[0113]
[0114] in, is the state weight of particle i in system k at the current moment, is the state weight of particle i in system k at the previous moment, q() is the importance function of easy sampling, p() is the posterior probability density function, For The state of particle i in the system at time k before, is the state of particle i in system k at the previous moment.
[0115] Specifically, the particle weight is normalized according to formula (7);
[0116]
[0117] Specifically, each system state outputted by calculation (8) is:
[0118]
[0119] Further based on the development strategy of Brownian motion, the Harris Eagle algorithm is used to iteratively optimize and improve the particle distribution, in which the progressive rapid dive soft encirclement strategy is used to update the position:
[0120]
[0121] Y(t)=X rabbit (t)-E|J·X rabbit (t)-X(t)| (14)
[0122] Z(t)=Y(t)+S+BM(D) (19)
[0123] Where f(·) is the fitness function, E is the escape energy of the prey, S is a D-dimensional random vector whose elements are random numbers between [0, 1], and BM() is the introduction of Brownian motion in the Harris hawk optimization process, using Brownian motion to replace Levy flight.
[0124] Further based on the development strategy of Brownian motion, the Harris Eagle algorithm is used to iteratively optimize and improve the particle distribution, in which the progressive fast dive hard encirclement strategy is used to update the position:
[0125]
[0126] Y(t)=X rabbit (t)-E|J·X rabbit (t)-X m (t)| (16)
[0127] Z(t)=Y(t)+S+BM(D) (19)
[0128] Among them, f(·) is the fitness function, E is the escape energy of the prey, X m (t) is the average position of the individual, S is a D-dimensional random vector whose elements are random numbers between [0, 1], and BM() introduces Brownian motion in the Harris hawk optimization process, using Brownian motion to replace Levy flight.
[0129] The verification of the effectiveness of the algorithm is given below.
[0130] In order to verify the effectiveness of the proposed algorithm, a nonlinear, non-Gaussian system model is selected in the simulation experiment, and Gaussian noise with a mean of 0 is introduced. The system observation equation and state equation are as follows:
[0131]
[0132] The simulation experiment uses MATLAB2019b to execute particle filter (PF), whale optimized particle filter (WOAPF), and Harris Eagle optimized particle filter (HHOPF) for comparative analysis under different numbers of particles, and uses ablation experiments to further verify the effectiveness of the improved algorithm.
[0133] By calculating the R of different algorithms MSE , evaluate the error of the algorithm, the root mean square error R of the algorithm result MSE The calculation formula is:
[0134]
[0135] The simulation experiment mainly consists of two experiments: Experiment 1 considers the changes of the state errors of the three algorithms with the time step under different numbers of particles (30, 50, 100), as well as the root mean square errors of the three algorithms, which are used to analyze the accuracy of filtering; Experiment 2 analyzes the diversity of particle distribution of the three algorithms under different numbers of iterations under a specific number of particles.
[0136] 1. Filter accuracy test
[0137] The state estimation and corresponding error results of the PF algorithm, WOAPF algorithm, and HHOPF algorithm when the particle number N is 30, 50, and 100, the system noise variance Q = 10, and the measurement noise variance R = 1 are as follows:
[0138] The prediction accuracy of particles of different algorithms is represented by curves of different colors. The more the curve coincides with the True curve, the higher the degree of agreement, and the smaller the error, the higher the particle prediction accuracy of the algorithm. Figure 2aIt can be seen that the difference between WOAPF and the true curve is the largest in the 0-10 step length, while the PF and HHOPF algorithms are relatively consistent after the 5th step, and after the 10th step length, PF, WOAPF, and HHOPF are consistent with the true curve; accordingly, Figure 2b It is shown that the error of the PF algorithm is the largest in the step size of 0-10.
[0139] Depend on Figure 3a It can be seen that the difference between WOAPF and the true curve is the largest in the step length of 0-5, and the difference between PF and the true curve is the largest in the step length of 30-35, while the HHOPF algorithm of the present invention always matches the true curve; accordingly, Figure 3b The error of WOAPF is the largest in the step length of 0-5, the error of PF is the largest in the step length of 30-35, and the error of the HHOPF algorithm of the present invention is the smallest.
[0140] Depend on Figure 4a It can be seen that as the number of particles increases, PF, WOAPF, and HHOPF are consistent with the true curve after step 5; and in the 0-5 step length, PF and WOAPF differ the most from the true curve; correspondingly, Figure 4b In the step size of 0-5, the PF and WOAPF have the largest errors, while the HHOPF algorithm of the present invention has the smallest error.
[0141] Therefore, the HHOPF proposed in the present invention shows a higher level of prediction accuracy than the traditional particle filter PF and WOAPF. When the number of particles is the same, the prediction curve of HHOPF is significantly better than that of PF and WOAPF in terms of the consistency with the actual state, and the error of HHOPF is smaller. In addition, it can be seen from Tables 1 and 2 that compared with the WOAPF algorithm, although the calculation time of HHOPF is slightly longer than that of WOAPF algorithm, the root mean square error is significantly reduced, which fully reflects that the accuracy of particle filtering is improved by increasing the calculation time.
