CN113612555A - Intelligent calibration algorithm and system based on mobile terminal wireless radio frequency signal intensity - Google Patents

Abstract

The invention proposes an intelligent calibration algorithm and system based on the wireless radio frequency signal strength of the mobile terminal, establishes a calibration model library according to the BP neural network calibration model with an improved nonlinear convergence factor proposed, records and saves the terminal model and the corresponding calibration model parameters; Whether the mobile terminal model is in the calibration model library, if not, use the raw RSSI observations of all APs collected by the standard mobile terminal at all collection points as the standard sampling data, and use the mobile terminal at the corresponding collection point. The observed value is used as the test sampling data; the original RSSI observation value received by the mobile terminal at any indoor position is used as the input of the calibration model, and the improved BP neural network calibration model is processed, and the final output is used as the calibration value. The invention can avoid the algorithm from falling into local optimum, has wide applicability, can complete the calibration work only through the hand-held mobile terminal, and effectively eliminates the heterogeneous differences of software and hardware of different mobile terminals.

Description

Intelligent calibration algorithm and system based on mobile terminal radio frequency signal strength

technical field

The invention relates to the field of radio frequency signal strength observation value calibration, in particular to an intelligent calibration algorithm and system based on the radio frequency signal strength of a mobile terminal.

Background technique

In the field of indoor positioning, the multi-path effect of the wireless radio frequency signal propagation of the mobile terminal can be used for environmental perception, and the fingerprint positioning can be carried out by obtaining the signal characteristics. However, because the software and hardware of different mobile terminals are different, the observed values of radio frequency signal strengths received by different mobile terminals at the same location are quite different, resulting in unusable fingerprint positioning. In order to solve the above problem, it is necessary to calibrate the observed value of the radio frequency signal strength, so that the observed value of the radio frequency signal strength received by other mobile terminals and the standard mobile terminal at the same location tends to be consistent.

The present invention proposes an intelligent calibration algorithm and system based on the strength of the radio frequency signal of the mobile terminal. Based on this, in order to further speed up the establishment and optimization of the calibration algorithm and avoid falling into local optimum, the present invention also proposes a nonlinear convergence factor a The improved BP neural network calibration model makes the observed values of RF signal strengths received by other mobile terminals and standard mobile terminals at the same location tend to be consistent. The user can complete the calibration work only by holding the mobile terminal, which effectively eliminates the heterogeneous differences of software and hardware of each mobile terminal. The operation is simple, and the adaptability to the environment can be improved.

SUMMARY OF THE INVENTION

The present invention is to solve the problems in the prior art that the received radio frequency signal strength observation values are quite different due to the heterogeneity of software and hardware of each mobile terminal, and the traditional calibration method has complex steps, narrow adaptability and low scalability.

In order to solve this technical problem, the present invention proposes an intelligent calibration algorithm and system based on the strength of radio frequency signals of mobile terminals, which can establish a corresponding improved BP neural network calibration model for each mobile terminal, wherein the method includes the following step:

According to the proposed BP neural network calibration model with improved nonlinear convergence factor, a calibration model library is established, and the terminal model and the corresponding calibration model parameters are recorded and saved;

Before holding a mobile terminal for calibration, determine whether the terminal model is in the calibration model library;

If not, take all the original radio frequency signal strength observations of all APs at all collection points of the standard mobile terminal as the standard sampling data, and use all the original radio frequency signal strength observations of all corresponding APs at the corresponding collection points of the mobile terminal as the test Sampling data; establish and train an improved BP neural network calibration model according to the standard sampling data and the test sampling data, and record and save the terminal model and the corresponding calibration model parameters;

If so, directly use the calibration model parameters corresponding to the calibration model library to calibrate;

When holding the mobile terminal for calibration, the original RF signal strength observation value received at any position in the room is used as the input of the calibration model, and it is processed by the improved BP neural network calibration model, and the final output is used as the calibration value to complete the calibration model. Smart calibration of mobile terminals.

