Global fusion method of temperature data based on PICI-IGWCA-ELM model

Abstract

The present invention discloses a global temperature data fusion method based on the PICI-IGWCA-ELM model. The method comprises setting up a smart greenhouse multi-sensor system, including edge sensing nodes and internal sensing nodes, wherein the edge sensing nodes contain a large amount of independent information, and the internal sensing nodes have a large common sensing area and high data redundancy; performing pairwise parallel fusion processing on the internal sensing nodes, and using the PICI model to calculate the fusion estimation value of each pair of internal sensing nodes and their corresponding covariance matrix; after the local fusion of all internal sensing nodes is completed, globally fusing the data of the edge sensing nodes with the local fusion results of all internal sensing nodes. The present invention has the characteristics of high fusion rate and stable performance.

Description

Technical Field

[0001] The present invention relates to the field of smart agriculture technology, and specifically to a global fusion method for temperature data based on a PICI-IGWCA-ELM model. Background Art

[0002] Multi-source data fusion technology, one of the core technologies in modern information processing, has, through in-depth exploration and widespread application, significantly improved system performance and decision-making quality in a variety of fields, including military, security, environment, transportation, healthcare, and commerce. By efficiently integrating data from multiple sensors, databases, and information systems, this technology aims to enhance system reliability, adaptability, detection range, and information comprehensiveness, while reducing the system's reliance on a single data source and improving overall robustness. However, despite its strong potential, multi-source data fusion technology still faces many challenges in practical application, particularly when processing temperature data in distributed systems.

[0003] Specifically, temperature data fusion in distributed systems is often limited by the difficulty in accurately solving the cross-covariance matrix between adjacent sensors, which directly affects the accuracy and real-time performance of the fusion results. When dealing with such problems, traditional methods often lack effective means to deal with the situation where the cross-covariance matrix is ​​unknown, resulting in a decrease in fusion accuracy or a significant delay in the fusion process. In addition, when fusing multi-source temperature data, traditional algorithms often have difficulties achieving ideal fusion effects due to the limitations of the algorithms themselves, such as unstable fusion accuracy caused by random assignment of initial parameters. Therefore, there is an urgent need for an innovative multi-source data fusion technology that can achieve high-precision fusion when the cross-covariance matrix is ​​unknown, while improving the fusion rate and solving the problems of low accuracy and unstable performance of traditional algorithms when fusing multi-source temperature data. Summary of the Invention

[0004] The object of the present invention is to provide a global fusion method of temperature data based on the PICI-IGWCA-ELM model to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solution: a global fusion method of temperature data based on the PICI-IGWCA-ELM model, comprising:

[0006] A smart greenhouse multi-sensor system is set up, including edge sensing nodes and internal sensing nodes. The edge sensing nodes contain a large amount of independent information, while the internal sensing nodes have a large common sensing area and high data redundancy.

[0007] Perform pairwise parallel fusion processing on internal perception nodes, and use the PICI model to calculate the fusion estimation value of each pair of internal perception nodes and their corresponding covariance matrix;

[0008] After the local fusion of all internal sensing nodes is completed, the data of the edge sensing nodes are globally fused with the local fusion results of all internal sensing nodes;

[0009] Then, based on the improved Great Wall construction algorithm, the extreme learning machine model is optimized and the labor population is initialized with chaos opposition.

[0010] Adaptive T-distribution mutation operator is used to perturb the individual positions of workers;

[0011] Optimize the global position of workers through vertical and horizontal cross-cutting strategies;

[0012] The global optimization capability of the IGWCA algorithm was verified using the IEEE CEC 2005 benchmark suite, and the IGWCA-ELM model was established based on the verification results.

[0013] The IGWCA algorithm is used to optimize the initial input weights and hidden layer bias of the ELM model.

[0014] According to the above technical solution, the fusion processing method includes:

[0015] A. Distributed fusion with known cross-covariance:

[0016] (1) Dual sensor fusion algorithm

[0017] When there are two sensors (X1, P1) and (X2, P2), the cross-covariance matrix P 12 Known, the Bar-Shalom algorithm is:

[0018] X BC =K1X1+K2X2,

[0019] X BC The corresponding fusion variance matrix P BC for:

[0020] P BC =P1-(P1-P 12 )(P1+P2-P 12 -P 21 ) -1 (P1-P 12 ) T ,in

[0021] The fusion coefficient matrix is:

[0022] K1=(P2-P 21 )(P1+P2-P 12 -P 21 ) -1 ,

[0023] K2=(P1-P 12 )(P1+P2-P 21 -P 12 ) -1 ,

[0024] The accuracy relationship satisfies:

[0025] P BC ≤P1 and P BC ≤P2,

[0026] trace(P BC )≤trace(P1) and trace(P BC )≤trace(P2).