[0142] Table 1 Different particle numbers R MSE contrast
[0143] Number of particles N / piece PF WOAPF HHOPF 30 4.8035 3.1807 1.9736 50 32134 2.9535 1.6170 100 2.8858 2.1965 1.2043
[0144] Table 2 Comparison of running time with different particle numbers
[0145] Number of particles N / piece PF WOAPF HHOPF 30 0.0271 0.0252 0.0208 50 0.0396 0.0347 0.0267 100 0.0579 0.0482 0.0373
[0146] 2. Particle diversity test
[0147] In order to evaluate the particle diversity, PF, WOAPF, and HHOPF are compared. In this invention, different parameter settings are selected, the number of particles is 100, and the particle diversity is analyzed when the number of particle filter iterations k = 30 and k = 50. The simulation results are shown in Figure 2. Figure 5 , Figure 6 shown.
[0148] from Figure 5 , Figure 6 It is not difficult to find that the state estimation of particles can be intuitively seen from the state space distribution of particles; in the particle iteration process, compared with PF and WOAPF, the particles of the HHOPF algorithm proposed in the present invention move to the high likelihood area as a whole, and also reasonably retain some particles in the low likelihood area, ensuring the particle diversity in the entire filtering process. Therefore, HHOPF not only improves the accuracy of particles, but also improves the impoverishment of particles and increases the diversity of particles. This experimental result further verifies that the algorithm proposed in the present invention has significant advantages in dealing with the challenges in particle filtering.
[0149] The above embodiments are preferred examples for implementing the present invention, and the present invention is not limited to the above embodiments. Any non-essential additions and substitutions made by those skilled in the art based on the technical features of the technical solution of the present invention shall fall within the protection scope of the present invention.
Claims
1. A HHOPF algorithm based on Harris Hawk particle filter, characterized in that: The following operations are included: 1) Set the number of particles N, the maximum number of iterations Max of Harris Eagle, the initialization of the state equation and the observation equation; 2) According to formulas (1) and (2), the state value and observation value of the particle are calculated, and the state value of the particle is used as the initial position of each Harris hawk individual; x k =f k (x k-1 ,v k-1 ) (1) Z k =h k (x k ,n k ) (2) f k (is the relationship between the system state at the previous moment and the current moment, h k is the relationship between the current state and its observed value, x k represents the state of the system k at the current time, x k-1 , represents the state of the system k at the previous moment, Z k represents the observed value of system k at the current moment, v k-1 and n k All noises are Gaussian distributed. 3) Using the Harris Eagle algorithm to iteratively optimize and improve the particle distribution, and updating the particle exploration behavior according to the search mechanism of sine and cosine as shown in formula (17) and the Brownian motion as shown in formula (18) to obtain the particle distribution; Among them, X(t) and X(t+1) are the positions of the particle in the current and next iteration respectively, t is the number of iterations, and X rand (t) is the position of a randomly selected particle, X rabbit (t) is the position of the particle with the best fitness, r1, r2, r6, q are all random numbers between [0, 1], r4 is a random number between [0, 2π], r5 is a random number between [0, 2], and T is the maximum number of iterations; 4) Calculate the weights of particles based on the particle filter algorithm; 5) Normalize the particle weights based on the particle filter algorithm; 6) Calculate the output state of each system based on the particle filter algorithm.
2. The HHOPF algorithm based on Harris Hawk particle filtering as claimed in claim 1, characterized in that: Calculate the weight of the particles according to formula (6); in, is the state weight of particle i in system k at the current moment, is the state weight of particle i in system k at the previous moment, q() is the importance function of easy sampling, p() is the posterior probability density function, is the state of particle i in system k at the current moment, is the state of particle i in system k at the previous moment.
3. The HHOPF algorithm based on Harris Hawk particle filtering as claimed in claim 2, characterized in that: The particle weight is normalized according to formula (7); 4. The HHOPF algorithm based on Harris Hawk particle filtering as claimed in claim 3, characterized in that: Each system state is calculated according to (8):
5. The HHOPF algorithm based on Harris Hawk particle filtering as claimed in claim 1, characterized in that: When the Harris Eagle algorithm is used to iteratively optimize and improve the particle distribution, the progressive rapid dive soft encirclement strategy is used to update the position: Y(t)=X rabbit (t)-E|J·X rabbit (t)-X(t)| (14) Z(t)=Y(t)+S+BM(D) (19) Where f(·) is the fitness function, E is the escape energy of the prey, J is a random number between [0, 2], S is a D-dimensional random vector whose elements are random numbers between [0, 1], and BM() is the introduction of Brownian motion in the Harris hawk optimization process, using Brownian motion to replace Levy flight.
6. The HHOPF algorithm based on Harris Hawk particle filtering as claimed in claim 1, characterized in that: When the Harris Eagle algorithm is used to iteratively optimize and improve the particle distribution, the progressive fast dive hard encirclement strategy is used to update the position: Y(t)=X rabbit (t)-E|J·X rabbit (t)-X m (t)| (16) Z(t)=Y(t)+S+BM(D) (19) Where f(·) is the fitness function, E is the escape energy of the prey, J is a random number between [0, 2], and X m (t) is the average position of the individual, S is a D-dimensional random vector whose elements are random numbers between [0, 1], and BM() introduces Brownian motion in the Harris hawk optimization process, using Brownian motion to replace Levy flight.
Citation Information
Cited By
Flexible job shop scheduling method and system based on multi-strategy improvement
CN119904078A