Based on the intelligent calibration algorithm of the wireless radio frequency signal strength of the mobile terminal, the method for establishing the calibration model library by using the improved BP neural network calibration model includes the following steps:

阈值

作为初始鲸鱼群位置向量X_i，设置鲸鱼种群大小N，当前鲸鱼种群迭代次数t＝0，鲸鱼种群最大迭代数t_max，当前BP神经网络迭代次数T，BP神经网络最大迭代次数T_max；Step 1: Select a number of collection points at random in the room. The standard mobile phone collects the RF signal strength at all the collection points, and comprehensively expresses it as the initial standard sampling data, and observes all the original RF signal strengths of all the corresponding APs at the corresponding collection point by the mobile terminal. The value is used as the test sampling data, and the test sampling data is the real input value of the improved BP neural network calibration model, and the initial standard sampling data is the real output value of the improved BP neural network calibration model. The weights of each layer

threshold

As the initial whale population position vector X _i , set the whale population size N, the current whale population iteration number t=0, the whale population maximum iteration number t _max , the current BP neural network iteration number T, and the BP neural network maximum iteration number T _max ;

Step 2: When the number of iterations t of the whale population is less than the maximum number of iterations t _max of the whale population, calculate the fitness value f(X _i ) of each whale, and find the best fitness and the corresponding optimal whale position X _best ;

Among them, _yi is the ith RSSI real value of the standard sampling data, y is the ith RSSI prediction value of the test sampling data, and n is the number of samples;

Step 3: In order to speed up the establishment and optimization of the calibration algorithm and improve the update iteration speed, a nonlinear convergence factor a is proposed to simulate the shrinking behavior of surrounding prey. The convergence factor a only changes dynamically with the current iteration number t, which can be It is more effective to avoid falling into local optimum;

Among them, t is the number of iterations of the current whale population, and t _max is the maximum number of iterations of the whale population; update the whale position parameters A and C;

A=2ar-a (3)

C=2r (4)

Among them, r is a random number in [0,1];

Step 4: Randomly generate probability p, determine whether p is less than 0.5, if p ≥ 0.5, then update the shrinking and surrounding position:

Among them, t is the number of iterations of the current whale population; X _best is the optimal whale position; X _i is the current whale position; A and C are the coefficient vectors obtained in step 3;

If p<0.5, and when |A|<1, perform a spiral walk:

表示当前鲸鱼与最优位置的距离；b为常数，定义对数螺旋线形状；l为[-1,1]中随机数；in,

Indicates the distance between the current whale and the optimal position; b is a constant, defining the shape of a logarithmic spiral; l is a random number in [-1,1];

When |A| ≥ 1, a random walk is performed according to the following formula:

Among them, X _rand is a randomly selected position vector;

Step 5: The number of whale population iterations t increases automatically, and the optimal position is updated according to step 2; when the maximum number of iterations t _max of the whale population is reached, output X _best , which is the optimal weight w _ij and threshold θ _j , and serve as the BP neural network The optimal initial parameters of the network;

Step 6: The BP neural network performs the forward propagation process, performs data processing through the connection weight w _ij between neurons and the neuron threshold θ _j , and uses the nonlinear Sigmoid activation function to obtain the predicted output value.

Among them, w _ij is the connection weight of neuron i to neuron j; θ _j is the threshold of neuron j; I _j is the input value of neuron j; O _j is the output value of neuron j; RSSI _x,i is the neuron The input value of i.

Step 7: The BP neural network performs the error back propagation process, obtains the predicted output value through the forward propagation process, and obtains the loss function E of the current iteration number according to the difference between the predicted output value of the smart device and the actual output value of the standard smart device _j , the error is back-propagated to the neurons of the previous layer to obtain the layer error, which is passed layer by layer to the top hidden layer, and the connection weights and thresholds are continuously adjusted based on the gradient descent method.

Among them, RSSI _y,j is the real output value of neuron j; RSSI′ _y,j is the predicted output value of output layer neuron j; w′ _ij is the updated weight; θ′ _j is the updated threshold; η∈(0 ,1) is the learning rate. If the value is too large, the convergence will be fast but it is easy to fall into the local optimum. If it is too small, the convergence will be slow but approach the global optimum.