[0027] (2) Multi-sensor fusion algorithm

[0028] In the smart greenhouse multi-source system, suppose there are n-source perception data (X1, P1), (X2, P2), ..., (X n ,P n ), and its corresponding cross-covariance P ij are all known precisely, where i, j = 1, 2, ..., n; when the local node variance matrix P i When both are known, the fusion formula of the state mean and covariance matrix is:

[0029]

[0030] The variables in the above two formulas are:

[0031] X=[X N1 ,…,X Nn ] T ; M=[I N1 ,…,I Nn ] T ;

[0032] Where n represents the number of observers and N represents the dimension of the state vector;

[0033] B. Distributed fusion of unknown cross-covariance:

[0034] (1) Covariance cross fusion

[0035] Assume there are two sensors in the system. are two unbiased estimates of the state variable X, namely:

[0036]

[0037] Where n = 1, 2 and the corresponding variance matrices are P1 and P2 respectively; when and The cross-covariance matrix P 12 When unknown, the CI fusion algorithm is as follows:

[0038]

[0039] in is the CI fusion output; P CI is the fusion variance matrix; the fusion coefficients ω1 and ω2 satisfy 0≤ω1,ω2≤1,ω1+ω2=1; the specific solution formulas for ω1 and ω2 are:

[0040]

[0041] When the number of system sensors is greater than or equal to three, the CI fusion formula is:

[0042]

[0043] in is the fusion output; P CI is the fusion variance; ω i is the fusion coefficient, i=1,2,…,n, 0≤ω i ≤1,ω1+…+ω n =1;ω i The specific solution formula is:

[0044]

[0045] (2) Inverse covariance cross fusion

[0046] Assume there are two sensors in the system and Then the ICI fusion output is:

[0047]

[0048] in is the ICI fusion output; P ICI is the fusion variance matrix; fusion coefficient ω∈[0,1]; K f With L f is the fusion coefficient matrix, which satisfies K f +L f =I, specifically defined as:

[0049]

[0050] The fusion coefficient ω can be obtained by minimizing P ICI The trace or determinant is used to solve, that is:

[0051]

[0052] According to the above technical solution, the method for initializing the chaotic opposition of the labor population is:

[0053] Chaos Tent Mapping:

[0054] The Tent mapping has a fast iteration speed and the chaotic sequence is evenly distributed in [0,1]. Its expression is as follows:

[0055]

[0056] where λ t is the chaotic number of the t-th iteration; t=0,1,…,T, where T is the maximum number of iterations; α is a constant in [0,1];

[0057] Adversarial Learning:

[0058] Assume that the position of a d-dimensional individual is:

[0059] Δ=(x1,x2,…,x d ),

[0060] Assuming that the lower and upper bounds of its individual position are lb and ub respectively, its opposite side position can be expressed as:

[0061]

[0062] When the upper and lower bounds of the objective function are symmetrical, the inverse solution will be an exact mirror image of the original solution. Considering the dual function, this mirror image solution has the same value as the initial objective solution, which makes it difficult to sort the two populations based on fitness, making it difficult to effectively improve the overall quality of the population.

[0063] Labor initialization strategy integrating chaotic tent mapping and adversarial learning:

[0064] Combining chaotic tent mapping with OBL, the specific mathematical model is:

[0065]

[0066] Where i=1,2,…,n, is the j-th dimension component of the opposing position of the i-th worker, is the dot product operator.

[0067] This strategy is equivalent to using the Tent mapping to uniformly vary the sum of the upper and lower bounds of the objective function, thereby dynamically compressing the original initial worker positions. Assuming the number of workers is n, the specific steps of this strategy are: first, generate n worker positions through random distribution; second, generate n chaotic opposing positions using an improved worker initialization strategy; finally, sort these 2n positions by fitness, and select the top n positions with the best fitness as the initial workers.

[0068] According to the above technical solution, the specific method of using the adaptive T-distribution mutation operator to perturb the individual positions of workers is:

[0069] The specific formula of the adaptive T-distribution mutation operator that perturbs the current optimal worker position is:

[0070]

[0071] where x new is the latest worker position after being perturbed by the adaptive T-distribution mutation operator; is the global optimal solution of the current iteration; iter is the current number of iterations, MaxIter is the total number of iterations; P iter Generate a random probability value for the current iteration. If it is greater than 0.8, perform adaptive T-distribution mutation perturbation. Otherwise, use the initial optimal solution. is the dot product operator; t(k) represents the probability density function of the Student's T distribution with k degrees of freedom.

