Indoor positioning method for optimizing neural network based on improved black-wing plinary algorithm

Abstract

The invention discloses an indoor positioning method for optimizing a neural network based on an improved black wing plinary algorithm, and belongs to the technical field of indoor positioning. According to the method, the algorithm performance is improved by utilizing the characteristics of high evolutionary ability and high search speed of a black-wing plinary optimization algorithm, the black-wing plinary algorithm is further improved, uniformly distributed initial individuals are generated by adopting Tent chaotic mapping, and the diversity of populations is improved, so that the algorithm is prevented from being converged to a locally optimal solution too early, and the global search ability is enhanced; through a lens imaging reverse learning strategy, the stability and robustness of the algorithm are improved; a golden sine strategy is introduced in an attack and migration stage, a solution space is cut according to a golden section coefficient, and the position of a black wing plenum individual is updated. Compared with the prior art, the black-wing supporting optimization algorithm is more excellent in performance, the RSSI ranging model based on the BP neural network is constructed through the black-wing supporting optimization algorithm, the weight and the threshold value of the neural network are optimized, and high-precision positioning in the complex indoor environment is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of indoor positioning, and in particular to an indoor positioning method based on an improved black-winged kite algorithm to optimize a neural network. Background Art

[0002] Positioning technology affects all areas of people's lives, and indoor positioning technology has expanded the application scenarios of traditional positioning technology. It has received widespread attention in recent years and has good development prospects. In indoor environments, there is a lack of line-of-sight signal transmission channels between satellites and indoor receivers. Existing outdoor positioning systems are difficult to accurately provide positioning services for indoor environments. Therefore, it is very important to develop a reliable and accurate indoor positioning system.

[0003] Although the indoor positioning algorithm based on RSSI has the advantages of easy implementation, low price, low energy consumption, and strong real-time performance, in the traditional positioning algorithm based on RSSI ranging, its model parameters are often selected through experience and the accuracy is not high. In addition, in complex indoor environments, there are interference factors such as multipath error and non-line-of-sight error, which makes it difficult to accurately construct a suitable ranging model, which may cause significant positioning errors.

[0004] In recent years, many scholars have used the good nonlinear fitting ability of neural networks to reduce the impact of RSSI fluctuations on distance estimation, thereby reducing the dependence of RSSI-based ranging models on environmental parameter selection. Li Yingyu et al. used neural networks to preprocess RSSI values ​​to reduce ranging errors, and then combined with the Newton positioning algorithm to correct the node position, reducing the RSSI ranging error. Zhao Shan et al. used a particle filter model to preprocess RSSI values ​​to reduce the impact of non-line-of-sight errors and multipath transmission in indoor environments. Fei Yang et al. used BP neural networks to construct a signal propagation path loss model. The results showed that the RSSI ranging error based on the BP neural network model was smaller than that of the traditional signal propagation path loss model. Shi Xin et al. used strong tracking filtering to adaptively track RSSI step changes, overcoming the uncertainty of unmodeled states in the traditional RSSI state space model. Li Tiansong et al. used an improved gray wolf optimization algorithm to optimize the three-sided positioning algorithm, which significantly improved the accuracy of the improved radio frequency identification indoor positioning algorithm. Lin Yiruo et al. proposed an RSSI ranging algorithm based on GWO-BP neural network, which enhanced the mapping relationship between RSSI and real physical distance, thereby improving ranging accuracy and stability. Liu Bo et al. optimized the RSSI value through hybrid filtering, and improved the sparrow search algorithm to construct a fitness function to optimize the search node position, thereby improving the positioning accuracy.

[0005] In the research of the above-mentioned RSSI-based ranging algorithm, the following challenges are mainly faced: ① The traditional path loss model is easily affected by environmental parameters and RSSI signal fluctuations, resulting in insufficient positioning accuracy; ② The ranging model based on the BP neural network has high accuracy, but the BP neural network randomly gives initial weights and thresholds, which leads to local optimal problems; ③ Although the optimization algorithm introduced in the above-mentioned literature has improved the ranging accuracy of the BP neural network model to a certain extent, there are still problems such as weak global search capability and insufficient convergence accuracy. Summary of the invention

[0006] The purpose of the present invention is to provide a method for realizing high-precision RSSI indoor positioning based on an improved black kite algorithm and optimized back propagation neural network, so as to solve the problems of insufficient positioning accuracy and long operation time caused by interference such as multipath error and non-line-of-sight error in complex indoor environments.