Step 8: After repeated learning and training, when the current iteration number T of the BP neural network reaches the maximum iteration number T _max of the BP neural network, the BP neural network with the smallest loss function E _j is selected as the final calibration model, and the current improved BP neural network calibration is saved. parameters of the model.

Step 9: Repeat steps 1 to 8 for multiple mobile terminals, and then establish a calibration model library.

The present invention also proposes an intelligent calibration system based on the strength of the wireless radio frequency signal of the mobile terminal, wherein the system includes:

The data acquisition module uses all the original RF signal strength observations of all APs at all collection points of the standard mobile terminal as the standard sampling data, and uses all the original RF signal strength observations of all corresponding APs at the corresponding collection points of other mobile terminals as the test. sampling data;

A model establishment module, establishes and trains an improved BP neural network calibration model according to the standard sampling data and the test sampling data, records and saves the terminal model and the corresponding calibration model parameters;

The intelligent calibration module takes the RF signal value received by the mobile terminal at any indoor position as the input of the calibration model, processes it through the improved BP neural network calibration model, and finally obtains the output as the calibration value to complete the intelligent calibration of the mobile terminal. Calibration.

Compared with the prior art, an intelligent calibration algorithm and system based on the radio frequency signal strength of a mobile terminal proposed by the present invention can avoid the algorithm from falling into local optimum, improve the adaptability to the environment, and has strong scalability and wide applicability. , the calibration work can be completed only by the handheld mobile terminal, which effectively eliminates the software and hardware heterogeneous differences of different mobile terminals, and the operation is simple.

Additional aspects and advantages of the present invention will be set forth, in part, from the following description, and in part will be apparent from the following description, or may be learned by practice of the invention.

Description of drawings

Fig. 1 is the principle frame diagram of Embodiment 1 of the present invention;

2 is a flowchart of Embodiment 1 of the present invention;

FIG. 3 is a schematic structural diagram of Embodiment 2 of the present invention.

Detailed ways

In order to facilitate understanding of the present invention, the present invention will be described more fully hereinafter with reference to the related drawings. Preferred embodiments of the invention are shown in the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terms used herein in the description of the present invention are for the purpose of describing specific embodiments only, and are not intended to limit the present invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

Example 1

Please refer to FIG. 1 and FIG. 2 , for an intelligent calibration algorithm based on the radio frequency signal strength of a mobile terminal proposed by the first embodiment of the present invention, the specific implementation method includes the following steps:

S101, establishing a calibration model library according to the proposed BP neural network calibration model with an improved nonlinear convergence factor, and recording and saving the terminal model and the corresponding calibration model parameters;

S102, before carrying out the calibration of the handheld mobile terminal, determine whether the terminal model is in the calibration model library, if not, take all the original radio frequency signal strength observations of all APs at all the collection points of the standard mobile terminal as the standard sampling data, and use the All raw radio frequency signal strength observations of all corresponding APs on the mobile terminal at the corresponding collection point are used as test sampling data;

S103, take the original radio frequency signal strength observation value received by the mobile terminal at any indoor position as the input of the calibration model, process it through the improved BP neural network calibration model, and finally obtain the output as the calibration value to complete the calibration of the mobile terminal. Smart calibration.

The calibration algorithm of the present application is to establish a calibration model between a standard mobile terminal and other mobile terminals. A mobile terminal has only one corresponding calibration model in the end, which is not directly related to the collection point, that is, the sampling data of the calibration model can be either in a The collection point can also be collected at many collection points, and the radio frequency signal refers to WiFi, Bluetooth, and so on. There is a one-to-one correspondence between the collection points in the standard sampling data and the test sampling data, and the AP and all its RSSIs are also in a one-to-one correspondence.

