A pseudo-satellite placement optimization method based on deep belief network and particle swarm

Abstract

The invention discloses a pseudo-satellite layout optimization method based on deep belief network and particle swarm, and relates to the technical field of base station deployment optimization. Firstly, a non-exposed space pseudo-satellite base station layout constraint index is constructed; secondly, a pseudo-satellite base station layout optimization performance evaluation index is determined; then, a pseudo-satellite base station layout optimization mathematical model is constructed; then, the collected data is trained based on the deep belief network to generate a DBN model; the training process includes an unsupervised pre-training stage and a supervised fine-tuning stage; finally, the best pseudo-satellite layout scheme under the current environment is obtained based on a DBN-PSO algorithm; the method provided by the invention improves the efficiency of the optimization process, reduces resource consumption, realizes the best layout scheme under multiple constraint conditions, and improves the overall performance of the navigation system by comprehensively evaluating the pseudo-satellite layout performance and combining the advantages of unsupervised and supervised learning.

Description

Technical Field

[0001] The present invention relates to the technical field of base station deployment optimization, and in particular to a pseudo-satellite layout optimization method based on deep belief network and particle swarm. Background Art

[0002] The rise of digital and intelligent technologies has promoted the rapid development of location-based services such as positioning and navigation. The demand for location information of people, vehicles, and objects is becoming more and more urgent, especially in non-exposed complex environments, such as airports, libraries, subways, underground parking lots, tunnels, and other environments. At present, non-exposed environment positioning technologies include positioning technologies based on external sources and natural sources. Positioning technologies based on external sources have the advantages of independent and high-precision positioning, mainly including wireless LAN positioning, Bluetooth positioning, near-ultrasound positioning, radio frequency identification positioning, ultra-wideband positioning, pseudo-satellite positioning, cellular mobile network positioning, etc., which have different advantages when facing indoor positioning needs in different non-exposed space scenarios. The positioning performance based on external sources in non-exposed environments is affected by factors such as non-line-of-sight propagation, multipath effect, and the layout of source base stations. In particular, the layout location of the source base station directly affects the line-of-sight propagation (LOS) and non-line-of-sight (NLOS) propagation of the signal, the measurement accuracy of the signal arrival time, etc., and thus affects the ranging accuracy and positioning solution accuracy. Therefore, the optimization of source base station layout is of vital significance to improve indoor positioning accuracy. In summary, this paper proposes a pseudo-satellite layout optimization method based on deep belief network and particle swarm to overcome the complex challenges of pseudo-satellite base station layout optimization in non-exposed environments. Summary of the invention

[0003] The purpose of the present invention is to provide a pseudolite layout optimization method based on deep belief network and particle swarm. The main factors affecting the pseudolite layout are analyzed from two aspects: plane position precision factor and far-near effect. Then, the pseudolite base station layout optimization performance evaluation index is constructed from three aspects: pseudolite system positioning robustness, positioning availability and positioning accuracy. A pseudolite base station layout optimization mathematical model is established. Finally, the optimal pseudolite layout scheme under the current environment is obtained based on the DBN-PSO algorithm.

[0004] To achieve the above object, the present invention provides a pseudolite layout optimization method based on deep belief network and particle swarm, comprising the following steps:

[0005] Step 1: Construct the layout constraint index of the non-exposed space pseudo-satellite base station, including the plane position precision factor and the near-far effect;

[0006] Step 2: Determine the pseudo-satellite base station layout optimization performance evaluation indicators, including Signal strength, number of visible satellites and HDOP;

[0007] Step 3, constructing a pseudo-satellite base station layout optimization mathematical model based on the non-exposed space pseudo-satellite base station layout constraint index constructed in step 1 and the pseudo-satellite base station layout optimization performance evaluation index determined in step 2;

[0008] Step 4: Train the collected data based on the deep belief network to generate a DBN model; the training process includes an unsupervised pre-training stage and a supervised fine-tuning stage;

[0009] Step 5: Obtain the optimal pseudo-satellite layout solution under the current environment based on the DBN-PSO algorithm.

