Multi-sensor error compensation method based on ISSA-BP neural network

Abstract

The present invention relates to a multi-sensor error compensation method based on ISSA-BP neural network, belonging to the field of sensor compensation. The method uses a trained ISSA-BP neural network to perform environmental error compensation on three sensors of temperature, humidity and pressure carried by the system respectively, thereby eliminating the detection errors caused by environmental temperature, humidity and air pressure. The method solves the problem that the prior art does not fully consider the influence of multiple environmental factors such as temperature, humidity and air pressure in the detection environment of the multi-sensor system on the detection data, resulting in poor robustness of the compensation method, and the problem that the traditional sensor error compensation algorithm has low accuracy and slow time.

Description

Technical Field

[0001] The invention belongs to the field of sensor compensation, and in particular relates to a multi-sensor error compensation method based on ISSA-BP neural network. Background Art

[0002] As the core of all atmospheric environmental parameter measurements, the multi-sensor system integrates multiple sensors such as temperature, humidity, and pressure to achieve in-situ detection of temperature, humidity, pressure, and other meteorological parameters in the space environment. However, the atmospheric environment is complex and changeable. During the detection process, the temperature, humidity, pressure, and other sensors carried by the multi-sensor system are exposed to the outside for a long time and will be affected by environmental changes such as dynamic temperature, humidity, and air pressure altitude, causing errors in the detection data. Therefore, in order to obtain more accurate detection data and improve the detection performance of the system, it is urgent to perform comprehensive error compensation on the multi-sensor detection data.

[0003] In the prior art, the detection data of the multi-sensor system are compensated by conventionally performing temperature compensation on the three sensors of temperature, humidity and pressure separately, mainly through hardware compensation and software compensation. Hardware compensation uses, for example, differential methods to offset the influence of temperature on the detection data, but it will increase the amount of system calculations, make wiring complex, and have high costs. Software compensation methods include interpolation, least squares fitting, back propagation (BP) neural network and other methods. Although they can achieve a certain compensation effect to a certain extent, there is still room for improvement. During the operation of the multi-sensor system, the detection data is not only affected by the ambient temperature, but the system will introduce many error interferences as the environmental state changes. From the above analysis, it can be seen that for the compensation of temperature, humidity and pressure of multiple sensors, the current research scheme not only considers the situation too single, is limited to the temperature compensation of the sensor, and does not consider the influence of other environmental parameters on the system, but also the compensation algorithm has its own limitations, the compensation accuracy is not high, and the processing time is too slow. Since the system is in a complex and disturbed environment, light radiation, air pressure altitude, temperature and humidity changes will affect the sensor's measurement data. Therefore, the compensation method needs to consider the impact of the environment as comprehensively as possible and obtain more accurate multi-sensor detection values ​​by optimizing the accuracy of the compensation algorithm.

[0004] In summary, the shortcomings of the existing technology and the technical reasons for these shortcomings are summarized as follows: (1) The existing technical solutions consider too single a situation. In the process of multi-sensor system detecting the environment in the three-dimensional space of the atmosphere, because the environmental conditions in different spaces and at different times are different, the compensation model should be considered more comprehensively, and the error caused by the change of environmental conditions should be eliminated as much as possible. In addition to the temperature measurement error caused by the heating of the sensor due to solar radiation, the errors caused by the changes in the temperature, humidity, pressure and other factors in the environment where the system is located should also be taken into account. The multi-sensor error compensation model should also consider these environmental factors as comprehensively as possible, comprehensively compensate for the impact caused by the changes in environmental temperature, humidity and pressure, and improve the accuracy of the detection data. (2) In addition, the existing compensation algorithm has too low accuracy, long processing time and poor robustness. Since the traditional BP neural network compensation algorithm is a nonlinear optimization, the initialization of the weights is random, so there will be local minima problems, and the newly added samples will affect the samples that have been learned, and the convergence speed of the learning algorithm is slow. Summary of the invention

[0005] In view of the above problems, the present invention provides a multi-sensor error compensation method based on an ISSA-BP neural network, which uses a trained ISSA-BP neural network to perform environmental error compensation on the three sensors of temperature, humidity and pressure carried by the system, thereby eliminating the detection errors caused by ambient temperature, humidity and air pressure. It solves the problem that the prior art does not fully consider the influence of multiple environmental factors such as temperature, humidity, air pressure in the detection environment of the multi-sensor system on the detection data, resulting in poor robustness of the compensation method, and the problem that the traditional sensor error compensation algorithm has low accuracy and slow time.

[0006] The present invention provides a multi-sensor error compensation method based on ISSA-BP neural network, comprising:

[0007] S1. Place the multi-sensor system in a test chamber, change the environmental parameters of the test chamber, and record the test values ​​of each sensor in the multi-sensor system;

[0008] S2, preprocessing the output value of each sensor in the multi-sensor system to obtain a neural network data set;

[0009] S3, establishing a correlation between the environmental parameters and the output values ​​of each sensor in the multi-sensor system in step S1; and building a BP neural network architecture based on the correlation;

[0010] S4, introducing the ISSA model into the BP neural network architecture to establish an initial ISSA-BP neural network; inputting the neural network data set obtained in step S2 into the initial ISSA-BP neural network, updating the weights and thresholds of the initial ISSA-BP neural network, and obtaining an updated ISSA-BP neural network;

[0011] S5. Acquire the data of each sensor in the multi-sensor system to be tested; input the data of each sensor into the updated ISSA-BP neural network, and output the predicted ambient temperature, humidity and air pressure values ​​after network compensation.

[0012] Preferably, step S1, placing the multi-sensor system in a test box, changing the environmental parameters of the test box; recording the standard value of the current environment and the test value of each sensor in the multi-sensor system;

[0013] Exemplarily, the output value of each sensor in the multi-sensor system includes the temperature, humidity and pressure of each sensor; the environmental parameters are temperature, humidity and pressure;

[0014] The multi-sensor system integrates multiple environmental parameter measurement sensors of temperature, humidity and pressure;

[0015] Preferably, in step S2, the output values ​​of each sensor in the multi-sensor system are preprocessed to obtain a neural network data set (H, P, T, H e ,P e ,T e );

[0016] Furthermore, we get the neural network dataset (H, P, T, H e ,P e ,T e ) The specific steps include:

[0017] The output values ​​of each sensor in the multi-sensor system and the real parameters of the environment are normalized and standardized in turn to obtain processed data; the processed data are classified by feature labels to obtain classified sensor data;