In S102, whether other mobile terminals have calibration models of the model are discussed in two cases:

a. If there is a calibration model of the mobile terminal model, then jump to step S103. Before calibrating the mobile terminal, when it is judged that the terminal model exists in the calibration library, the mobile terminal can be used to receive any location in the room. The observed value of the original RF signal strength is used as the input of the calibration model, which is processed by the improved BP neural network calibration model, and the final output is used as the calibration value to complete the intelligent calibration of the mobile terminal;

b. If the terminal model is not in the calibration library, use all the original radio frequency signal strength observations of all APs at all acquisition points of the standard mobile terminal as the standard sampling data, and use all the original radio frequency signals of all APs at the corresponding acquisition points of the terminal as the standard sampling data. The intensity observation value is used as the test sampling data, and then jumps to step S101, that is, an improved BP neural network calibration model is established and trained according to the standard sampling data and the test sampling data, and the terminal model and the corresponding calibration model are recorded and saved. model parameters, and finally jump to step S103 to complete the intelligent calibration of the mobile terminal.

It should be pointed out here that the collected radio frequency signal is a signal source that already exists in the indoor environment, and there is no need to deploy APs again.

In addition, in order to ensure that the standard sampling data and test sampling data are stable and reliable, the sampling time at each collection point is set to 10 minutes, and the sampling frequency is 5 seconds.

Further, the original data sending format of the standard sampling data obtained from the standard mobile terminal or the test sampling data obtained from the other mobile terminals is:

{P ₁ {(AP ₁₁ ,RSSI ₁ ,...,RSSI _i ,...,RSSI _k ),...,(AP _1n ,RSSI ₁ ,...,RSSI _i ,...,RSSI _k )},

......

P _i {(AP _i1 ,RSSI ₁ ,...,RSSI _i ,...,RSSI _k ),...,(AP _in ,RSSI ₁ ,...,RSSI _i ,...,RSSI _k ) },

......

P _j {(AP _j1 ,RSSI ₁ ,...,RSSI _i ,...,RSSI _k ),...,(AP _jn ,RSSI ₁ ,...,RSSI _i ,...,RSSI _k ) }}

Among them, Pi is the position of the _ith collection point, _i =1,2,...,j,j is the number of collection points, AP _in is the nth AP received at the collection point Pi, RSSI _k is the acquired kth original RF signal strength observation value.

S101 , a calibration model library is established based on the BP neural network calibration model improved according to the proposed nonlinear convergence factor, and the terminal model and corresponding calibration model parameters are recorded and saved.

In this step, its specific implementation process is as follows:

Step 1: Select a number of collection points at will in the room (if there are multiple environments in the room, such as corridors, rooms, stairwells, etc., in each scene, several collection points will be selected to represent this scene. environment), use a standard mobile phone to collect RF signal strength at all collection points, and comprehensively express it as the initial standard sampling data (for example, if n1 pieces of data are collected at point a, and n2 pieces of data are collected at point b, the synthesis is n1+n2 pieces of data) , and use all the original radio frequency signal strength observations of all corresponding APs at the corresponding collection point of the mobile terminal as the test sampling data;

阈值

作为初始鲸鱼群位置向量X_i，设置鲸鱼种群大小N，当前鲸鱼种群迭代次数t＝0，鲸鱼种群最大迭代次数t_max，当前BP神经网络迭代次数T，BP神经网络最大迭代次数T_max；Taking the test sampling data as the real input value of the improved BP neural network calibration model, using the initial standard sampling data as the real output value of the improved BP neural network calibration model, and randomly initializing the weights of each layer of the neural network

threshold

As the initial whale population position vector X _i , set the whale population size N, the current whale population iteration number t=0, the whale population maximum iteration number t _max , the current BP neural network iteration number T, and the BP neural network maximum iteration number T _max ;

Step 2: Calculate the fitness value f(X _i ) of each whale, find out the best fitness and the corresponding optimal whale position X _best ;

Among them, _yi is the ith RSSI real value of the standard sampling data, y is the ith RSSI prediction value of the test sampling data, and n is the number of samples;