[0010] Preferably, the calculation process of the plane position precision factor is as follows:

[0011] Construct the pseudo-satellite positioning observation equation, the expression is as follows:

[0012] (1);

[0013] (2);

[0014] In the formula, Indicates the observed value of the carrier phase at the Beidou B1 frequency point. Beidou The wavelength of the frequency point; is the distance from the corresponding pseudo-satellite base station to the receiver; represents the speed of light, is the receiver clock error; is the clock error of the indoor pseudo-satellite base station; is the estimated integer ambiguity; is the corresponding other noise error; and are the three-dimensional coordinates of the pseudo-satellite base station and the receiver respectively;

[0015] The least square method is used to obtain the weight coefficient matrix of formula (1), which is expressed as follows:

[0016] (3);

[0017] In the formula, is the weight matrix, is the geometric matrix, express Direction positioning error variance, express Direction positioning error variance, express Direction positioning error variance, represents the receiver clock error variance, Indicates The diagonal elements of the weight matrix, where ;

[0018] Get the plane position precision factor HDOP, the expression is as follows:

[0019] (4).

[0020] Preferably, the specific content of the near-far effect is as follows: According to the near-far effect, at the same position in the measured environment, the power value range received by the receiver is obtained as follows:

[0021] (5);

[0022] , (6);

[0023] In the formula, is the power that the receiver can receive, Indicates that the receiver is at position Click to receive the Maximum power of pseudo-satellite signal; Indicates that the receiver is at position Receive the first Maximum signal power of pseudo-satellite; Indicated by The measured area consists of points; the calculation expression of the power that the receiver can receive is as follows:

[0024] (7);

[0025] In the formula, is the signal power transmitted by the pseudo-satellite; is the receiver antenna gain; is the spatial propagation loss of radio signals; represents the receiver RF front-end loss; the calculation expression of the radio signal spatial propagation loss is as follows:

[0026] (8);

[0027] In the formula, represents the speed of light in a vacuum; Indicates the signal frequency, unit Hz; is the propagation distance in meters.

[0028] Preferably, The specific value ranges of signal strength, number of visible satellites NOVS and HDOP are as follows:

[0029] (9);

[0030] (10);

[0031] (11).

[0032] Preferably, the pseudo-satellite base station layout optimization mathematical model constructed in step 3 is as follows:

[0033] (12);

[0034] In the formula, For the receiver The first received Pseudosatellite data, Indicated in The position is calculated by value, Indicated in The receiver collects the value; Indicates that the receiver is at position Click to receive the Maximum power of pseudo-satellite signal; Indicates that the receiver is at position Receive the first Maximum signal power of pseudo-satellite; Objective function It represents the average value of the C / N0 signal strength of each pseudo-satellite in the measured environment collected by the receiver, which is used to evaluate the robustness of the system. The larger the result, the better. Objective function Indicates the calculated value in the tested environment The average value is used to evaluate the positioning accuracy of the system. The smaller the result, the better. The objective function Indicates the data collected by the receiver in the test environment. , the bigger the better.

[0035] Preferably, the unsupervised pre-training stage trains RBM layer by layer, uses contrast divergence algorithm to update weights, and stacks layer by layer to form a network structure to achieve pseudo-satellite detection. Feature extraction and representation learning of signal strength, number of visible satellites NOVS and HDOP; Initialize network weights by training RBM layer by layer; Each layer of RBM is trained to learn the distribution of input data, and the output of the previous layer is used as the input of the next layer; The specific process is as follows:

[0036] S41. Constructing joint energy function , the expression is as follows:

[0037] (13);

[0038] In the formula, represents the parameters of the restricted Boltzmann machine; A vector representing the visible variables; Represents the number of neurons in the visible layer; A vector representing hidden variables; Represents the number of neurons in the hidden layer; represents the weight of the edge between the visible layer neurons and the hidden layer neurons, The dimension is ; represents the bias of the visible layer neurons, Represents the bias of the hidden layer neurons, each component corresponds to a separate bias for each node, represents the jth hidden unit, represents the i-th visible unit;

[0039] S42. According to the determined And the constructed joint energy function obtains the vector of visible variables and a vector of hidden variables The joint probability function of is as follows:

[0040] (14);

[0041] Substituting formula (13) into formula (14), the joint probability function is:

[0042] (15);

[0043] In the formula, is the normalization factor, and its expression is as follows:

[0044] (16);

[0045] S43. Calculate the maximum likelihood estimate from the marginal distribution of the joint probability function, as follows:

[0046] (17);

[0047] S44. Given a vector of hidden variables In the case of state, calculate the vector of visible variables The i-th visible unit The probability of being turned on is expressed as follows:

[0048] (18);

[0049] S45. Given a vector of visible variables In the case of , calculate the vector of hidden variables The jth hidden unit The probability of being turned on is expressed as follows:

[0050] (19);

[0051] In the formula, It is a commonly used activation function ;

[0052] S46, parameters , and The gradient expression is as follows:

[0053] (20);

[0054] In the formula, Indicates that given the training data In the case of and hidden units The average probability of being activated simultaneously; Indicates that given the model parameters In the case of and hidden units The average probability of being activated simultaneously; Indicates that given the training data In the case of The average probability of being activated; Indicates that given the model parameters In the case of The average probability of being activated; Indicates that given the training data In the case of The average probability of being activated; Indicates that given the model parameters In the case of The average probability of being activated;

[0055] S47, according to Criteria, Update Parameters , and , the expression is as follows:

[0056] (twenty one);

[0057] In the formula, is the learning rate.