[0072] According to the above technical solution, the specific method of optimizing the global position of workers through the vertical and horizontal cross strategy is:

[0073] Horizontal cross:

[0074] The specific calculation formula is:

[0075]

[0076] in and They are and The d-dimensional labor after horizontal cross processing; r1 and r2 are random numbers in [0,1]; c1 and c2 are random numbers in [-1,1];

[0077] Vertical cross:

[0078]

[0079] in For workers The offspring workers are generated by vertical crossover; r is a random number in [0,1].

[0080] According to the above technical solution, the establishment process of the IGWCA-ELM model specifically includes:

[0081] Step A1: Set the initial parameters of the Great Wall construction algorithm and ELM, such as the initial population size, the number of hidden neural nodes, and the number of iterations;

[0082] Step A2: Build the IGWCA optimized ELM network framework and form the corresponding initial labor based on the ELM input weights and hidden layer bias;

[0083] Step A3: Calculate the fitness of the workers and determine the initial optimal workers, i.e. the emperor agent, based on the fitness;

[0084] Step A4: Continuously iterate to update the global optimal position of the workers;

[0085] Step A5: When the number of iterations is reached, stop the optimization, otherwise return to step A4;

[0086] Step A6: Use the IGWCA global optimal output as the optimal input weight and hidden layer bias of ELM, and train the optimal ELM model based on the dataset to output the global optimal prediction.

[0087] Compared with the prior art, the present invention has the following beneficial effects:

[0088] (1) By introducing the Parallel Inverse Covariance Intersection (PICI) model, it is possible to achieve local multi-source fusion of temperature data in a distributed fusion environment, even when the cross-covariance matrix is ​​unknown, to achieve optimal fusion accuracy, and effectively reduce the shortcomings of traditional methods in dealing with the problem of solving the cross-covariance matrix between adjacent sensors. The parallel fusion architecture of the PICI model also enables the algorithm to flexibly respond to changes in sensor order, thereby further improving the fusion rate and system adaptability. Simulation results also show that the PICI model effectively solves the fusion delay problem existing in traditional models, ensuring the real-time and accuracy of temperature data;

[0089] (2) Through chaotic opposition initialization, adaptive T-distribution mutation operator and vertical and horizontal cross strategies, the global optimization ability and fusion accuracy are significantly improved; by optimizing the initial input weights and hidden layer biases of ELM, the problems of poor fusion accuracy and unstable prediction performance caused by random assignment of traditional ELM models are overcome, making the model show higher stability and reliability in the global multi-source fusion task of temperature data in complex environments such as smart greenhouses. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0091] Figure 1 Schematic diagram of a global fusion method for temperature data based on the PICI-IGWCA-ELM model according to an embodiment of the present invention

[0092] Figure 2 This is a schematic diagram of the IGWCA-ELM process according to an embodiment of the present invention;

[0093] Figure 3 Schematic diagram of geometric comparison of covariance ellipses according to an embodiment of the present invention;

[0094] Figure 4 Schematic diagram of the F1 convergence curve of an embodiment of the present invention;

[0095] Figure 5 Schematic diagram of the F2 convergence curve of an embodiment of the present invention;

[0096] Figure 6 This is a schematic diagram of the F3 convergence curve of an embodiment of the present invention;

[0097] Figure 7 Schematic diagram of the F4 convergence curve of an embodiment of the present invention;

[0098] Figure 8 Schematic diagram of the F5 convergence curve of an embodiment of the present invention;

[0099] Figure 9 Schematic diagram of the F6 convergence curve of an embodiment of the present invention;

[0100] Figure 10 Schematic diagram of the F7 convergence curve of an embodiment of the present invention;

[0101] Figure 11 Schematic diagram of the F8 convergence curve of an embodiment of the present invention;

[0102] Figure 12 Schematic diagram of the F9 convergence curve of an embodiment of the present invention;

[0103] Figure 13 Schematic diagram of the F10 convergence curve of an embodiment of the present invention;

[0104] Figure 14 This is a schematic diagram of the F11 convergence curve of an embodiment of the present invention;

[0105] Figure 15 Schematic diagram of the F12 convergence curve of an embodiment of the present invention;

[0106] Figure 16 Schematic diagram of the F13 convergence curve of an embodiment of the present invention;

[0107] Figure 17Schematic diagram of the F14 convergence curve of an embodiment of the present invention;

[0108] Figure 18 Schematic diagram of the F15 convergence curve of an embodiment of the present invention;

[0109] Figure 19 Schematic diagram of the F16 convergence curve of an embodiment of the present invention;

[0110] Figure 20 Schematic diagram of the F17 convergence curve of an embodiment of the present invention;