Step 3: In order to speed up the establishment and optimization of the calibration algorithm and improve the update iteration speed, a nonlinear convergence factor a is proposed to simulate the shrinking behavior of surrounding prey. The convergence factor a only changes dynamically with the current iteration number t, which can be It is more effective to avoid falling into a local optimum, and update the whale position parameters a, A, and C:

A=2ar-a (3)

C=2r (4)

Among them, t is the number of iterations of the current whale population, t _max is the maximum number of iterations, and r is a random number in [0,1];

Step 4: Randomly generate probability p, determine whether p is less than 0.5, if p ≥ 0.5, then update the shrinking and surrounding position:

Among them, t is the number of iterations of the current whale population; X _best is the optimal whale position; X _i is the current whale position; A and C are the coefficient vectors obtained in step 3;

If p<0.5, and when |A|<1, perform a spiral walk:

表示当前鲸鱼与最优位置的距离；b为常数，定义对数螺旋线形状；l为[-1,1]中随机数；in,

Indicates the distance between the current whale and the optimal position; b is a constant, defining the shape of a logarithmic spiral; l is a random number in [-1,1];

When |A| ≥ 1, a random walk is performed according to the following formula:

Among them, X _rand is a randomly selected position vector;

Step 5: The number of whale population iterations t increases automatically, and the optimal position is updated according to step 2; when the maximum number of iterations t _max of the whale population is reached, output X _best , which is the optimal weight w _ij and threshold θ _j , and serve as the BP neural network The optimal initial parameters of the network;

Step 6: The BP neural network performs the forward propagation process, performs data processing through the connection weight w _ij between neurons and the neuron threshold θ _j , and uses the nonlinear Sigmoid activation function to obtain the predicted output value.

Among them, w _ij is the connection weight of neuron i to neuron j; θ _j is the threshold of neuron j; I _j is the input value of neuron j; O _j is the output value of neuron j; RSSI _x,i is the neuron The input value of i.

Step 7: The BP neural network performs the error back propagation process, obtains the predicted output value through the forward propagation process, and obtains the loss function E of the current iteration number according to the difference between the predicted output value of the smart device and the actual output value of the standard smart device _j , the error is back-propagated to the neurons of the previous layer to obtain the layer error, which is passed layer by layer to the top hidden layer, and the connection weights and thresholds are continuously adjusted based on the gradient descent method.

Among them, RSSI _y,j is the real output value of neuron j; RSSI′ _y,j is the predicted output value of output layer neuron j; w′ _ij is the updated weight; θ′ _j is the updated threshold; η∈(0 ,1) is the learning rate. If the value is too large, the convergence will be fast but it is easy to fall into the local optimum. If it is too small, the convergence will be slow but approach the global optimum.

Step 8: After repeated learning and training, when the number of iterations T of the BP neural network reaches the maximum number of iterations T _max of the BP neural network, select the BP neural network with the smallest loss function E _j as the final calibration model, and save the current improved BP neural network calibration model. parameter.

Step 9: Repeat steps 1 to 8 for multiple mobile terminals, and then establish a calibration model library.

When a pedestrian needs to calibrate a handheld mobile terminal, the system will automatically match the terminal model in the calibration database of the server. If it exists, the mobile terminal can be used to calibrate the received RF signal strength observation value at any location in the room. The test sampling data should be collected at the collection point first, and the improved BP neural network calibration model of the terminal model should be established before calibration.

What needs to be added here is that any location in the room can be the collection point of the sampling data, or it can be another location. At any location, the mobile terminal receives the raw RF signal strength observations of all APs and will perform calibration in real time.