[0018] Based on the classified sensor data, the neural network data set (H, P, T, H e ,P e ,T e ), the data set is divided into a training set and a test set according to a certain ratio; where H e is the actual humidity of the environment, P e is the actual ambient pressure, T eis the actual temperature of the environment, H is the ambient humidity value measured by the multi-sensor system, P is the ambient pressure value measured by the multi-sensor system, and T is the ambient temperature value measured by the multi-sensor system;

[0019] Preferably, the specific steps of establishing the association relationship between the environmental parameters and the output values ​​of each sensor in the multi-sensor system in step S3 include:

[0020] S31, calculating the Pearson correlation coefficient between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system; judging the correlation between the environmental parameter and the multi-sensor system based on the Pearson correlation coefficient, and obtaining the correlation relationship between the environmental parameter and the multi-sensor system;

[0021] It can be understood that the relationship between the environmental parameter and the multi-sensor system is the existence or non-existence of correlation between the real value of the environmental parameter and the test value of each sensor in the multi-sensor system;

[0022] Specifically, if the Pearson correlation coefficient between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system is close to 0,

[0023] Then the correlation between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system is weak;

[0024] If the correlation between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system is close to 1 or -1, then there is a strong positive or negative correlation between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system;

[0025] Furthermore, the expression of the correlation between the environmental parameters and the multi-sensor system is:

[0026]

[0027] Among them, H e is the actual humidity of the environment, P e is the actual ambient pressure, T e is the actual temperature of the environment, and f(H,P,T) is the actual humidity of the environment H e Relationship with the ambient humidity test value H, ambient air pressure test value P and ambient temperature test value T; g(H,P,T) is the actual ambient air pressure P e The relationship between the ambient humidity test value H, the ambient air pressure test value P and the ambient temperature test value T; h(H,P,T) is the actual ambient temperature T e Relationship with the ambient humidity test value H, ambient air pressure test value P and ambient temperature test value T.

[0028] S32, establishing a BP neural network architecture based on the association between the environmental parameters and the multi-sensor system;

[0029] Furthermore, the specific steps of establishing the BP neural network architecture based on the association between the environmental parameters and the multi-sensor system in S32 include:

[0030] S321, determining the output of the initial BP neural network architecture; building the initial BP neural network architecture based on the output and the test values ​​of each sensor in the multi-sensor system;

[0031] The expression of the initial BP neural network framework is:

[0032]

[0033] Among them, H p is the humidity prediction value output by the initial BP neural network, P p is the predicted value of air pressure output by the initial BP neural network, T p The temperature prediction value output by the initial BP neural network; represents the number of neurons in each hidden layer of the humidity prediction network, represents the humidity prediction network p 1 The weights of the hidden layer, represents the weight of the first hidden layer of the humidity prediction network, represents the humidity prediction network p 1 The threshold of the hidden layer, represents the threshold of the first hidden layer of the humidity prediction network, p 1 Represents the number of hidden layers of the humidity prediction network, m 1 represents the humidity prediction network p 1 The number of neurons in the hidden layer, r 1 Represents the number of neurons in the first hidden layer of the humidity prediction network; represents the number of neurons in each hidden layer of the air pressure prediction network, represents the pressure prediction network p 2 The weights of the hidden layer, represents the weight of the first hidden layer of the air pressure prediction network, represents the threshold of the neurons in the first hidden layer of the air pressure prediction network, represents the pressure prediction network p 2 The threshold of neurons in the hidden layer, p 2 Represents the number of hidden layers in the air pressure prediction network, m 2 represents the pressure prediction network p 2 The number of neurons in the hidden layer, r 2 They represent the number of neurons in the first hidden layer of the air pressure prediction network; represents the number of neurons in each hidden layer of the temperature prediction network, represents the temperature prediction network p 3 The weights of the hidden layer, represents the weight of the first hidden layer of the temperature prediction network, The threshold of the neurons in the first hidden layer, represents the temperature prediction network p 3 The threshold of neurons in the hidden layer, p 3 Represents the number of hidden layers of the temperature prediction network, m 3 represents the temperature prediction network p 3 The number of neurons in the hidden layer, r 3 The number of neurons in the first hidden layer of the temperature prediction network; σ is the activation function between each layer, N input is the number of neurons in the input layer of the constructed BP neural network, and (H, P, T) is the data set consisting of the measured humidity values, measured air pressure values, and measured temperature values ​​of the multi-sensor system.

[0034] S322, based on the neural network data set (H, P, T, H e ,P e ,T e ) and the loss function perform gradient descent update on the initial BP neural network architecture to obtain the BP neural network architecture;

[0035] The expression of the loss function is:

[0036]

[0037] Among them, MSE H is the mean square error of the humidity prediction value output by the initial BP neural network, MSE P is the mean square error of the air pressure prediction value output by the initial BP neural network, MSE T is the mean square error of the temperature prediction value output by the initial BP neural network; H p,o is the predicted value of humidity sample o output by the initial BP neural network, O represents the total number of humidity samples; P p,q is the predicted value of the air pressure sample q output by the initial BP neural network, Q represents the total number of air pressure samples, T p,r is the predicted value of the temperature sample r output by the initial BP neural network, and R represents the total number of temperature samples.

[0038] S4, introducing the ISSA model into the BP neural network architecture to establish an initial ISSA-BP neural network; inputting the neural network data set into the initial ISSA-BP neural network, updating the weights and thresholds of the initial ISSA-BP neural network, and obtaining an updated ISSA-BP neural network;

[0039] It can be understood that the original ISSA-BP neural network includes multiple groups of candidate weights and thresholds; each group of candidate weights and thresholds is defined as an individual, and all individuals constitute a population;

[0040] The population includes a plurality of individuals, and the individuals are the candidate weights or thresholds;

[0041] The individuals include discoverers and joiners; the discoverers and joiners include alerters and non-alerters;

[0042] It can be understood that the discoverer is an individual with low fitness; the joiner is an individual with high fitness; the vigilant is randomly selected from all individuals in the population;

[0043] In the establishment of the model, the fitness value of the population individual depends on the mean square error calculated by bringing the parameters of the population individual into the ISSA-BP neural network and using the data set. Individuals with good fitness will become discoverers, and individuals with poor fitness will become joiners.