[0111] Figure 21 Schematic diagram of the F18 convergence curve of an embodiment of the present invention;

[0112] Figure 22 Schematic diagram of the F19 convergence curve of an embodiment of the present invention;

[0113] Figure 23 Schematic diagram of the F20 convergence curve of an embodiment of the present invention;

[0114] Figure 24 Schematic diagram of the F21 convergence curve of an embodiment of the present invention;

[0115] Figure 25 Schematic diagram of the F22 convergence curve of an embodiment of the present invention;

[0116] Figure 26 Schematic diagram of the F23 convergence curve of an embodiment of the present invention;

[0117] Figure 27 This is a comparison chart of the global fusion results of various models in the embodiment of the present invention;

[0118] Figure 28 This is a comparison chart of the global fusion errors of various models in the embodiment of the present invention;

[0119] Figure 29 This is a comparison chart of ELM global fusion results in an embodiment of the present invention;

[0120] Figure 30 is an ELM global fusion error graph according to an embodiment of the present invention;

[0121] Figure 31 This is a comparison diagram of the GWO-ELM global fusion results of an embodiment of the present invention;

[0122] Figure 32 : This is a GWO-ELM global fusion error graph according to an embodiment of the present invention;

[0123] Figure 33 This is a comparison chart of WOA-ELM global fusion results in an embodiment of the present invention;

[0124] Figure 34 : This is a WOA-ELM global fusion error graph according to an embodiment of the present invention;

[0125] Figure 35 This is a comparison chart of the SSA-ELM global fusion results of an embodiment of the present invention;

[0126] Figure 36 This is an SSA-ELM global fusion error graph according to an embodiment of the present invention;

[0127] Figure 37 This is a comparison chart of DBO-ELM global fusion results in an embodiment of the present invention;

[0128] Figure 38 This is a DBO-ELM global fusion error graph of an embodiment of the present invention;

[0129] Figure 39 This is a comparison chart of the SCHO-ELM global fusion results of an embodiment of the present invention;

[0130] Figure 40 SCHO-ELM global fusion error map of an embodiment of the present invention;

[0131] Figure 41 This is a comparison chart of the IGWCA-ELM global fusion results of an embodiment of the present invention;

[0132] Figure 42 This is an IGWCA-ELM global fusion error diagram of an embodiment of the present invention;

[0133] Figure 43 This is a bar chart comparing evaluation indicators of various models in an embodiment of the present invention;

[0134] Figure 44 This is a comparison chart of evaluation index instruments for each model in an embodiment of the present invention;

[0135] Figure 45 This is a radar comparison chart of evaluation indicators of various models in an embodiment of the present invention. DETAILED DESCRIPTION

[0136] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0137] See also Figure 1 The present invention provides a technical solution: a global fusion method of temperature data based on the PICI-IGWCA-ELM model, comprising:

[0138] S101. Setting up a smart greenhouse multi-sensor system, including edge sensing nodes and internal sensing nodes, wherein the edge sensing nodes contain a large amount of independent information, and the internal sensing nodes have a large common sensing area and high data redundancy;

[0139] S102, performing a pairwise parallel fusion process on the internal perception nodes, using the PICI model to calculate the fusion estimation value of each pair of internal perception nodes and its corresponding covariance matrix, so as to reduce data redundancy and improve the fusion rate while maintaining a high fusion accuracy;

[0140] S103. After the local fusion of all internal sensing nodes is completed, the data of the edge sensing nodes are globally fused with the local fusion results of all internal sensing nodes to preserve the independent information of the edge sensing data and integrate the observation data of all sensors. Therefore, in the PICI model, the internal sensing nodes are fused in parallel in pairs, thereby achieving the purpose of improving the fusion rate and reducing the computing resource usage without losing the fusion accuracy.

[0141] S104, then optimizing the extreme learning machine model based on the improved Great Wall construction algorithm, and performing chaotic opposition initialization on the labor population;

[0142] S105, using an adaptive T-distribution mutation operator to perturb the individual positions of workers;

[0143] S106, optimize the overall position of labor through vertical and horizontal cross-strategies;

[0144] S107. The global optimization capability of the IGWCA algorithm was verified using the IEEE CEC 2005 benchmark suite, and an IGWCA-ELM model was established based on the verification results. Simulation results show that the IGWCA-ELM model can effectively overcome the shortcomings of the traditional ELM model, such as poor fusion accuracy and unstable prediction performance caused by random assignment of initial input weights and hidden layer biases. Therefore, this model is beneficial for solving the global multi-source fusion problem of temperature data in smart greenhouses.

[0145] S108. Use the IGWCA algorithm to optimize the initial input weights and hidden layer biases of the ELM model; to overcome the problems of poor fusion accuracy and unstable prediction performance caused by random assignment in the traditional ELM model.