[0058] Preferably, the supervised fine-tuning stage is specifically as follows: first, one or more fully connected layers are added to the top layer of the pre-trained model as supervised learning layers; then, supervised learning is performed using a labeled pseudo-satellite base station layout optimization performance evaluation index data set; in this process, the output is calculated through network forward propagation, and the mean square error MSE is used as the loss function to quantify the difference between the model prediction value and the true label; then, the gradient of the loss function relative to each weight is calculated using the back-propagation algorithm, and the weights are updated according to these gradients to reduce the difference between the predicted value and the actual label; this iterative process continuously optimizes the model parameters until a specific stopping condition is met, and the specific stopping condition is that the loss no longer decreases or reaches a predetermined number of iterations; the ultimate goal is to improve the model's generalization ability for unknown data, ensuring that it not only performs well on training data, but also makes accurate predictions on unseen data. Among them, the calculation expression of MSE is as follows:

[0059] (twenty two);

[0060] In the formula, It is Real pseudo-satellite base station layout optimization performance evaluation index data, The predicted value of the training output, is the number of labeled samples.

[0061] Preferably, the specific process of obtaining the optimal pseudolite layout solution under the current environment based on the DBN-PSO algorithm in step 5 is as follows:

[0062] S51, Initialize the population: Initialize the position and velocity of the particles. The positions of the pseudo-satellites are combined into one dimensional particles, a total of particles form a particle group, where particles The position is represented by , the particle velocity is expressed as ;

[0063] S52, fitness value calculation: Generate the fitness value of each particle in the initialization position based on the DBN model Signal strength, number of visible satellites NOVS and HDOP;

[0064] S53. Update particle position :The constructed pseudo-satellite base station layout optimization mathematical model is used as the particle fitness value calculation function, according to each particle in S52 Signal strength, number of visible satellites NOVS and HDOP, obtain population The optimal fitness value in the , and use it as the current global optimal value to update the position of each particle and speed , the calculation expression is as follows:

[0065] (twenty three);

[0066] (twenty four);

[0067] In the formula, is the inertia weight; Representing the moment particle exist In the iteration Speed ​​in dimension; , is the acceleration constant, which represents the weight of the particle tracking its own historical optimal value and the group optimal value, and is generally set to 2; , For Region A random number between is the historical optimal position of a single particle, is the best position found so far for all particles; Represents particles In the Position on the dimension; Representing the moment particle exist In the iteration Position on the dimension; Representing the moment particle exist In the iteration Position on the dimension; is the compression factor, which is used to constrain the flying speed of particles; Representing the moment particle exist In the iteration Speed ​​in dimension;

[0068] S54, update inertia weight , the calculation expression is as follows:

[0069] (25);

[0070] In the formula, , Respectively represent the maximum and minimum values ​​of the initial weight of the particle swarm algorithm; is the current iteration number, is the maximum number of iterations of the algorithm;

[0071] S55. Recalculate the population After updating the position, the fitness value of each particle is updated, and the historical optimal solution and historical optimal value of each particle, the global optimal solution and the global optimal value are updated at the same time;

[0072] S56, determine whether the maximum number of termination condition iterations is reached or the fitness value is within the set range, if so, end the algorithm and output the final result, otherwise return to S53.

[0073] Therefore, the present invention adopts the above-mentioned pseudolite layout optimization method based on deep belief network and particle swarm, which has the following beneficial effects:

[0074] (1) This paper proposes a multi-dimensional performance evaluation index system covering data characteristics such as C / N0 signal strength, number of visible satellites and HDOP. The system comprehensively evaluates the performance of pseudolite layout through three core dimensions: positioning robustness, visibility and accuracy, and establishes a detailed performance evaluation factor database, which provides scientific guidance for optimizing layout.