[0007] On the one hand, the present invention utilizes the characteristics of the black kite optimization algorithm, which has strong evolutionary ability and fast search speed, to improve the algorithm performance, and further improves the black kite algorithm, adopts Tent chaotic mapping to generate uniformly distributed initial individuals, improves the diversity of the population, thereby avoiding the algorithm from converging to the local optimal solution too early, and enhances the global search capability; improves the stability and robustness of the algorithm through the lens imaging reverse learning strategy; introduces the golden sine strategy in the attack and migration stages, cuts the solution space according to the golden section coefficient, and updates the individual positions of the black kites. The improved black kite optimization algorithm has better performance, and uses the black kite optimization algorithm to construct an RSSI ranging model based on the BP neural network, optimizes the weights and thresholds of the neural network, and realizes high-precision positioning in complex indoor environments.

[0008] The technical solution adopted by the present invention is: an indoor positioning method based on an improved black kite algorithm to optimize a neural network, comprising the following steps:

[0009] S1: Initialize the parameters of the black kite algorithm;

[0010] S2: Initialize the population using Tent mapping and lens imaging strategies;

[0011] S3: Calculate the fitness of each black-winged kite and rank them;

[0012] S4: Select the black-winged kite with the best fitness value as Leader XL;

[0013] S5: Update the position of the black kite in the attack phase according to equations (1) and (2):

[0014]

[0015] In the formula, and are the positions of the ith black-winged kite in the jth dimension in the tth and t+1th iteration steps, respectively; r is a random number in [0,1], p is a constant value of 0.9; T is the total number of iterations; t is the current number of iterations;

[0016] S6: Update the position of the black kite during the migration phase according to equations (3) and (4):

[0017]

[0018] In the formula, is the score of the black kite leader in the jth dimension in t iterations so far, F i represents the position of the black-winged kite individual in the jth dimension during the tth iteration; F ri is the fitness value obtained, C(0,1) is the Cauchy mutation, which is a normally distributed random number;

[0019] S7: Update the position of the black kite according to equations (5), (6) and (7) combined with the golden sine strategy:

[0020] c1=a(1-h)+bh (5)

[0021] c2=ah+b(1-h) (6)

[0022] x gnew =x i |sin(r1)|-r2sin(r2)|c1x b -c2x i | (7)

[0023] In the formula, a and b are the initial values ​​of the golden section coefficients, and a=π, b=-π; h is the golden section ratio, r1 is a random number in [0,2π], which indicates the moving length of the individual to the optimal solution; r2 is a random number in [0,π], which indicates the moving direction of the individual to the optimal solution; x b is the optimal position of the individual; x gnew is the position of the individual after iteration; is the best attack position of the black-winged kite in the jth dimension in the tth iteration, x i is the position of the i-th black-winged kite;

[0024] S8: Calculate the objective function values ​​of all black-winged kites and output the global optimal value;

[0025] S9: Determine whether the current number of iterations has reached the maximum. If it has reached iter m , then output the position of the optimal individual and execute the next step; otherwise, increase the current iteration number and return to S5;

[0026] S10: Output the optimal solution of the black kite individual, and use the improved black kite algorithm to select the optimal weights and thresholds of the BP neural network, build the IBKA-BP neural network model, and perform nonlinear fitting of the relationship between RSSI value and distance;

[0027] S11: Complete the IBKA-BP neural network training, perform simulation prediction, output the location of the point to be predicted, and complete high-precision indoor positioning.

[0028] Compared with the prior art, the present invention has the following advantages and technical effects:

[0029] The IBKA-BP algorithm proposed in this paper introduces the golden sine strategy to effectively balance the exploration stage and the development stage in the black kite search process, and improves the convergence speed of the IBKA-BP algorithm; at the same time, the introduction of Tent mapping and lens imaging reverse learning strategies expands the range of effective solutions, allowing black kite individuals to jump out of the local optimum, improving the population quality, and making IBKA-BP perform better in positioning accuracy.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) The present invention utilizes the characteristics of the black kite optimization algorithm, namely, strong evolutionary ability and fast search speed, to improve the algorithm performance, and further improves the black kite algorithm by using Tent chaotic mapping to generate uniformly distributed initial individuals, thereby improving the diversity of the population, thereby avoiding the algorithm from converging to the local optimal solution too early and enhancing the global search capability; by using the lens imaging reverse learning strategy, the stability and robustness of the algorithm are improved; by introducing the golden sine strategy in the attack and migration stages, the solution space is cut according to the golden section coefficient, and the individual positions of the black kites are updated, so that the improved black kite optimization algorithm has better performance.