[0146] Fusion processing methods include:

[0147] A. Distributed fusion with known cross-covariance:

[0148] (1) Dual sensor fusion algorithm

[0149] When there are two sensors (X1, P1) and (X2, P2), the cross-covariance matrix P12 Known, the Bar-Shalom algorithm is:

[0150] X BC =K1X1+K2X2,

[0151] X BC The corresponding fusion variance matrix P BC for:

[0152] P BC =P1-(P1-P 12 )(P1+P2-P 12 -P 21 ) -1 (P1-P 12 ) T ,

[0153] in At the same time BC It can also be written as:

[0154] P BC =P2-(P2-P 21 )(P1+P2-P 21 -P 12 ) -1 (P2-P 21 ) T ,

[0155] The fusion coefficient matrix is:

[0156] K1=(P2-P 21 )(P1+P2-P 12 -P 21 ) -1 ,

[0157] K2=(P1-P 12 )(P1+P2-P 21 -P 12 ) -1 ,

[0158] The accuracy relationship satisfies:

[0159] P BC ≤P1 and P BC ≤P2,

[0160] trace(P BC )≤trace(P1) and trace(P BC )≤trace(P2).

[0161] (2) Multi-sensor fusion algorithm

[0162] In the smart greenhouse multi-source system, suppose there are n-source perception data (X1, P1), (X2, P2), ..., (X n ,P n ), and its corresponding cross-covariance P ij are all known precisely, where i, j = 1, 2, ..., n; when the local node variance matrix P i When both are known, the fusion formula of the state mean and covariance matrix is:

[0163]

[0164] The variables in the above two formulas are:

[0165] X=[X N1 ,…,X Nn ] T ; M=[I N1 ,…,I Nn ] T ; Where n represents the number of observers and N represents the dimension of the state vector;

[0166] B. Distributed fusion of unknown cross-covariance:

[0167] (1) Covariance cross fusion

[0168] In distributed multi-sensor systems, the covariance cross-fusion algorithm developed by Julier and Uhlman combines the upper bounds of local error variances through scalar weighting, eliminating the need to understand the inter-node correlation information, thereby simplifying the calculation and improving the accuracy of local estimation.

[0169] Assume there are two sensors in the system. are two unbiased estimates of the state variable X, namely:

[0170]

[0171] Where n = 1, 2 and the corresponding variance matrices are P1 and P2 respectively; when and The cross-covariance matrix P 12 When unknown, the CI fusion algorithm is as follows:

[0172]

[0173] in is the CI fusion output; P CI is the fusion variance matrix; the fusion coefficients ω1 and ω2 satisfy 0≤ω1,ω2≤1,ω1+ω2=1; the specific solution formulas for ω1 and ω2 are:

[0174]

[0175] When the number of system sensors is greater than or equal to three, the CI fusion formula is:

[0176]

[0177] in is the fusion output; P CI is the fusion variance; ω i is the fusion coefficient, i=1,2,…,n, 0≤ω i ≤1,ω1+…+ω n =1;ω i The specific solution formula is:

[0178]

[0179] (2) Inverse covariance cross fusion

[0180] In the absence of inter-node mutual correlation information, the CI fusion algorithm uses a convex combination method to fuse data based on the upper limit of the local error variance. Although this can improve estimation accuracy, as the number of sensors increases, the CI fusion results become increasingly conservative, which can easily affect the system fusion accuracy. Therefore, the ICI fusion algorithm came into being. This algorithm not only solves the over-conservatism problem of the CI algorithm, but also further improves fusion accuracy. The ICI algorithm is summarized as follows:

[0181] Assume there are two sensors in the system and Then the ICI fusion output is:

[0182]

[0183] in is the ICI fusion output; P ICI is the fusion variance matrix; fusion coefficient ω∈[0,1]; K f With L f is the fusion coefficient matrix, which satisfies K f +L f =I, specifically defined as:

[0184]

[0185] The fusion coefficient ω can be obtained by minimizing P ICI The trace or determinant is used to solve, that is:

[0186]

[0187] The method for initializing the chaotic opposition of the labor population is:

[0188] Chaos Tent Mapping:

[0189] The Tent mapping has a fast iteration speed and the chaotic sequence is evenly distributed in [0,1]. Its expression is as follows:

[0190]

[0191] where λ t is the chaos number for the tth iteration; t = 0, 1, …, T, where T is the maximum number of iterations; α is a constant in the range [0, 1]. As a method based on nonlinear theory, chaos mapping exhibits significant characteristics such as nonlinearity, universality, and randomness. It can achieve non-repeated exhaustion of all possible states within a specific boundary, which promotes the generation of new solutions and improves population diversity during the intelligent algorithm optimization process.