[0075] (2) The present invention proposes to use a deep belief network (DBN) to construct a pseudo-satellite layout and performance mapping model, which significantly reduces the amount of data collection and complex calculation requirements in traditional methods; this method combines the advantages of unsupervised and supervised learning, improves the efficiency of the optimization process, and reduces resource consumption;

[0076] (3) The DBN-PSO deep belief network and particle swarm optimization algorithm provided by the present invention efficiently calculates fitness through the DBN model, dynamically adjusts the particle position, realizes the optimal layout solution under multiple constraints, and improves the overall performance of the navigation system.

[0077] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is an overall flow chart of a pseudolite layout optimization method based on deep belief network and particle swarm of the present invention;

[0079] Figure 2 This is a diagram of the DBN model architecture constructed in an embodiment of the present invention;

[0080] Figure 3 This is a flow chart of DBN-PSO algorithm optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0082] See also Figure 1-Figure 3 , a pseudo-satellite layout optimization method based on deep belief network and particle swarm, comprising the following steps:

[0083] Step 1. Construct the layout constraint indicators of pseudo-satellite base stations in non-exposed space, including the plane position precision factor and the near-far effect. The near-far effect is as follows: pseudo-satellite base stations are deployed in non-exposed space. Compared with GNSS, the pseudo-satellite is very close to the receiver. When the distance between the two changes relatively, the signal strength will change dramatically. When the distance between the two is very close, the signal power will be very strong, and when the distance is relatively far, the signal will become weaker. At the same time, the C / A code length of the pseudo-satellite positioning system is 1023. When the power of the strong signal is 24dB greater than that of the weak signal, the multiple access interference between the two signals will be very serious, and the near-far effect will appear.

[0084] The calculation process of the plane position precision factor is as follows:

[0085] Construct the pseudo-satellite positioning observation equation, the expression is as follows:

[0086] (1);

[0087] (2);

[0088] In the formula, Indicates the observed value of the carrier phase at the Beidou B1 frequency point. Beidou The wavelength of the frequency point; is the distance from the corresponding pseudo-satellite base station to the receiver; represents the speed of light, is the receiver clock error; is the clock error of the indoor pseudo-satellite base station; is the estimated integer ambiguity; is the corresponding other noise error; and are the three-dimensional coordinates of the pseudo-satellite base station and the receiver respectively;

[0089] The least square method is used to obtain the weight coefficient matrix of formula (1), which is expressed as follows:

[0090] (3);

[0091] In the formula, is the weight matrix, is the geometric matrix, express Direction positioning error variance, express Direction positioning error variance, express Direction positioning error variance, represents the receiver clock error variance, Indicates The diagonal elements of the weight matrix, where ;

[0092] Get the plane position precision factor HDOP, the expression is as follows:

[0093] (4).

[0094] The specific content of the near-far effect is as follows: According to the near-far effect, at the same position in the measured environment, the power value range received by the receiver is as follows:

[0095] (5);

[0096] , (6);

[0097] In the formula, is the power that the receiver can receive, Indicates that the receiver is at position Click to receive the Maximum power of pseudo-satellite signal; Indicates that the receiver is at position Receive the first Maximum signal power of pseudo-satellite; Indicated by The measured area consists of points; the calculation expression of the power that the receiver can receive is as follows:

[0098] (7);

[0099] In the formula, is the signal power transmitted by the pseudo-satellite; is the receiver antenna gain; is the spatial propagation loss of radio signals; represents the receiver RF front-end loss; the calculation expression of the radio signal spatial propagation loss is as follows:

[0100] (8);

[0101] In the formula, represents the speed of light in a vacuum; Indicates the signal frequency, unit Hz; is the propagation distance in meters.