[0044] Furthermore, the fitness of the discoverer individual is good, that is, the ISSA-BP neural network uses the weights and thresholds of the discoverer individual, and the mean square error is small; the discoverer is responsible for guiding the search area of ​​the optimal solution in the entire population, so that the population can find a better individual, that is, find the optimal weights and thresholds of the BP neural network to minimize the mean square error;

[0045] The fitness of the participant is poor, that is, the ISSA-BP neural network uses the weights and thresholds of the participant, and the mean square error is large; the participant is responsible for approaching the search area where the optimal finder is located, so as to improve the fitness of the individual and optimize the mean square error;

[0046] The role of the alerter is to alert other individuals nearby to prevent the population from falling into the local optimum in its search for the optimal solution; the non-alerts will continue the optimization and solution process of the individual as a discoverer or joiner;

[0047] Furthermore, the specific steps of obtaining the updated ISSA-BP neural network in step S4 include:

[0048] S41, initializing the weight w and threshold b of the ISSA-BP neural network to obtain an initial ISSA-BP neural network;

[0049] S42, based on the true values ​​of the environmental parameters and the initialized ISSA-BP neural network, obtaining multiple sets of candidate weights and thresholds;

[0050] The weights and thresholds of the plurality of groups of candidates are defined as individuals respectively, to obtain a plurality of groups of individuals, wherein the plurality of groups of individuals constitute a population;

[0051] Obtaining the fitness corresponding to the population; obtaining the initial position of the population based on the corresponding fitness;

[0052] S43, optimizing the initial position of the population to obtain an updated position of the population;

[0053] S44, obtaining the fitness corresponding to the updated position of the population;

[0054] S45, obtaining an update weight and an update threshold based on the fitness corresponding to the update position of the population;

[0055] The initial ISSA-BP neural network is updated based on the update weight and update threshold to obtain an updated ISSA-BP neural network.

[0056] Furthermore, the specific steps of obtaining and updating the ISSA-BP neural network include:

[0057] SS41, initialize the neural network hyperparameters, weight w and threshold b of the ISSA-BP neural network to obtain the initial ISSA-BP neural network;

[0058] SS42, based on the true values ​​of environmental parameters and the initial ISSA-BP neural network, multiple sets of candidate weights and thresholds are obtained;

[0059] The weight and threshold corresponding to each group are defined as individuals, and a plurality of individuals are obtained, and the plurality of individuals constitute a population;

[0060] Obtaining the fitness corresponding to the population; obtaining the initial position of the population based on the corresponding fitness;

[0061] Furthermore, the population includes a humidity population X, an air pressure population Y, and a temperature population Z;

[0062] The humidity population X, the air pressure population Y and the temperature population Z each include a plurality of individuals;

[0063] Furthermore, the expressions of the humidity population X, the pressure population Y and the temperature population Z are respectively:

[0064]

[0065] Among them, X g represents the weight and threshold corresponding to the individual g in the humidity population, g = 1, 2, 3, ..., G; G is the total number of individuals in the humidity population, Y n represents the weight and threshold corresponding to the individual n of the pressure population, n = 1, 2, 3...N, N is the total number of individuals in the pressure population, Z mRepresents the weight and threshold corresponding to individual m in the temperature population, m = 1, 2, 3…M, M is the total number of individuals in the temperature population.

[0066] Specifically, the specific steps of obtaining the initial position of the population based on the corresponding fitness in step S42 include:

[0067] Get the fitness corresponding to humidity population X, pressure population Y and temperature population Z respectively;

[0068] The fitness corresponding to the humidity population X, the air pressure population Y and the temperature population Z respectively include the fitness of multiple humidity individuals, the fitness of multiple air pressure individuals and the fitness of multiple temperature individuals;

[0069] The fitness F of multiple humidity individuals Hg Sort by fitness to get the initial position of the humidity population;

[0070] The fitness F of multiple pressure individuals Pn Sort by fitness to get the initial position of the air pressure population;

[0071] The individual fitness F of multiple temperatures Tm Sort by fitness to get the initial position of the temperature population;

[0072] Exemplarily, the fitness expressions corresponding to the humidity population X, the pressure population Y, and the temperature population Z are respectively:

[0073]

[0074] Among them, F H Indicates the fitness of the humidity population; F P Indicates the fitness of the air pressure population; F T Indicates the fitness of the temperature population; F Hg represents the humidity fitness of individual g in the humidity population, F Pn represents the fitness of the air pressure of individual n in the air pressure population, F Tm Represents the temperature fitness of individual m in the temperature population.

[0075] SS43, let t = 1, when t = 1, are the initial positions of the initial ISSA-BP neural network and the population respectively; the initial position t of the population includes the initial position of the discoverer individual, the initial position of the joiner individual and the initial position of the vigilant individual;

[0076] Further, the population includes a humidity population, an air pressure population or a temperature population;

[0077] SS44, obtain ISSA-BP neural network A t; Based on the ISSA-BP neural network A t Obtain the position of the discoverer individual c corresponding to the position t of the population The position of the joiner individual j And the position of the vigilant individual z

[0078] Use the edge evolution algorithm for the position of the discoverer individual c The position of the joiner individual j Optimize to obtain the updated position of the discoverer individual c The updated position of the joiner individual j

[0079] Use the adaptive step size algorithm for the position of the vigilant individual z Optimize to obtain the updated position of the vigilant individual z

[0080] Furthermore, step S44 further includes: performing boundary detection on the parameter population while updating the positions of the discoverer individual, the joiner individual, and the vigilant individual to obtain a boundary detection result; in order to prevent the situation that the position update range is too large during the position update process of the population, resulting in unstable position update and non-convergence of the training process;

[0081] Furthermore, the expression for the updated position of the discoverer individual is:

[0082]

[0083] Wherein, Represents the position of the discoverer individual c in the population at the iteration number t, Represents the position of the discoverer individual c in the population at the iteration number t + 1, c ∈ N, N represents the total number of individuals; t represents the current iteration number, α ∈ (0,1) is a random number; iter max Represents the maximum number of iterations in the training process; R 2 Is the early warning value, as a random number, R 2 ∈ [0,1]; ST is the safety threshold, ST ∈ [0.5,1]; when R 2 < ST, it means that this discoverer individual will continue to search for the optimal solution near the current position; when R 2 > ST, it means that this individual will immediately leave this point, and this movement process is random, but the discoverer with poor fitness will also be attracted by the discoverer with good fitness and move towards the latter, and this movement tendency is reflected by the evolution factor η, realizing the retention of the movement tendency of the discoverer with excellent fitness and the expansion of the movement tendency of the discoverer with poor fitness, thereby enhancing the algorithm's ability to search for the global optimal solution; η represents the evolution factor of the discoverer p 1 is the proportion of the discoverer PD; Q represents a random number that obeys a normal distribution with a mean of 0 and a variance of 1, and L represents a 1×D matrix of all 1s. represents the position of the best individual among all individuals before the position update; A represents a 1×D matrix with random elements of 1 or -1, A + =A T (AA T ) -1 .