[0192] Adversarial Learning:

[0193] Assume that the position of a d-dimensional individual is:

[0194] Δ=(x1,x2,…,x d ),

[0195] Assuming that the lower and upper bounds of its individual position are lb and ub respectively, its opposite side position can be expressed as:

[0196]

[0197] When the upper and lower bounds of the objective function are symmetrical, it can be seen from the above formula that the reverse solution will form an exact mirror image correspondence with the original solution. When considering the dual function, this mirror image solution has the same value as the initial target solution, which makes it inconvenient to sort the two populations based on fitness, making it difficult to effectively improve the overall quality of the population; in the initialization stage of the population, some randomly generated individuals of the intelligent algorithm are often located in invalid or marginal areas far away from the optimal solution, which slows down the search efficiency of the population. By implementing the opposing gene learning strategy and introducing a pair of random solutions and their opposing solutions at the initial stage of the population, the initial quality of the population is significantly optimized compared to adding only two unrelated random solutions.

[0198] Labor initialization strategy integrating chaotic tent mapping and adversarial learning:

[0199] Combining chaotic tent mapping with OBL, the specific mathematical model is:

[0200]

[0201] Where i=1,2,…,n, is the j-th dimension component of the opposing position of the i-th worker, is the dot product operator.

[0202] This strategy is equivalent to using the Tent mapping to uniformly vary the sum of the upper and lower bounds of the objective function, thereby dynamically compressing the original initial worker positions. Assuming the number of workers is n, the specific steps of this strategy are: first, generate n worker positions through random distribution; second, generate n chaotic opposing positions using an improved worker initialization strategy; finally, sort these 2n positions by fitness, and select the top n positions with the best fitness as the initial workers.

[0203] The specific method of using the adaptive T-distribution mutation operator to perturb the individual positions of workers is as follows:

[0204] The specific formula of the adaptive T-distribution mutation operator that perturbs the current optimal worker position is:

[0205]

[0206] where x new is the latest worker position after being perturbed by the adaptive T-distribution mutation operator; is the global optimal solution of the current iteration; iter is the current number of iterations, MaxIter is the total number of iterations; P iter Generate a random probability value for the current iteration. If it is greater than 0.8, perform adaptive T-distribution mutation perturbation. Otherwise, use the initial optimal solution. is the dot product operator; t(k) represents the probability density function of the Student's T distribution with k degrees of freedom.

[0207] The specific method of optimizing the global position of workers through vertical and horizontal cross-strategy is:

[0208] Horizontal cross:

[0209] The specific calculation formula is:

[0210]

[0211] in and They are and The d-dimensional worker after horizontal cross processing; r1 and r2 are random numbers in [0,1]; c1 and c2 are random numbers in [-1,1]. By dividing the solution space, it is reconstructed into a semi-group hypercube with binary operation characteristics, and new worker individuals are created in the hypercube and its boundary area, and then marginal exploration is carried out. This strategy can not only significantly improve the search efficiency of the GWCA algorithm, but also effectively reduce the uncovered area in the search, thereby greatly enhancing the global search capability of the GWCA algorithm.

[0212] After performing horizontal crossover, a new generation of workers is first generated using a random pairing mechanism. Secondly, the fitness values ​​of the new workers are compared with those of their predecessors to screen out the best workers and retain them, thereby improving the convergence efficiency of the GWCA algorithm.

[0213] Vertical cross:

[0214]

[0215] in For workers The offspring workers are generated through vertical crossover; r is a random number in [0,1]. Compared with the local optimality and early convergence problems of the traditional GWCA algorithm, this step introduces mutant molecules through the vertical crossover strategy and then crosses the worker dimension, thus jumping out of the local optimality and enhancing the global optimization performance.

[0216] By improving the traditional GWCA algorithm using chaotic opposition population initialization, vertical and horizontal cross optimization and adaptive T distribution mutation perturbation strategy, it can not only enhance the global optimization performance of GWCA and jump out of local optimality in time, but also improve the algorithm convergence rate and reduce the optimization cost.

[0217] The establishment process of the IGWCA-ELM model specifically includes:

[0218] Step A1: Set the initial parameters of the Great Wall construction algorithm and ELM, such as the initial population size, the number of hidden neural nodes, and the number of iterations;

[0219] Step A2: Build the IGWCA optimized ELM network framework and form the corresponding initial labor based on the ELM input weights and hidden layer bias;

[0220] Step A3: Calculate the fitness of the workers and determine the initial optimal workers, i.e. the emperor agent, based on the fitness;

[0221] Step A4: Continuously iterate to update the global optimal position of the workers;

[0222] Step A5: When the number of iterations is reached, stop the optimization, otherwise return to step A4;

[0223] Step A6: Use the IGWCA global optimal output as the optimal input weight and hidden layer bias of ELM, and train the optimal ELM model based on the dataset to output the global optimal prediction.