Example 2

Referring to FIG. 3 , an intelligent calibration system based on the strength of a wireless radio frequency signal of a mobile terminal proposed by the second embodiment of the present invention includes a data acquisition module 11 , a model establishment module 12 and an intelligent calibration module 13 connected in sequence;

The data acquisition module 11 is specifically used for:

All raw radio frequency signal strength observations of all APs at all collection points of the standard mobile terminal are used as standard sampling data, and all original radio frequency signal strength observations of all corresponding APs at corresponding collection points by the mobile terminal are used as test sampling data;

The model building module 12 is specifically used for:

Establish and train an improved BP neural network calibration model according to the standard sampling data and the test sampling data, record and save the terminal model and the corresponding calibration model parameters;

The intelligent calibration module 13 is specifically used for:

The radio frequency signal value received by the mobile terminal at any indoor position is used as the input of the calibration model, which is processed by the improved BP neural network calibration model, and the final output is used as the calibration value to complete the intelligent calibration of the mobile terminal.

Those skilled in the art can understand that all or part of the steps in the method of the above embodiments can be implemented by instructing the relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, the steps described in the above method are included. The storage medium includes: ROM/RAM, magnetic disk, optical disk, etc.

The above-mentioned embodiments only represent several embodiments of the present invention, and the descriptions thereof are specific and detailed, but should not be construed as a limitation on the scope of the patent of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can also be made, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be subject to the appended claims.

Claims (3)

1. Intelligent calibration algorithm based on mobile terminal wireless radio frequency signal intensity is characterized in that: the calibration algorithm establishes a corresponding improved BP neural network calibration model for each mobile terminal, and comprises the following steps:

s1, establishing a calibration model library according to the improved BP neural network calibration model obtained by the proposed nonlinear convergence factor, and recording and storing the terminal model and the corresponding calibration model parameters;

s2, before a certain mobile terminal is held by hand for calibration, judging whether the terminal model is in a calibration model library;

if not, taking all original radio frequency signal intensity observed values of all APs of the standard mobile terminal on all acquisition points as standard sampling data, taking all original radio frequency signal intensity observed values of all APs of the mobile terminal on corresponding acquisition points as test sampling data, taking the test sampling data as a real input value of the improved BP neural network calibration model, taking the standard sampling data as a real output value of the improved BP neural network calibration model, repeating the step S1 to establish and train to obtain calibration model parameters of the terminal model in the improved BP neural network calibration model, and then entering the step S3;

if yes, directly entering step S3 to calibrate by using the calibration model parameters corresponding to the calibration model library;

s3, when the mobile terminal is held by hand for calibration, the original radio frequency signal intensity observed value received at any indoor position is used as the input of a calibration model, the calibration model is processed by the improved BP neural network calibration model, and the finally obtained output is used as the calibration value to finish the intelligent calibration of the mobile terminal.

2. The intelligent calibration algorithm based on the wireless radio frequency signal strength of the mobile terminal according to claim 1, wherein the method for establishing the calibration model library according to the improved BP neural network calibration model obtained from the proposed nonlinear convergence factor comprises the following steps:

s1-1: a plurality of acquisition points are randomly selected indoors, the standard mobile phone acquires the radio frequency signal intensity on all the acquisition points and comprehensively expresses the radio frequency signal intensity as initial standard sampling data,taking all original radio frequency signal intensity observed values of all corresponding APs on a corresponding acquisition point of a mobile terminal as test sampling data, taking the test sampling data as a real input value of an improved BP neural network calibration model, and taking the initial standard sampling data as a real output value of the improved BP neural network calibration model; randomly initializing weights of each layer of neural network

Threshold value

As initial whale flock position vector X_iSetting the whale population size N, setting the current whale population iteration time t to be 0, and setting the maximum iteration time t of the whale population_maxCurrent BP neural network iteration number T, maximum BP neural network iteration number T_max；

S1-2: when the current iteration times t of the whale population is less than the maximum iteration times t of the whale population_maxThen, the fitness value f (X) of each whale was calculated_i) Finding out the best fitness and the corresponding optimal whale position X_best：

In the formula, y_iThe real value of the ith RSSI of the standard sampling data is obtained, y is the predicted value of the ith RSSI of the test sampling data, and n is the number of samples;

s1-3: in order to accelerate the establishment and optimization of a calibration algorithm and improve the updating iteration speed, a nonlinear convergence factor a is provided for simulating the shrinkage behavior of a surrounding prey, the convergence factor a only dynamically changes along with the current iteration time t, the algorithm can be effectively prevented from falling into local optimum, and the calculation formula of the nonlinear convergence factor a is as follows:

wherein t is the iteration number of the current whale population, and t_maxThe maximum iteration number of the whale population is obtained; the whale location parameter A, C is updated as follows:

A＝2ar-a (3)

C＝2r (4)

wherein r is a random number of [0,1 ];

s1-4: randomly generating probability p, judging whether p is less than 0.5, if p is more than or equal to 0.5, updating the contraction surrounding position:

in the formula, t is the iteration times of the current whale population; x_bestIs the optimal whale position; x_iIs the current whale position; a and C are coefficient vectors obtained in the step S1-3;

if p <0.5, and when | A | <1, a helical walk is performed:

in the formula,

representing the distance of the current whale from the optimal position; b is a constant and defines the shape of a logarithmic spiral; l is [ -1,1 [ ]]A medium random number;

when | A | ≧ 1, random walk is performed according to equation (6):

in the formula, X_randIs a randomly selected position vector;

s1-5: the iteration times t of the whale population are increased automatically, and the optimal position is updated according to the comparison in the step S1-2; maximum iteration when whale population is reachedNumber of times t_maxThen, output X_bestI.e. the optimal weight w_ijThreshold value theta_jAnd is used as the optimal initial parameter of the BP neural network;

s1-6: the BP neural network carries out the forward propagation process and passes the connection weight w between the neurons_ijAnd neuron threshold θ_jAnd (3) processing data, and obtaining a predicted output value by adopting a nonlinear Sigmoid activation function:

in the formula, w_ijThe connection weight from the neuron i to the neuron j is obtained; theta_jIs neuron j threshold; i is_jInputting a value for neuron j; o is_jOutputting a value for neuron j; RSSI_x，iIs the input value of neuron i;

s1-7: the BP neural network carries out an error back propagation process, obtains a predicted output value through the forward propagation process, and obtains a loss function E of the current iteration times according to the difference between the predicted output value of the intelligent equipment and the real output value of the standard intelligent equipment_jAnd reversely propagating the error to the neuron at the upper layer to obtain the error at the layer, transmitting the error layer by layer until the hidden layer at the uppermost layer, and continuously adjusting the connection weight and the threshold value based on a gradient descent method:

wherein RSSI_y，jTrue output values for neuron j; RSSI'_y，jPredicting an output value for output layer neuron j; w'_ijIs the updated weight value; theta'_jIs an updated threshold; eta belongs to (0,1) as a learning rate, if the value is larger, the convergence is fast but the local optimum is easy to fall into, and if the value is smaller, the convergence is slow but the global optimum is approached;

s1-8: after repeated learning and training, when the iteration number T of the BP neural network reaches the maximum iteration number T of the BP neural network_maxThen, selecting a loss function E_jThe minimum BP neural network is used as a final calibration model, and the parameters of the current improved BP neural network calibration model and the corresponding mobile terminal model are saved;

s1-9: and repeating the steps S1-1 to S1-8 on a plurality of mobile terminals, and further establishing a standard model library.

3. Intelligent calibration system based on mobile terminal radio frequency signal intensity, its characterized in that: the calibration system comprises a data acquisition module, a model establishment module and an intelligent calibration module;

the data acquisition module is used for taking all original radio frequency signal intensity observed values of all APs of a standard mobile terminal on all acquisition points as standard sampling data, and taking all original radio frequency signal intensity observed values of all corresponding APs of other mobile terminals on corresponding acquisition points as test sampling data;

the model establishing module is used for establishing and training an improved BP neural network calibration model according to the standard sampling data and the test sampling data, and recording and storing the model number of the terminal and corresponding calibration model parameters;

the intelligent calibration module is used for taking a radio frequency signal value received by the mobile terminal at any indoor position as a calibration model input, processing the radio frequency signal value through the improved BP neural network calibration model, and taking the finally obtained output as a calibration value to finish the intelligent calibration of the mobile terminal.

Images