[0084] Furthermore, the expression of the update position of the joiner individual is:

[0085]

[0086] in, represents the position of the individual j in the population at iteration number t, Indicates the position of the individual j joining the population at iteration number t+1, j∈N; Indicates the position of the worst individual among all the joined individuals in the current iteration number t, represents the position of the best individual among all the discoverers that have been updated to the t+1 iteration number; μ represents the evolution factor of the joiner, when When , it means that the fitness of the jth joiner with poor fitness is too poor and needs to move to other positions to find a solution; when When , it means that the individual belongs to the individual with better fitness among the joiners. Therefore, in addition to the joiner movement tendency of the individual itself, it is also considered to give the finder movement tendency to these individuals to a certain extent, so that these joiners with better fitness not only execute the joiner movement tendency, but also have a certain degree of finder behavior, foraging around themselves. This movement tendency is reflected by the evolution factor μ, which realizes the retention of the movement tendency of joiners with poor fitness and the expansion of the movement tendency of joiners with excellent fitness, thereby enhancing the algorithm's ability to find the global optimal solution.

[0087] Furthermore, the expression of the updated position of the individual sentinel is:

[0088]

[0089] in, represents the previous position of the vigilant individual z in the population at iteration number t, represents the position of the vigilant individual z in the population at iteration number t+1, z∈N; β∈[-1,1] is a random number that obeys the normal distribution, with a mean of 0 and a variance of 1; δ represents the adaptive step coefficient, Where t represents the current number of iterations, iter max represents the maximum number of iterations; K∈[-1,1] is a random number representing the direction of individual movement; f z represents the current fitness of the individual, f g represents the global optimal fitness value, f w Represents the global worst fitness value, ∈ is an infinitely small constant to prevent the denominator from being 0; when f z ≠f g When f , it means that the individual is at the edge of the population, and there is a certain distance from the optimal solution position in the solution space, and it needs to move closer to the current optimal solution position; when f z =f g When , it means that the individual with the best fitness in the population may fall into the local optimal state and needs to move to the position of other individuals to reduce the possibility of falling into the local optimal state. At the same time, the adaptive step coefficient δ is introduced. This coefficient is only for the case when the sentinel contains the individual with the best fitness. The adaptive step coefficient δ realizes that in the early stage of the search, the best discoverer can expand its global search range through a larger step size, and in subsequent iterations, the step size is gradually reduced by the coefficient δ to enhance its local search ability.

[0090] The boundary detection expression is:

[0091]

[0092] Among them, u is the upper boundary value; l is the lower boundary value, ub is the upper boundary value; lb is the lower boundary value.

[0093] SS45, based on the updated position of the discoverer individual c Updated position of joiner individual j and the updated position of the sentinel individual z Obtaining the updated position t+1 of the population;

[0094] SS46, obtaining the fitness corresponding to the updated position t+1 of the population, and selecting the individual corresponding to the best fitness among the fitness corresponding to the updated position t+1 of the population;

[0095] SS47, using the individual corresponding to the optimal fitness as the update weight and update threshold of the ISSA-BP neural network;

[0096] Based on the update weight and update threshold of the ISSA-BP neural network, an updated ISSA-BP neural network A is obtained. t+1 , the updated ISSA-BP neural network A t+1 Evaluated by the test set;

[0097] SS48, determine whether t+1 is greater than T, if not, set t=t+1 and return to step S44; if greater, end the training, obtain the updated position t+1 of the population and the updated ISSA-BP neural network A t+1 .

[0098] S5. Input the multi-sensor system to be tested into the updated ISSA-BP neural network, and output the predicted ambient temperature, humidity and air pressure values ​​after network compensation based on the updated weights and updated thresholds.

[0099] The present invention proposes to realize multi-sensor detection parameter error compensation based on an improved sparrow search algorithm optimized BP (ISSA-BP) neural network, which not only enables the compensation model to adapt to different input parameters, but also significantly outperforms the traditional model in accuracy, computational efficiency and adaptability.

[0100] Compared with the prior art, the present invention has at least the following beneficial effects:

[0101] (1) The multi-sensor compensation method proposed in the present invention makes up for the defect that the traditional compensation method only performs temperature compensation. The multi-sensor system is in a scene where multiple environmental factors are coupled. The sensor detection data is not only affected by temperature and produces temperature drift, but the dynamically changing humidity and air pressure in the environment will also affect the sensor, causing errors in the detection data. The technical solution of the present invention comprehensively considers the impact of multiple environmental factors on the system detection data, and at the same time compensates for the errors caused by changes in temperature, humidity, and air pressure in the system, significantly improving the accuracy of the compensation algorithm and the robustness of the sensor system.

[0102] (2) The technical solution of the present invention combines the advantages of the SSA algorithm, which has strong global and local search capabilities, fewer parameters, no need for gradient information, and high training efficiency, and the BP neural network, which has strong learning and generalization capabilities. While overcoming the slow convergence speed of traditional neural networks, it improves the accuracy of model prediction, meets the needs of multi-parameter error compensation for multi-sensor systems, and realizes high-efficiency and high-precision prediction of environmental temperature, humidity, and pressure.

[0103] (3) The ISSA-BP neural network in the technical solution of the present invention sets the population and individuals through the SSA algorithm, simulates the foraging and anti-foraging behaviors of sparrows, gives the joiner a movement tendency in the position update of the discoverer individual, and gives the discoverer a movement tendency in the position update of the joiner individual. An adaptive step size coefficient is added in the process of updating the position of the individual that serves as the optimal individual and the alert individual at the same time, which not only enhances the global optimization ability and local optimization ability of the algorithm, but also improves the training efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] The drawings are only for the purpose of illustrating particular embodiments and are not to be construed as limiting the invention.

[0105] Figure 1 A flow chart of a multi-sensor error compensation method based on ISSA-BP neural network disclosed in an embodiment of the present invention;

[0106] Figure 2 A schematic diagram of a comparison chart of error compensation results of humidity values ​​measured by multiple sensors in different environments using the method disclosed in this embodiment. DETAILED DESCRIPTION

[0107] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. In addition, the present invention can also be implemented in other ways different from those described herein, and therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0108] A specific embodiment of the present invention, as Figure 1-2 , discloses a multi-sensor error compensation method based on ISSA-BP neural network,

[0109] In order to illustrate the effectiveness of the method proposed in the present invention, the above technical solution of the present invention is described in detail below through a specific embodiment, and the specific implementation steps are as follows:

[0110] The present invention provides a multi-sensor error compensation method based on ISSA-BP neural network, comprising:

[0111] S1. Place the multi-sensor system in a test chamber, change the environmental parameters of the test chamber, and record the test values ​​of each sensor in the multi-sensor system;

[0112] S2, preprocessing the output value of each sensor in the multi-sensor system to obtain a neural network data set;

[0113] S3, establishing a correlation between the environmental parameters and the output values ​​of each sensor in the multi-sensor system in step S1; and building a BP neural network architecture based on the correlation;

[0114] S4, introducing the ISSA model into the BP neural network architecture to establish an initial ISSA-BP neural network; inputting the neural network data set obtained in step S2 into the initial ISSA-BP neural network, updating the weights and thresholds of the initial ISSA-BP neural network, and obtaining an updated ISSA-BP neural network;

[0115] S5. Acquire the data of each sensor in the multi-sensor system to be tested; input the data of each sensor into the updated ISSA-BP neural network, and output the predicted ambient temperature, humidity and air pressure values ​​after network compensation.