[0224] Example 1

[0225] Verifying the performance of the PICI model

[0226] (1) Performance comparison

[0227] In actual crop production, the mutual covariance matrices between multi-source sensors are often complex and unknown. Therefore, this embodiment uses the unknown mutual covariance distributed fusion algorithm as the core component of the smart greenhouse secondary fusion center. In addition, because ICI distributed fusion technology is superior to CI fusion in terms of fusion accuracy in multi-source environments, this embodiment uses the ICI fusion algorithm over the above two types of unknown mutual covariance distributed fusion technologies. The specific proof is:

[0228] Proof: Assume that the fusion coefficient ω * =1-ω, then:

[0229]

[0230]

[0231] Perform diagonal transformation on P1 and P2, and we have M1=HP1H T 、M2=HP2H T , and then we have:

[0232]

[0233] Take the diagonal elements of matrices M1 and M2 Then we have:

[0234]

[0235] According to the inequality (a 2 +b 2 )≥2|ab|, we can get:

[0236]

[0237] The right-hand side of the above formula is:

[0238] (1-ω) 2 +ω 2 +2ω(1-ω)-1=0,

[0239]

[0240] Therefore, (M) ii ≥0, that is, all diagonal elements of M are positive, so M is a positive definite matrix, then:

[0241] P ICI (1-ω)≤P CI (ω).

[0242] That is: for any fusion coefficient ω∈[0,1], there is always ω * ∈[0,1], satisfying: P ICI (ω * )≤PCI (ω), that is, the fusion performance of the ICI model is stronger than that of the CI model in terms of accuracy, which is proved.

[0243] (2) Covariance ellipse geometric analysis

[0244] The following is a comparative analysis of the fusion performance of the CI algorithm, the SICI algorithm, and the PICI algorithm of this embodiment from a geometric perspective, that is, using covariance ellipsoid comparative analysis, taking a four-sensor system as an example:

[0245]

[0246] Where P1, P2, P3 and P4 are the estimated covariance matrices of the two-dimensional state vector of the four sensors respectively.

[0247] Depend on Figure 3 It can be seen that although the above three fusion algorithms can all achieve high fusion accuracy, the covariance ellipse generated by the ICI algorithm is always within the range of the CI algorithm. Since the tighter the ellipse, the higher the accuracy, the ICI algorithm can show better fusion performance than the CI algorithm in terms of fusion accuracy, which perfectly conforms to the accuracy comparison results proved from the formula perspective in the previous section. Among them, the covariance ellipses of the SICI algorithm and the PICI algorithm of this embodiment completely overlap, indicating that the two have the same accuracy, thus proving that the PICI fusion structure proposed in this embodiment is fully comparable to the SICI structure in terms of fusion performance.

[0248] (3) Comparison of algorithm fusion rates

[0249] The comparison of the fusion rates of the CI model, the traditional SICI model, and the PICI model proposed in this embodiment in the above four-sensor system is shown in Table 1 below:

[0250] Table 1 Fusion rate comparison table

[0251]

[0252]

[0253] As can be seen from the above table, in the 20 rounds of fusion experiments, the PICI algorithm proposed in this embodiment has the fastest fusion rate, with an average time consumption of only 0.452 milliseconds; the SICI algorithm has the longest average time consumption, which is as high as 25.564 milliseconds; although the CI algorithm has a lower average time consumption, its fusion accuracy is the worst, so it is not discussed here.

[0254] In summary, the PICI algorithm has the advantages of high precision, few iterations and fast speed when dealing with the local fusion problem of multi-source data in smart greenhouses. The above experiments show that compared with the traditional CI algorithm and SICI algorithm, the PICI algorithm can effectively solve the time delay problem that sequential fusion cannot overcome without sacrificing fusion accuracy, thereby greatly improving the fusion rate, reducing the computing resource occupancy rate, and enabling system resources to be more effectively scheduled to other tasks, thereby achieving the purpose of improving global fusion efficiency. At the same time, the PICI model proposed in the present invention can also significantly reduce the energy consumption of multi-source perception, thereby effectively improving the life cycle of the wireless sensor network, which is particularly important for maintaining distributed wireless sensor networks and is suitable for some smart greenhouses with high maintenance costs.