[0102] Step 2: Determine the pseudo-satellite base station layout optimization performance evaluation indicators, including Signal strength, number of visible satellites and HDOP; improve the robustness, availability and accuracy of pseudo-satellite system positioning; robustness is the ability of the system to maintain relatively accurate and reliable positioning in the face of various uncertainties and challenges, such as multipath effects, signal blocking, dynamic changes in the environment (such as crowd movement, new obstacles), etc. In the non-exposed space pseudo-satellite positioning system, the pseudo-satellite receiver Signal strength is an important factor affecting the robustness of the system. Strong signal strength reduces positioning drift and errors caused by signal fluctuations or interference, helps to improve the robustness of positioning, and makes the system more reliable at different times and under different environments. Reasonable pseudo-satellite layout helps to reduce the obstruction of obstacles, improve the direct signal path, reduce signal blind spots and multipath effects, and increase signal strength. Therefore, the present invention analyzes each pseudo-satellite signal collected by the receiver The signal strength is used to evaluate the system availability performance after the non-exposed space pseudo-satellite base station is optimized. Similar to the outdoor satellite navigation system, non-exposed space positioning also requires at least four independent pseudo-satellite signals for three-dimensional positioning. The more pseudo-satellites there are and the wider the distribution, the higher the robustness and accuracy of the positioning. In a non-exposed environment, due to the obstruction of buildings, the number of visible pseudo-satellites is limited. A reasonable pseudo-satellite layout can reduce the blind area of ​​pseudo-satellite signal coverage and increase the availability of the pseudo-satellite positioning system, especially in application scenarios requiring high-precision positioning. Therefore, the present invention evaluates the system availability performance after the non-exposed space pseudo-satellite base station is optimized by analyzing the number of visible pseudo-satellites collected by the receiver. In satellite navigation, the positioning error is equal to the measurement error of the pseudo-range multiplied by the geometric precision factor formed by the satellite layout, so the precision factor to some extent characterizes the navigation positioning accuracy. The non-exposed space array pseudo-satellite positioning base station developed in this paper adopts a homologous multi-element design, and the time-frequency reference of the multi-channel pseudo-satellite signal is controlled and generated by the same clock source, and there is no clock difference between the pseudo-satellite base stations. Therefore, the present invention evaluates the system positioning accuracy performance after the non-exposed space pseudo-satellite base station is optimized by calculating the HDOP value.

[0103] in, The specific value ranges of signal strength, number of visible satellites NOVS and HDOP are as follows:

[0104] (9);

[0105] (10);

[0106] (11).

[0107] Step 3: construct a pseudo-satellite base station layout optimization mathematical model based on the non-exposed space pseudo-satellite base station layout constraint index constructed in step 1 and the pseudo-satellite base station layout optimization performance evaluation index determined in step 2; wherein the constructed pseudo-satellite base station layout optimization mathematical model is as follows:

[0108] (12);

[0109] In the formula, For the receiver The first received Pseudosatellite data, Indicated in The position is calculated by value, Indicated in The receiver collects the value; Indicates that the receiver is at position Click to receive the Maximum power of pseudo-satellite signal; Indicates that the receiver is at position Receive the first Maximum signal power of pseudo-satellite; Objective function It represents the average value of the C / N0 signal strength of each pseudo-satellite in the measured environment collected by the receiver, which is used to evaluate the robustness of the system. The larger the result, the better. Objective function Indicates the calculated value in the tested environment The average value is used to evaluate the positioning accuracy of the system. The smaller the result, the better. The objective function Indicates the data collected by the receiver in the test environment. , the bigger the better.

[0110] Step 4: Train the collected data based on the deep belief network to generate a DBN model; the training process includes an unsupervised pre-training stage and a supervised fine-tuning stage;

[0111] In the unsupervised pre-training stage, pseudo-satellites are detected by training RBM layer by layer, updating weights using contrast divergence algorithm, and stacking layer by layer to form a network structure. Feature extraction and representation learning of signal strength, number of visible satellites NOVS and HDOP; Initialize network weights by training RBM layer by layer; Each layer of RBM is trained to learn the distribution of input data, and the output of the previous layer is used as the input of the next layer; The specific process is as follows:

[0112] S41. Constructing joint energy function , the expression is as follows:

[0113] (13);

[0114] In the formula, represents the parameters of the restricted Boltzmann machine; A vector representing the visible variables; Represents the number of neurons in the visible layer; A vector representing hidden variables; Represents the number of neurons in the hidden layer; represents the weight of the edge between the visible layer neurons and the hidden layer neurons, The dimension is ; represents the bias of the visible layer neurons, Represents the bias of the hidden layer neurons, each component corresponds to a separate bias for each node, represents the jth hidden unit, represents the i-th visible unit;

[0115] S42. According to the determined And the constructed joint energy function obtains the vector of visible variables and a vector of hidden variables The joint probability function of is as follows:

[0116] (14);

[0117] Substituting formula (13) into formula (14), the joint probability function is:

[0118] (15);

[0119] In the formula, is the normalization factor, and its expression is as follows:

[0120] (16);

[0121] S43. Calculate the maximum likelihood estimate from the marginal distribution of the joint probability function, as follows:

[0122] (17);

[0123] S44. Given a vector of hidden variables In the case of state, calculate the vector of visible variables The i-th visible unit The probability of being turned on is expressed as follows:

[0124] (18);

[0125] S45. Given a vector of visible variables In the case of , calculate the vector of hidden variables The jth hidden unit The probability of being turned on is expressed as follows:

[0126] (19);

[0127] In the formula, It is a commonly used activation function ;

[0128] S46, parameters , and The gradient expression is as follows:

[0129] (20);

[0130] In the formula, Indicates that given the training data In the case of and hidden units The average probability of being activated simultaneously; Indicates that given the model parameters In the case of and hidden units The average probability of being activated simultaneously; Indicates that given the training data In the case of The average probability of being activated; Indicates that given the model parameters In the case of The average probability of being activated; Indicates that given the training data In the case of The average probability of being activated; Indicates that given the model parameters In the case of The average probability of being activated;

[0131] S47, according to Criteria, Update Parameters , and , the expression is as follows:

[0132] (twenty one);

[0133] In the formula, is the learning rate.

[0134] The supervised fine-tuning stage is as follows: first, add one or more fully connected layers as supervised learning layers on the top layer of the pre-trained model; then, use a labeled pseudo-satellite base station layout optimization performance evaluation index data set for supervised learning; in this process, the output is calculated through network forward propagation, and the mean square error MSE is used as the loss function to quantify the difference between the model prediction value and the true label; then, the back-propagation algorithm is used to calculate the gradient of the loss function relative to each weight, and the weight is updated according to these gradients to reduce the difference between the predicted value and the actual label; this iterative process continuously optimizes the model parameters until a specific stopping condition is met, and the specific stopping condition is that the loss no longer decreases or reaches a predetermined number of iterations; the ultimate goal is to improve the model's generalization ability for unknown data, ensuring that it can not only perform well on training data, but also make accurate predictions on unseen data. Among them, the calculation expression of MSE is as follows:

[0135] (twenty two);

[0136] In the formula, It is Real pseudo-satellite base station layout optimization performance evaluation index data, The predicted value of the training output, is the number of labeled samples.

[0137] Step 5: Obtain the optimal pseudolites layout scheme under the current environment based on the DBN-PSO algorithm; the specific process is as follows:

[0138] S51, Initialize the population: Initialize the position and velocity of the particles. The positions of the pseudo-satellites are combined into one dimensional particles, a total of particles form a particle group, where particles The position is represented by , the particle velocity is expressed as ;

[0139] S52, fitness value calculation: Generate the fitness value of each particle in the initialization position based on the DBN model Signal strength, number of visible satellites NOVS and HDOP;

[0140] S53. Update particle position :The constructed pseudo-satellite base station layout optimization mathematical model is used as the particle fitness value calculation function, according to each particle in S52 Signal strength, number of visible satellites NOVS and HDOP, obtain population The optimal fitness value in the , and use it as the current global optimal value to update the position of each particle and speed , the calculation expression is as follows:

[0141] (twenty three);

[0142] (twenty four);

[0143] In the formula, is the inertia weight; Representing the moment particle exist In the iteration Speed ​​in dimension; , is the acceleration constant, which represents the weight of the particle tracking its own historical optimal value and the group optimal value, and is generally set to 2; , For Region A random number between is the historical optimal position of a single particle, is the best position found so far for all particles; Represents particles In the Position on the dimension; Representing the moment particle exist In the iteration Position on the dimension; Representing the moment particle exist In the iteration Position on the dimension; is the compression factor, which is used to constrain the flying speed of particles; Representing the moment particle exist In the iteration Speed ​​in dimension;

[0144] S54, update inertia weight , the calculation expression is as follows:

[0145] (25);

[0146] In the formula, , Respectively represent the maximum and minimum values ​​of the initial weight of the particle swarm algorithm; is the current iteration number, is the maximum number of iterations of the algorithm;

[0147] S55. Recalculate the population After updating the position, the fitness value of each particle is updated, and the historical optimal solution and historical optimal value of each particle, the global optimal solution and the global optimal value are updated at the same time;

[0148] S56, determine whether the maximum number of termination condition iterations is reached or the fitness value is within the set range, if so, end the algorithm and output the final result, otherwise return to S53.