[0116] Preferably, step S1, placing the multi-sensor system in a test box, changing the environmental parameters of the test box; recording the standard value of the current environment and the test value of each sensor in the multi-sensor system;

[0117] Exemplarily, the output value of each sensor in the multi-sensor system includes the temperature, humidity and pressure of each sensor; the environmental parameters are temperature, humidity and pressure;

[0118] The multi-sensor system integrates multiple environmental parameter measurement sensors of temperature, humidity and pressure;

[0119] Preferably, in step S2, the output values ​​of each sensor in the multi-sensor system are preprocessed to obtain a neural network data set (H, P, T, H e ,P e ,T e );

[0120] Furthermore, we get the neural network dataset (H, P, T, H e ,P e ,T e ) The specific steps include:

[0121] The output values ​​of each sensor in the multi-sensor system and the real parameters of the environment are normalized and standardized in turn to obtain processed data; the processed data are classified by feature labels to obtain classified sensor data;

[0122] Based on the classified sensor data, the neural network data set (H, P, T, H e ,P e ,T e ), the data set is divided into a training set and a test set according to a certain ratio; where H e is the actual humidity of the environment, P e is the actual ambient pressure, T e is the actual temperature of the environment, H is the ambient humidity value measured by the multi-sensor system, P is the ambient pressure value measured by the multi-sensor system, and T is the ambient temperature value measured by the multi-sensor system;

[0123] Preferably, the specific steps of establishing the association relationship between the environmental parameters and the output values ​​of each sensor in the multi-sensor system in step S3 include:

[0124] S31, calculating the Pearson correlation coefficient between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system; judging the correlation between the environmental parameter and the multi-sensor system based on the Pearson correlation coefficient, and obtaining the correlation relationship between the environmental parameter and the multi-sensor system;

[0125] It can be understood that the relationship between the environmental parameter and the multi-sensor system is the existence or non-existence of correlation between the real value of the environmental parameter and the test value of each sensor in the multi-sensor system;

[0126] Specifically, if the Pearson correlation coefficient between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system is close to 0,

[0127] Then the correlation between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system is weak;

[0128] If the correlation between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system is close to 1 or -1, then there is a strong positive or negative correlation between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system;

[0129] Furthermore, the expression of the correlation between the environmental parameters and the multi-sensor system is:

[0130]

[0131] Among them, H e is the actual humidity of the environment, P e is the actual ambient pressure, T e is the actual temperature of the environment, and f(H,P,T) is the actual humidity of the environment H e Relationship with the ambient humidity test value H, ambient air pressure test value P and ambient temperature test value T; g(H,P,T) is the actual ambient air pressure P e The relationship between the ambient humidity test value H, the ambient air pressure test value P and the ambient temperature test value T; h(H,P,T) is the actual ambient temperature T e Relationship with the ambient humidity test value H, ambient air pressure test value P and ambient temperature test value T.

[0132] S32, establishing a BP neural network architecture based on the association between the environmental parameters and the multi-sensor system;

[0133] Furthermore, the specific steps of establishing the BP neural network architecture based on the association between the environmental parameters and the multi-sensor system in S32 include:

[0134] S321, determining the output of the initial BP neural network architecture; building the initial BP neural network architecture based on the output and the test values ​​of each sensor in the multi-sensor system;

[0135] The expression of the initial BP neural network framework is:

[0136]

[0137] Among them, H p is the humidity prediction value output by the initial BP neural network, P p is the predicted value of air pressure output by the initial BP neural network, T p The temperature prediction value output by the initial BP neural network; represents the number of neurons in each hidden layer of the humidity prediction network, represents the humidity prediction network p 1 The weights of the hidden layer, represents the weight of the first hidden layer of the humidity prediction network, represents the humidity prediction network p 1 The threshold of the hidden layer, represents the threshold of the first hidden layer of the humidity prediction network, p 1 Represents the number of hidden layers of the humidity prediction network, m 1 represents the humidity prediction network p 1 The number of neurons in the hidden layer, r 1 Represents the number of neurons in the first hidden layer of the humidity prediction network; represents the number of neurons in each hidden layer of the air pressure prediction network, represents the pressure prediction network p 2 The weights of the hidden layer, represents the weight of the first hidden layer of the air pressure prediction network, represents the threshold of the neurons in the first hidden layer of the air pressure prediction network, represents the pressure prediction network p 2 The threshold of neurons in the hidden layer, p 2 Represents the number of hidden layers in the air pressure prediction network, m 2 represents the pressure prediction network p 2 The number of neurons in the hidden layer, r 2 They represent the number of neurons in the first hidden layer of the air pressure prediction network; represents the number of neurons in each hidden layer of the temperature prediction network, represents the temperature prediction network p 3 The weights of the hidden layer, represents the weight of the first hidden layer of the temperature prediction network, The threshold of the neurons in the first hidden layer, represents the temperature prediction network p 3 The threshold of neurons in the hidden layer, p 3 Represents the number of hidden layers of the temperature prediction network, m 3 represents the temperature prediction network p 3 The number of neurons in the hidden layer, r 3 are the number of neurons in the first hidden layer of the temperature prediction network; σ is the activation function between each layer, Ninput is the number of neurons in the input layer of the constructed BP neural network, and (H, P, T) is a data set consisting of the measured humidity values, measured air pressure values, and measured temperature values ​​of the multi-sensor system.

[0138] S322, based on the neural network data set (H, P, T, H e ,P e ,T e ) and the loss function perform gradient descent update on the initial BP neural network architecture to obtain the BP neural network architecture;

[0139] The expression of the loss function is:

[0140]

[0141] Among them, MSE H is the mean square error of the humidity prediction value output by the initial BP neural network, MSE P is the mean square error of the air pressure prediction value output by the initial BP neural network, MSE T is the mean square error of the temperature prediction value output by the initial BP neural network; H p,o is the predicted value of humidity sample o output by the initial BP neural network, O represents the total number of humidity samples; P p,q is the predicted value of the air pressure sample q output by the initial BP neural network, Q represents the total number of air pressure samples, T p,r is the predicted value of the temperature sample r output by the initial BP neural network, and R represents the total number of temperature samples.