[0032] (2) The present invention obtains the initial weights and thresholds of the network by improving the powerful global search capability of the Black Kite algorithm, effectively preventing the network from falling into a local optimal solution, thereby improving the reliability and accuracy of the overall positioning, allowing the network to jump out of the suboptimal solution and find a better solution, making the overall performance and generalization ability of the model better.

[0033] (3) The indoor positioning algorithm based on the improved Black Kite algorithm optimized neural network of the present invention shows excellent performance in experimental verification. The IBKA-BP proposed in this paper requires fewer iterations to reach the optimal fitness value, indicating that the golden sine strategy effectively balances the exploration stage and the development stage in the Black Kite search process and improves the convergence speed of the IBKA-BP algorithm. The median error, average error, maximum error and root mean square error are 8.71 cm, 15.52 cm, 58.2 cm and 21.42 cm respectively, indicating that IBKA-BP effectively improves the indoor positioning accuracy and can provide a reference for indoor positioning algorithms based on ranging. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is the initial population distribution map of the black-winged kite algorithm;

[0035] Figure 2 This is the population distribution map after the introduction of the tent mapping optimization black kite algorithm;

[0036] Figure 3 Algorithm parameter setting diagram for basic test function comparison;

[0037] Figure 4 It is a basic test function table used by the present invention to compare functions;

[0038] Figure 5 This is the comparison result of basic test functions between IBKA and other algorithms;

[0039] Figure 6 It is the structure diagram of BP neural network;

[0040] Figure 7 It is a flow chart of the ranging algorithm of IBKA-BP constructed by the present invention;

[0041] Figure 8 It is a parameter setting diagram of different algorithms when the present invention performs indoor positioning result measurement;

[0042] Fig. 9 It is an iteration diagram of different algorithms when the present invention performs indoor positioning result measurement;

[0043] Fig.10 It is a distribution diagram of accumulated errors of different algorithms when the present invention performs indoor positioning result measurement;

[0044] Fig.11 This is a graph of the maximum error, average error, and root mean square error results of different algorithms when the present invention performs actual measurement of indoor positioning results. DETAILED DESCRIPTION

[0045] The following is a clear and complete description of the technical solution of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The indoor positioning algorithm based on the improved black kite algorithm to optimize the neural network includes the following steps:

[0047] S1. Initialize the parameters of the black-winged mandarin duck algorithm and use tent mapping and lens imaging to initialize the population. S1 specifically includes:

[0048] S11. Establish a random solution for initialization. The position of each black-winged mandarin duck is shown in formula (8), which can obtain the position of each black-winged kite and generate an initialized population:

[0049]

[0050] Where BK ij is the jth dimension of the i-th black-winged bird; pop is the number of solutions to the equation; dim is the dimension size of the given problem;

[0051] S12. Select the position of a black kite in the population as the basic variable, and continuously map it to the solution space according to its position in the segmented interval as shown in formula (9), generating a chaotic sequence of the solution space:

[0052]

[0053] In the formula, β is the control parameter; X ij is the position of the jth dimension of the ith individual, X′ ij is the position of the individual after mapping;

[0054] S13. Map the generated chaotic sequence to the corresponding dimension range of the population search space according to formula (10) to form a newly generated black-winged kite population individual:

[0055] X″ ij =lb j +X′ ij ×(ub j -lb j ) (10)

[0056] In the formula, lb j andub j are the lower and upper limits of the j-th dimension position of the black-winged kite, X″ ij The location of the Black Winged Kite after remapping;

[0057] S14. Repeat S12 and S13 for each dimension until a complete initial population is generated, thus completing the process of population tent mapping;

[0058] S15. When the coverage area in the search space is small, in order to increase the probability of the optimal solution, the lens imaging strategy is used to shorten the search process according to formula (11), thereby making the black kite individual optimization faster:

[0059]

[0060] In the formula, a j represents the minimum value, b j represents the maximum value, K is the ratio of the object height to the imaging height in the lens imaging strategy, and X j and It represents the best individual of the current black-winged kite and the lens imaging strategy;

[0061] S16. This algorithm uses Tent chaotic mapping to generate uniformly distributed initial individuals to improve the diversity of the population, thereby avoiding the algorithm from converging to the local optimal solution too early. The population distribution diagram before optimization is shown in Figure 1 As shown in the figure, the optimized population distribution diagram is as follows Figure 2 As shown, the S1 stage is now completed;

[0062] S2. Calculate the fitness value of each black-winged mandarin duck and sort them, select the black-winged mandarin duck with the best fitness value as the leader XL, and update the position of the black-winged mandarin duck in the attack phase and migration phase by combining the lens imaging reverse learning strategy and the golden sine strategy. S2 specifically includes:

[0063] S21. The individual with the best fitness value is set as the leader XL in the initial group, and is defined as the optimal position of the black-winged kite; the initial leader XL, as shown in formula f best =min(f(X i )) As shown in i ) is the defined fitness function, which is the error value of indoor positioning in the present invention;

[0064] S22. As shown in equations (1) and (2), the position of the black-winged kite is updated in the attack phase so as to find the best black-winged kite individual in the current phase, so that the group members can effectively explore the solution space;

[0065] S23. Based on the mathematical model of black kite migration behavior, as shown in equations (3) and (4), the best performing individual is selected as the leader according to the fitness function. As the algorithm continues to iterate, the leader will also change so that the population moves towards the optimal solution:

[0066] S24. The golden sine strategy is introduced into the Black Kite algorithm. As shown in equations (5) and (6), the golden section coefficients c1 and c2 are first calculated, and then the pace of the individual approaching the optimal position is adjusted to balance the ability of exploration and development, so that the algorithm is not easily trapped in the local optimum:

[0067] S25. The golden section coefficient will cut the solution space once every iteration, and the mathematical model will be updated as shown in formula (7):

[0068] S26. The local operator Gold-SA is introduced to improve the migration phase, combining the golden section and sine wave characteristics to enhance the ergodicity of the algorithm, reduce the optimization space, and accelerate the convergence speed;

[0069] S28. The improved mathematical model is shown in formula (12). By introducing the golden sine strategy, the optimal black kite position is found and the search mechanism is improved, thereby achieving a stronger optimization effect:

[0070]

[0071] In the formula, r1 is a random number in [0,2π], which indicates the moving length of the individual to the optimal solution; r2 is a random number in [0,π], which indicates the moving direction of the individual to the optimal solution; and are the positions of the i-th black-winged kite in the j-th dimension in the t-th and t+1-th iteration steps, respectively. is the score of the black-winged kite leader in the jth dimension in the t iterations so far, r is a random number in [0,1], p is a constant value of 0.9, c1 and c2 are shown in equations (5) and (6);

[0072] S29. The performance of the black kite optimization algorithm proposed in the present invention is verified by simulation experiments of the benchmark test function. The parameters of the benchmark test function are as follows: Figure 3 As shown, the benchmark function is Figure 4 As shown, the test results of BKA, IBKA, GWO, SSA, and WOA are as follows Figure 5 As shown in the figure, the experiment proves that the Black Kite Optimization Algorithm has the best performance among these algorithms and can be used for optimization processes that require high performance.

[0073] S3. To meet the requirements of high-precision indoor positioning, the present invention constructs an RSSI ranging model based on a BP neural network to accurately capture the relationship between RSSI and distance in a complex environment and effectively improve positioning accuracy. S3 specifically includes:

[0074] S31. The present invention takes into account the shadow fading effect in the environment, empirically fits actual data, describes the propagation loss of wireless signals in complex environments, and selects Gaussian white noise to represent interference, forming a log-normal path loss model, as shown in formula (13):

[0075]

[0076] Where PL(d) and PL(d0) represent the path loss at d and d0 meters from the transmitter, respectively. Usually d0 = 1m. The shadow fading X σ It follows a normal distribution, η is the path loss factor, which is related to the environment;

[0077] S32. At this time, the RSSI value of the receiving position d meters away from the transmitting end can be expressed as formula (14), where P t is the transmitted signal power:

[0078] RSSI(d)=P t -PL(d) (14)

[0079] S33. In addition, by substituting formula (13) into formula (14), the relationship between RSSI and distance can be obtained as shown in formula (15):

[0080]

[0081] S34. It can be seen from (15) that the parameters A and η will significantly affect the relationship between RSSI and distance. Selecting appropriate parameters is crucial to building an accurate log-normal path loss model. Therefore, the present invention uses a BP neural network to find the required parameters and build a neural network ranging model. The ranging model built by the present invention is as follows: Figure 6 As shown;

[0082] S4. Further, in order to improve the accuracy of the indoor positioning algorithm, the optimal weights and thresholds of the BP neural network are selected and trained through the improved black-winged mandarin duck optimization algorithm, and the established neural network RSSI ranging model is used for position prediction. S4 specifically includes:

[0083] S41. Select the multi-strategy optimized black kite optimization algorithm to find the optimal weights and thresholds, and construct a neural network ranging flow chart, such as Figure 7 As shown;