[0149] Therefore, the present invention adopts the above-mentioned pseudolite layout optimization method based on deep belief network and particle swarm. First, the main factors affecting the pseudolite layout are analyzed from two aspects: plane position precision factor and far-near effect. Then, the pseudolite base station layout optimization performance evaluation index is constructed from three aspects: pseudolite system positioning robustness, positioning availability and positioning accuracy. A pseudolite base station layout optimization mathematical model is established. Finally, the optimal pseudolite layout scheme in the current environment is obtained based on the DBN-PSO algorithm; thereby, the optimal pseudolite base station layout scheme in the current non-exposed environment is realized.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A pseudo-satellite layout optimization method based on deep belief network and particle swarm, characterized in that: The following steps are involved: Step 1: Construct the layout constraint index of the non-exposed space pseudo-satellite base station, including the plane position precision factor and the near-far effect; Step 2: Determine the pseudo-satellite base station layout optimization performance evaluation indicators, including Signal strength, number of visible satellites and HDOP; Step 3, constructing a pseudo-satellite base station layout optimization mathematical model based on the non-exposed space pseudo-satellite base station layout constraint index constructed in step 1 and the pseudo-satellite base station layout optimization performance evaluation index determined in step 2; Step 4: Train the collected data based on the deep belief network to generate a DBN model; the training process includes an unsupervised pre-training stage and a supervised fine-tuning stage; Step 5: Obtain the optimal pseudolite layout solution under the current environment based on the DBN-PSO algorithm; The calculation process of the plane position precision factor is as follows: Construct the pseudo-satellite positioning observation equation, the expression is as follows: （1）； （2）； In the formula, It represents the observed value of the carrier phase at the Beidou B1 frequency point. Beidou The wavelength of the frequency point; is the distance from the corresponding pseudo-satellite base station to the receiver; represents the speed of light, is the receiver clock error; is the clock error of the indoor pseudo-satellite base station; is the estimated integer ambiguity; is the corresponding other noise error; and are the three-dimensional coordinates of the pseudo-satellite base station and the receiver respectively; The least square method is used to obtain the weight coefficient matrix of formula (1), which is expressed as follows: （3）； In the formula, is the weight matrix, is the geometric matrix, express Direction positioning error variance, express Direction positioning error variance, express Direction positioning error variance, represents the receiver clock error variance, Indicates The diagonal elements of the weight matrix, where ; Get the plane position precision factor HDOP, the expression is as follows: （4）； The specific content of the near-far effect is as follows: According to the near-far effect, at the same position in the measured environment, the power value range received by the receiver is as follows: （5）； ， （6）； In the formula, is the power that the receiver can receive, Indicates that the receiver is at position Click to receive the Maximum power of pseudo-satellite signal; Indicates that the receiver is at position Receive the first Maximum signal power of pseudo-satellite; Indicated by The measured area consists of points; the calculation expression of the power that the receiver can receive is as follows: （7）； In the formula, is the signal power transmitted by the pseudo-satellite; is the receiver antenna gain; is the spatial propagation loss of radio signals; represents the receiver RF front-end loss; the calculation expression of the radio signal spatial propagation loss is as follows: （8）； In the formula, represents the speed of light in a vacuum; Indicates the signal frequency, unit Hz; is the propagation distance in meters; The pseudo-satellite base station layout optimization mathematical model constructed in step 3 is as follows: （12）； In the formula, For the receiver The first received Pseudosatellite data, Indicated in The position is calculated by value, Indicated in The receiver collects the value; Indicates that the receiver is at position Click to receive the Maximum power of pseudo-satellite signal; Indicates that the receiver is at position Receive the first Maximum signal power of pseudo-satellite; Objective function It represents the average value of the C / N0 signal strength of each pseudo-satellite in the measured environment collected by the receiver, which is used to evaluate the robustness of the system; objective function Indicates the calculated value in the tested environment Average value, used to evaluate the positioning accuracy of the system; objective function Indicates the data collected by the receiver in the test environment. .