[0142] S4, introducing the ISSA model into the BP neural network architecture to establish an initial ISSA-BP neural network; inputting the neural network data set into the initial ISSA-BP neural network, updating the weights and thresholds of the initial ISSA-BP neural network, and obtaining an updated ISSA-BP neural network;

[0143] It can be understood that the original ISSA-BP neural network includes multiple groups of candidate weights and thresholds; each group of candidate weights and thresholds is defined as an individual, and all individuals constitute a population;

[0144] The population includes a plurality of individuals, and the individuals are the candidate weights or thresholds;

[0145] The individuals include discoverers and joiners; the discoverers and joiners include alerters and non-alerters;

[0146] It can be understood that the discoverer is an individual with low fitness; the joiner is an individual with high fitness; the vigilant is randomly selected from all individuals in the population;

[0147] In the establishment of the model, the fitness value of the population individual depends on the mean square error calculated by bringing the parameters of the population individual into the ISSA-BP neural network and using the data set. Individuals with good fitness will become discoverers, and individuals with poor fitness will become joiners.

[0148] Furthermore, the fitness of the discoverer individual is good, that is, the ISSA-BP neural network uses the weights and thresholds of the discoverer individual, and the mean square error is small; the discoverer is responsible for guiding the search area of ​​the optimal solution in the entire population, so that the population can find a better individual, that is, find the optimal weights and thresholds of the BP neural network to minimize the mean square error;

[0149] The fitness of the participant is poor, that is, the ISSA-BP neural network uses the weights and thresholds of the participant, and the mean square error is large; the participant is responsible for approaching the search area where the optimal finder is located, so as to improve the fitness of the individual and optimize the mean square error;

[0150] The role of the alerter is to alert other individuals nearby to prevent the population from falling into the local optimum in its search for the optimal solution; the non-alerts will continue the optimization and solution process of the individual as a discoverer or joiner;

[0151] Furthermore, the specific steps of obtaining the updated ISSA-BP neural network in step S4 include:

[0152] S41, initializing the weight w and threshold b of the ISSA-BP neural network to obtain an initial ISSA-BP neural network;

[0153] S42, based on the true values ​​of the environmental parameters and the initialized ISSA-BP neural network, obtaining multiple sets of candidate weights and thresholds;

[0154] The weights and thresholds of the plurality of groups of candidates are defined as individuals respectively, to obtain a plurality of groups of individuals, wherein the plurality of groups of individuals constitute a population;

[0155] Obtaining the fitness corresponding to the population; obtaining the initial position of the population based on the corresponding fitness;

[0156] S43, optimizing the initial position of the population to obtain an updated position of the population;

[0157] S44, obtaining the fitness corresponding to the updated position of the population;

[0158] S45, obtaining an update weight and an update threshold based on the fitness corresponding to the update position of the population;

[0159] The initial ISSA-BP neural network is updated based on the update weight and update threshold to obtain an updated ISSA-BP neural network.

[0160] Furthermore, the specific steps of obtaining and updating the ISSA-BP neural network include:

[0161] SS41, initialize the neural network hyperparameters, weight w and threshold b of the ISSA-BP neural network to obtain the initial ISSA-BP neural network;

[0162] SS42, based on the true values ​​of environmental parameters and the initial ISSA-BP neural network, multiple sets of candidate weights and thresholds are obtained;

[0163] The weight and threshold corresponding to each group are defined as individuals, and a plurality of individuals are obtained, and the plurality of individuals constitute a population;

[0164] Obtaining the fitness corresponding to the population; obtaining the initial position of the population based on the corresponding fitness;

[0165] Furthermore, the population includes a humidity population X, an air pressure population Y, and a temperature population Z;

[0166] The humidity population X, the air pressure population Y and the temperature population Z each include a plurality of individuals;

[0167] Furthermore, the expressions of the humidity population X, the pressure population Y and the temperature population Z are respectively:

[0168]

[0169] Among them, X g represents the weight and threshold corresponding to the individual g in the humidity population, g = 1, 2, 3, ..., G; G is the total number of individuals in the humidity population, Y n represents the weight and threshold corresponding to the individual n of the pressure population, n = 1, 2, 3...N, N is the total number of individuals in the pressure population, Z m Represents the weight and threshold corresponding to individual m in the temperature population, m = 1, 2, 3...M, M is the total number of individuals in the temperature population.

[0170] Specifically, the specific steps of obtaining the initial position of the population based on the corresponding fitness in step S42 include:

[0171] Get the fitness corresponding to humidity population X, pressure population Y and temperature population Z respectively;

[0172] The fitness corresponding to the humidity population X, the air pressure population Y and the temperature population Z respectively include the fitness of multiple humidity individuals, the fitness of multiple air pressure individuals and the fitness of multiple temperature individuals;

[0173] The fitness F of multiple humidity individuals Hg Sort by fitness to get the initial position of the humidity population;

[0174] The fitness F of multiple pressure individuals Pn Sort by fitness to get the initial position of the air pressure population;

[0175] The individual fitness F of multiple temperatures Tm Sort by fitness to get the initial position of the temperature population;

[0176] Exemplarily, the fitness expressions corresponding to the humidity population X, the pressure population Y, and the temperature population Z are respectively:

[0177]

[0178] Among them, F H Indicates the fitness of the humidity population; F P Indicates the fitness of the air pressure population; F T Indicates the fitness of the temperature population; F Hg represents the humidity fitness of individual g in the humidity population, F Pn represents the fitness of the air pressure of individual n in the air pressure population, F Tm Represents the temperature fitness of individual m in the temperature population.