[0084] S42. Arrange the simulation environment to simulate the indoor positioning environment of 6m×6m. Arrange four anchor nodes at the four corners (0,0), (0,6, (6,0) and (6,6) respectively. The relevant indoor parameter settings are as follows: Figure 8As shown in the figure, 100 nodes to be located are randomly generated in the same simulated indoor environment, and the positioning accuracy of the ranging algorithms based on BKA-BP (Black Kite Optimized BP Neural Network), GWO-BP (Grey Wolf Algorithm Optimized BP Neural Network), ISSA-BP (Improved Sparrow Search Algorithm Optimized BP Neural Network) and IBKA-BP are compared;

[0085] S43. Iteration curves of different algorithms are shown in Figure 4. Fig. 9 As shown, it can be seen that the IBKA-BP proposed in the present invention requires fewer iterations to reach the optimal fitness value, indicating that the golden sine strategy effectively balances the exploration phase and the development phase in the black-winged kite search process and improves the convergence speed of the IBKA-BP algorithm;

[0086] S44. The CDF diagram of the cumulative error of different algorithms is as follows Fig.10 As shown, the maximum error, average error and mean square error are Fig.11 As shown. It can be seen that the positioning errors of IBKA-BP, ISSA-BP, BKA-BP and GWO-BP less than 10cm account for 61%, 45%, 40% and 29% respectively, and the median positioning errors are 8.71cm, 11.35cm, 15.12cm and 22.3cm respectively. The maximum error, average error and root mean square error of IBKA-BP are better than those of other algorithms. The simulation results verify that the introduction of Tent mapping and lens imaging reverse learning strategies in the present invention expands the range of effective solutions, enables black-winged kite individuals to jump out of the local optimum, improves the population quality, and makes IBKA-BP perform better in terms of positioning accuracy, which proves the effectiveness of the indoor positioning method proposed by the present invention based on the improved black-winged kite algorithm to optimize the neural network.

Claims

1. An indoor positioning method based on an improved black kite algorithm to optimize a neural network, characterized in that: The method is as follows: first, the Tent chaotic mapping is used to improve the initial population distribution and improve the convergence accuracy; then, the lens imaging reverse learning strategy is introduced to improve the quality of the initial black kite population; finally, the golden sine strategy is established to update the position of the black kite to improve the optimization efficiency of the algorithm; and then the IBKA-BP neural network ranging model is established by optimizing the initial weights and initial thresholds of the neural network algorithm through the improved black kite algorithm to achieve high-precision positioning.

2. A method for indoor positioning based on an improved black kite algorithm and optimized neural network according to claim 1, characterized in that: The following steps are involved: S1: Initialize the parameters of the black kite algorithm; S2: Initialize the population using Tent mapping and lens imaging strategies; S3: Calculate the fitness of each black-winged kite and rank them; S4: Select the black-winged kite with the best fitness value as Leader XL; S5: Update the position of the black kite in the attack phase according to equations (1) and (2): In the formula, and are the positions of the ith black-winged kite in the jth dimension in the tth and t+1th iteration steps respectively; r is a random number in [0,1], p is a constant value of 0.9; T is the total number of iterations; t is the current number of iterations; S6: Update the position of the black kite during the migration phase according to equations (3) and (4): In the formula, is the score of the black kite leader in the jth dimension in t iterations so far, F i represents the position of the black-winged kite individual in the jth dimension during the tth iteration; F ri is the fitness value obtained, C(0,1) is the Cauchy mutation, which is a random number of normal distribution; S7: Update the position of the black kite according to equations (5), (6) and (7) combined with the golden sine strategy: c1=a(1-h)+bh (5) c2=ah+b(1-h) (6) x gnew =x i |sin(r1)|-r2sin(r2)|c1x b -c2x i | (7) In the formula, a and b are the initial values ​​of the golden section coefficients, and a=π, b=-π; h is the golden section ratio, r1 is a random number in [0,2π], which indicates the moving length of the individual to the optimal solution; r2 is a random number in [0,π], which indicates the moving direction of the individual to the optimal solution; x b is the optimal position of the individual; x gnew is the position of the individual after iteration; is the best attack position of the black-winged kite in the jth dimension in the tth iteration, x i is the position of the i-th black-winged kite; S8: Calculate the objective function values ​​of all black-winged kites and output the global optimal value; S9: Determine whether the current number of iterations has reached the maximum. If it has reached iter m , then output the position of the optimal individual and execute the next step; otherwise, increase the current iteration number and return to S5; S10: Output the optimal solution of the black kite individual, and use the improved black kite algorithm to select the optimal weights and thresholds of the BP neural network, build the IBKA-BP neural network model, and perform nonlinear fitting of the relationship between RSSI value and distance; S11: Complete the IBKA-BP neural network training, perform simulation prediction, output the location of the point to be predicted, and complete high-precision indoor positioning.

Images