2. The pseudo-satellite layout optimization method based on deep belief network and particle swarm according to claim 1, characterized in that: The specific value ranges of signal strength, number of visible satellites NOVS and HDOP are as follows: （9）； （10）； （11）。 3. The pseudo-satellite layout optimization method based on deep belief network and particle swarm according to claim 2 is characterized in that: In the unsupervised pre-training stage, the pseudo-satellite detection is realized by training RBM layer by layer, using contrast divergence algorithm to update weights and stacking layer by layer to form a network structure. Feature extraction and representation learning of signal strength, number of visible satellites NOVS and HDOP; The network weights are initialized by training RBM layer by layer; each layer of RBM is trained to learn the distribution of input data, and the output of the previous layer is used as the input of the next layer; the specific process is as follows: S41. Constructing joint energy function , the expression is as follows: （13）； In the formula, represents the parameters of the restricted Boltzmann machine; A vector representing the visible variables; Represents the number of neurons in the visible layer; A vector representing hidden variables; Represents the number of neurons in the hidden layer; represents the weight of the edge between the visible layer neurons and the hidden layer neurons, The dimension is ; represents the bias of the visible layer neurons, Represents the bias of the hidden layer neurons, each component corresponds to a separate bias for each node, represents the jth hidden unit, represents the i-th visible unit; S42. According to the determined And the constructed joint energy function obtains the vector of visible variables and a vector of hidden variables The joint probability function of is expressed as follows: （14）； Substituting formula (13) into formula (14), the joint probability function is: （15）； In the formula, is the normalization factor, and its expression is as follows: （16）； S43. Calculate the maximum likelihood estimate from the marginal distribution of the joint probability function, as follows: （17）； S44. Given a vector of hidden variables In the case of state, calculate the vector of visible variables The i-th visible unit The probability of being turned on is expressed as follows: （18）； S45. Given a vector of visible variables In the case of , calculate the vector of hidden variables The jth hidden unit The probability of being turned on is expressed as follows: （19）； In the formula, It is a commonly used activation function ; S46, parameters , and The gradient expression is as follows: （20）； In the formula, Indicates that given the training data In the case of and hidden units The average probability of being activated simultaneously; Indicates that given the model parameters In the case of and hidden units The average probability of being activated simultaneously; Indicates that given the training data In the case of The average probability of being activated; Indicates that given the model parameters In the case of The average probability of being activated; Indicates that given the training data In the case of The average probability of being activated; Indicates that given the model parameters In the case of The average probability of being activated; S47, according to Criteria, Update Parameters , and , the expression is as follows: （21）； In the formula, is the learning rate.

4. The pseudo-satellite layout optimization method based on deep belief network and particle swarm according to claim 3 is characterized in that: The supervised fine-tuning stage is as follows: first, add one or more fully connected layers as supervised learning layers on the top layer of the pre-trained model; then, use the labeled pseudo-satellite base station layout optimization performance evaluation indicator dataset for supervised learning; In this process, the output is calculated through the network forward propagation, and the mean square error MSE is used as the loss function to quantify the difference between the model prediction value and the true label; then, the gradient of the loss function relative to each weight is calculated using the backpropagation algorithm, and the weight is updated according to these gradients; this iterative process continues to optimize the model parameters until a specific stopping condition is met, which is that the loss no longer decreases or reaches a predetermined number of iterations; where the calculation expression of MSE is as follows: （22）； In the formula, It is Real pseudo-satellite base station layout optimization performance evaluation index data, The predicted value of the training output, is the number of labeled samples.

5. The pseudo-satellite layout optimization method based on deep belief network and particle swarm according to claim 4, characterized in that: The specific process of obtaining the optimal pseudolite layout solution in the current environment based on the DBN-PSO algorithm in step 5 is as follows: S51, Initialize the population: Initialize the position and velocity of the particles. The positions of the pseudo-satellites are combined into one dimensional particles, a total of particles form a particle group, where particles The position is represented by , the particle velocity is expressed as ; S52, fitness value calculation: Generate the fitness value of each particle in the initialization position based on the DBN model Signal strength, number of visible satellites NOVS and HDOP; S53. Update particle position :The constructed pseudo-satellite base station layout optimization mathematical model is used as the particle fitness value calculation function, according to the Signal strength, number of visible satellites NOVS and HDOP, get population The optimal fitness value in the , and use it as the current global optimal value to update the position of each particle and speed , the calculation expression is as follows: （23）； （24）； In the formula, is the inertia weight; Representing the moment particle exist In the iteration Speed ​​in dimension; , is the acceleration constant, which represents the weight of the particle tracking its own historical optimal value and the group optimal value; , For Region A random number between is the historical optimal position of a single particle, is the best position found so far for all particles; Represents particles In the Position on the dimension; Representing the moment particle exist In the iteration Position on the dimension; Representing the moment particle exist In the iteration Position on the dimension; is the compression factor, which is used to constrain the flying speed of particles; Representing the moment particle exist In the iteration Speed ​​in dimension; S54, update inertia weight , the calculation expression is as follows: （25）； In the formula, , Respectively represent the maximum and minimum values ​​of the initial weight of the particle swarm algorithm; is the current iteration number, is the maximum number of iterations of the algorithm; S55. Recalculate the population After updating the position, the fitness value of each particle is updated, and the historical optimal solution and historical optimal value of each particle, the global optimal solution and the global optimal value are updated at the same time; S56, determine whether the maximum number of termination condition iterations is reached or the fitness value is within the set range, if so, end the algorithm and output the final result, otherwise return to S53.

Images