[0179] SS43, let t = 1, when t = 1, are the initial positions of the initial ISSA-BP neural network and the population respectively; the initial position t of the population includes the initial position of the discoverer individual, the initial position of the joiner individual and the initial position of the vigilant individual;

[0180] Further, the population includes a humidity population, an air pressure population or a temperature population;

[0181] SS44, obtain ISSA-BP neural network A t Based on the ISSA-BP neural network A t Get the position of the discoverer individual c corresponding to the population position t The position of the participant j and the individual z position of the sentinel

[0182] Use the edge evolution algorithm for the position of the discoverer individual c The position of the joiner individual j Optimize to obtain the updated position of the discoverer individual c The updated position of the joiner individual j

[0183] Use the adaptive step size algorithm for the position of the vigilant individual z Optimize to obtain the updated position of the vigilant individual z

[0184] Furthermore, step S44 also includes: performing boundary detection on the parameter population while updating the positions of the discoverer individual, the joiner individual, and the vigilant individual to obtain a boundary detection result; in order to prevent the situation that the position update range of the population is too large during the position update process, resulting in unstable position updates and the inability to converge during the training process;

[0185] Furthermore, the expression for the updated position of the discoverer individual is:

[0186]

[0187] Wherein, represents the position of the discoverer individual c in the population at iteration number t, represents the position of the discoverer individual c in the population at iteration number t + 1, c ∈ N, N represents the total number of individuals; t represents the current iteration number, α ∈ (0, 1) is a random number; iter max represents the maximum number of iterations in the training process; R 2 is the warning value, as a random number, R 2 ∈ [0, 1]; ST is the safety threshold, ST ∈ [0.5, 1]; when R 2 < ST, it means that this discoverer individual will continue to search for the optimal solution near the current position; when R 2 > ST, it means that this individual will immediately leave this point, and this movement process is random, but the discoverer with a poorer fitness will also be attracted by the discoverer with a better fitness and move towards the latter, and this movement tendency is reflected by the evolution factor η, realizing the retention of the movement tendency of the discoverer with excellent fitness and the expansion of the movement tendency of the discoverer with poorer fitness, thereby enhancing the algorithm's ability to search for the global optimal solution; η represents the evolution factor of the discoverer, p 1 is the proportion of the discoverer PD; Q represents a random number that follows a normal distribution with a mean of 0 and a variance of 1, and L represents a 1×D all-1 matrix, represents the position of the best individual among all individuals before the position update; A represents a 1×D matrix with random elements of 1 or -1, A + =A T (AA T ) -1 .

[0188] Furthermore, the expression of the update position of the joiner individual is:

[0189]

[0190] in, represents the position of the individual j in the population at iteration number t, Indicates the position of the individual j joining the population at iteration number t+1, j∈N; Indicates the position of the worst individual among all the joined individuals in the current iteration number t, represents the position of the best individual among all the discoverers that have been updated to the t+1 iteration number; μ represents the evolution factor of the joiner, when When , it means that the fitness of the jth joiner with poor fitness is too poor and needs to move to other positions to find a solution; when When , it means that the individual belongs to the individual with better fitness among the joiners. Therefore, in addition to the joiner movement tendency of the individual itself, it is also considered to give the finder movement tendency to these individuals to a certain extent, so that these joiners with better fitness not only execute the joiner movement tendency, but also have a certain degree of finder behavior, foraging around themselves. This movement tendency is reflected by the evolution factor μ, which realizes the retention of the movement tendency of joiners with poor fitness and the expansion of the movement tendency of joiners with excellent fitness, thereby enhancing the algorithm's ability to find the global optimal solution.

[0191] Furthermore, the expression of the updated position of the individual sentinel is:

[0192]

[0193] in, represents the previous position of the vigilant individual z in the population at iteration number t, represents the position of the vigilant individual z in the population at iteration number t+1, z∈N; β∈[-1,1] is a random number that obeys the normal distribution, with a mean of 0 and a variance of 1; δ represents the adaptive step coefficient, Where t represents the current number of iterations, iter max represents the maximum number of iterations; K∈[-1,1] is a random number representing the direction of individual movement; f z represents the current fitness of the individual, fg represents the global optimal fitness value, f w Represents the global worst fitness value, ∈ is an infinitely small constant to prevent the denominator from being 0; when f z ≠f g When f , it means that the individual is at the edge of the population, and there is a certain distance from the optimal solution position in the solution space, and it needs to move closer to the current optimal solution position; when f z =f g When , it means that the individual with the best fitness in the population may fall into the local optimal state and needs to move to the position of other individuals to reduce the possibility of falling into the local optimal state. At the same time, the adaptive step coefficient δ is introduced. This coefficient is only for the case when the sentinel contains the individual with the best fitness. The adaptive step coefficient δ realizes that in the early stage of the search, the best discoverer can expand its global search range through a larger step size, and in subsequent iterations, the step size is gradually reduced by the coefficient δ to enhance its local search ability.

[0194] The boundary detection expression is:

[0195]

[0196] Among them, u is the upper boundary value; l is the lower boundary value, ub is the upper boundary value; lb is the lower boundary value.

[0197] SS45, based on the updated position of the discoverer individual c Updated position of joiner individual j and the updated position of the sentinel individual z Obtaining the updated position t+1 of the population;

[0198] SS46, obtaining the fitness corresponding to the updated position t+1 of the population, and selecting the individual corresponding to the best fitness among the fitness corresponding to the updated position t+1 of the population;

[0199] SS47, using the individual corresponding to the optimal fitness as the update weight and update threshold of the ISSA-BP neural network;

[0200] Based on the update weight and update threshold of the ISSA-BP neural network, an updated ISSA-BP neural network A is obtained. t+1 , the updated ISSA-BP neural network A t+1 Evaluated by the test set;

[0201] SS48, determine whether t+1 is greater than T, if not, set t=t+1 and return to step S44; if greater, end the training, obtain the updated position t+1 of the population and the updated ISSA-BP neural network A t+1 .

[0202] S5. Input the multi-sensor system to be tested into the updated ISSA-BP neural network, and output the predicted ambient temperature, humidity and air pressure values ​​after network compensation based on the updated weights and updated thresholds.

[0203] The present invention proposes to realize multi-sensor detection parameter error compensation based on an improved sparrow search algorithm optimized BP (ISSA-BP) neural network, which not only enables the compensation model to adapt to different input parameters, but also significantly outperforms the traditional model in accuracy, computational efficiency and adaptability.

[0204] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-sensor error compensation method based on ISSA-BP neural network, characterized in that: include: S1. Place the multi-sensor system in a test chamber and change the environmental parameters of the test chamber; Record the test values ​​of each sensor in the multi-sensor system; S2, preprocessing the output value of each sensor in the multi-sensor system to obtain a neural network data set; the output value of each sensor in the multi-sensor system includes the temperature, humidity and pressure of each sensor; S3, establishing a correlation between the environmental parameters and the output values ​​of each sensor in the multi-sensor system in step S1; and building a BP neural network architecture based on the correlation; S4, introducing the ISSA model into the BP neural network architecture to establish an initial ISSA-BP neural network; inputting the neural network data set obtained in step S2 into the initial ISSA-BP neural network, updating the weights and thresholds of the initial ISSA-BP neural network, and obtaining an updated ISSA-BP neural network; S5, acquiring data of each sensor in the multi-sensor system to be tested; inputting the data of each sensor into the updated ISSA-BP neural network, and outputting a predicted value after network compensation.

2. The multi-sensor error compensation method according to claim 1, characterized in that: The specific steps of building a BP neural network architecture based on the association relationship in step S3 include: S31, calculating the Pearson correlation coefficient between the true value of the environmental parameter and the test value of each sensor in the multi-sensor system; judging the correlation between the environmental parameter and the multi-sensor system based on the Pearson correlation coefficient, and obtaining the correlation relationship between the environmental parameter and the multi-sensor system; S32: Establish a BP neural network architecture based on the correlation between the environmental parameters and the multi-sensor system.

3. The multi-sensor error compensation method according to claim 2, characterized in that: The expression of the correlation between the environmental parameters and the multi-sensor system is: Among them, H e is the actual humidity of the environment, P e is the actual ambient pressure, T e is the actual temperature of the environment, and f(H,P,T) is the actual humidity of the environment H e Relationship with the ambient humidity test value H, ambient air pressure test value P and ambient temperature test value T; g(H,P,T) is the actual ambient air pressure P e The relationship between the ambient humidity test value H, the ambient air pressure test value P and the ambient temperature test value T; h(H,P,T) is the actual ambient temperature T e Relationship with the ambient humidity test value H, ambient air pressure test value P and ambient temperature test value T.

4. The multi-sensor error compensation method according to claim 3, characterized in that: The specific steps of establishing the BP neural network architecture based on the association between the environmental parameters and the multi-sensor system in S32 include: S321, determining the output of the initial BP neural network architecture; building the initial BP neural network architecture based on the output and the test values ​​of each sensor in the multi-sensor system; S322, performing gradient descent update on the initial BP neural network architecture based on the neural network data set and the loss function to obtain the BP neural network architecture.

5. The multi-sensor error compensation method according to claim 4, characterized in that: The expression of the loss function is: Among them, MSE H is the mean square error of the humidity prediction value output by the initial BP neural network, MSE P is the mean square error of the air pressure prediction value output by the initial BP neural network, MSE T is the mean square error of the temperature prediction value output by the initial BP neural network; H p,o is the predicted value of the humidity sample output by the initial BP neural network, O represents the total number of humidity samples; P p,q is the predicted value of the air pressure sample output by the initial BP neural network, Q represents the total number of air pressure samples, T p,r is the predicted value of the temperature sample output by the initial BP neural network, and R represents the total number of temperature samples.

6. The multi-sensor error compensation method according to claim 5, characterized in that: The specific steps of updating the weights and thresholds of the ISSA-BP neural network in step S4 include: S41, initializing the weight w and threshold b of the ISSA-BP neural network to obtain an initial ISSA-BP neural network; S42, based on the true values ​​of the environmental parameters and the initialized ISSA-BP neural network, obtaining multiple sets of candidate weights and thresholds; The weights and thresholds of the plurality of groups of candidates are defined as individuals respectively, to obtain a plurality of groups of individuals, wherein the plurality of groups of individuals constitute a population; S43, optimizing the initial position of the population to obtain an updated position of the population; S44, obtaining the fitness corresponding to the updated position of the population; S45, obtaining an update weight and an update threshold based on the fitness corresponding to the update position of the population; The initial ISSA-BP neural network is updated based on the update weight and update threshold to obtain an updated ISSA-BP neural network.

7. The multi-sensor error compensation method according to claim 6, characterized in that: The fitness corresponding to the population includes the fitness corresponding to the humidity population X, the air pressure population Y and the temperature population Z respectively; The fitness expressions corresponding to humidity population X, pressure population Y and temperature population Z are: Among them, F H Indicates the fitness of the humidity population; F P Indicates the fitness of the air pressure population; F T Indicates the fitness of the temperature population; F Hg represents the humidity fitness of individual g in the humidity population; F Pn represents the fitness of the air pressure of individual n in the air pressure population; F Tm Represents the temperature fitness of individual m in the temperature population.

8. The multi-sensor error compensation method according to claim 7, characterized in that: The specific steps of obtaining and updating the ISSA-BP neural network in step S4 include: SS41, initialize the neural network hyperparameters, weights and thresholds of the ISSA-BP neural network to obtain the initialized ISSA-BP neural network; SS42, based on the true values ​​of environmental parameters and the initialized ISSA-BP neural network, multiple sets of candidate weights and thresholds are obtained; The weights and thresholds of the plurality of groups of candidates are defined as individuals respectively, to obtain a plurality of groups of individuals, wherein the plurality of groups of individuals constitute a population; Obtaining the fitness corresponding to the population; obtaining the initial position of the population based on the corresponding fitness; SS43, let t = 1, when t = 1, are the initial positions of the initial ISSA-BP neural network and the population respectively; the initial position t of the population includes the initial position of the discoverer individual, the initial position of the joiner individual and the initial position of the vigilant individual; SS44, obtain ISSA-BP neural network A t Based on the ISSA-BP neural network A t Get the position of the discoverer individual c corresponding to the population position t The position of the participant j and the individual z position of the sentinel For the finder individual c position The position of the participant j and the individual z position of the sentinel Optimize and get the updated position of the discoverer individual c Updated position of participant j and the updated position of the sentinel individual z SS45, based on the updated position of the discoverer individual c Updated position of participant j and the updated position of the sentinel individual z Obtaining the updated position t+1 of the population; SS46. Obtain the fitness corresponding to the updated position t+1 of the population; SS47, obtaining an update weight and an update threshold of the ISSA-BP neural network based on the update position t+1 of the population; Based on the update weight and update threshold of the ISSA-BP neural network, an updated ISSA-BP neural network A is obtained. t+1 , the updated ISSA-BP neural network A t+1 Evaluated by the test set; SS48, determine whether t+1 is greater than T, if not, set t=t+1 and return to step SS44; if greater, end the training, obtain the updated position t+1 of the population and the updated ISSA-BP neural network A t+1 .

9. The multi-sensor error compensation method according to claim 8, characterized in that: The expression of the updated position of the discoverer individual is: in, represents the position of the discoverer individual c in the population at the iteration number t, It represents the position of the discoverer individual c in the population at iteration number t+1, c∈N, N represents the total number of individuals; t represents the current iteration number, α is a random number; iter max represents the maximum number of iterations of the training process; R2 is the warning value; ST is the safety threshold; η represents the evolution factor of the discoverer; Q represents a random number that obeys a normal distribution with a mean of 0 and a variance of 1, and L represents a 1×D matrix of all 1s. represents the position of the best individual among all individuals before the position update; A represents a 1×D matrix with random elements of 1 or -1, A + =A T (AA T ) -1 .

Images