[0255] Example 2

[0256] The traditional classical PSO algorithm, modern popular GWO, MFO, WOA and GWCA algorithms were simulated and compared with the IGWCA algorithm of the present invention in the IEEE CEC 2005 benchmark suite; the experimental computer configuration is Intel (R) Core (TM) i3-9100 CPU @ 3.60GHz processor, 8.00GB running memory, 64-bit Windows 11 operating system, and MATLAB R2022b simulation software;

[0257] like Figure 4-Figure 26 It can be seen that:

[0258] (1) In all five unimodal test functions, the IGWCA algorithm demonstrates excellent convergence performance. Compared with traditional GWCA, IGWCA has a significantly improved global search capability on F1 to F4, and its convergence performance is significantly improved on F5.

[0259] (2) Among the seven basic multi-modal test functions, IGWCA performs best on functions F8, F9, and F11. At the same time, although IGWCA has strong optimization performance on F6, F7, and F12, its convergence speed is slow. On the contrary, although IGWCA shows strong convergence performance on F10, its optimization accuracy is poor compared with the traditional WOA model.

[0260] (3) Among the two extended multimodal test functions, IGWCA shows better convergence rate and global optimization performance on F14; while on F13, although it shows stronger optimization performance, its convergence rate is lower.

[0261] (4) Among the nine composite combination test functions, IGWCA also performs quite well, and the nine optimization results are very close to the theoretical minimum. Among them, IGWCA performs best on F15, F16, F17, F20, F21, F22 and F23, while its convergence performance on F18 and F19 is slightly inferior.

[0262] In summary, the performance of the IGWCA algorithm has been improved compared with the traditional GWCA algorithm. Although the improvement is not obvious on several basic multimodal test functions, it performs well on unimodal test functions, extended multimodal test functions, and composite combination test functions. This shows that the IGWCA algorithm has good optimization performance and can solve most problems.

[0263] To validate the global fusion performance of the IGWCA-ELM model, field temperature data from a strawberry greenhouse in Xinhuang Town, Liuhe District, Nanjing, Jiangsu Province, from January 31 to February 24, 2024, were used as the research sample. Continuous monitoring was performed for 8 hours per day at a 30-minute sensing interval, resulting in a total of 400 sets of field data over 25 days. Of these 400 sets, 320 sets of field data were randomly selected as training samples for the IGWCA-ELM model, and the remaining 80 sets served as testing samples.

[0264] This example uses five indicators to evaluate the global fusion performance of ELM and each improved model, namely, mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), and goodness of fit (R2). The specific calculation formulas are:

[0265]

[0266] Where n is the number of test set samples; is the model prediction value; x k (d) is the true value of the temperature data in the test sample, is the average value of temperature data in the test sample.

[0267] This example compares the global fusion accuracy of the IGWCA-ELM algorithm with popular algorithms such as GWO-ELM, WOA-ELM, SSA-ELM, DBO-ELM, and SCHO-ELM. Except for the IGWCA model, the remaining algorithms are traditional unimproved versions and are not described here. The specific parameters of the IGWCA-ELM algorithm are shown in Table 2:

[0268] Table 2 IGWCA-ELM algorithm parameter selection table

[0269]

[0270] In Table 2, the number of neurons in the input layer of the ELM model is the number of rows of the input matrix after data normalization of the training set; the number of neurons in the output layer of the ELM model is the number of rows of the output matrix after data normalization of the training set; with regard to the selection of the number of neurons in the hidden layer, in order to ensure the scientific rationality of each model when making predictions, all ELM models in this embodiment are set to a single hidden layer structure, and the number of neurons in the hidden layer is uniformly set to 40.

[0271] Comparison of global fusion performance of each model Figures 27-42 As shown in Table 3, the comparison of evaluation indicators is as follows: Figure 43-45 As shown:

[0272] Table 3 Comparison of evaluation indicators of each model

[0273]

[0274] The smaller the evaluation indicators MAE, MSE, RMSE and MAPE, the better the model; 2 Contrary to the above indicators, the larger the value, the better; at the same time, the smaller the area of ​​the radar pentagon, the better. Figures 27-42 , Table 3 and Figure 43-45 It can be seen that the global fusion performance of the IGWCA-ELM model proposed in this embodiment is the best among all the above models. Its MAE, MSE, RMSE and MAPE are reduced by 93.65%, 99.66%, 94.17% and 84.28% respectively compared with the traditional ELM model. 2 It increased by 23.48%, indicating that IGWCA can effectively overcome the problems of poor prediction accuracy and unstable fusion performance caused by random assignment of initial input weights and hidden layer biases in the traditional ELM model. Therefore, the IGWCA-ELM model is very suitable as a global accurate fusion algorithm for temperature data in smart greenhouses.

[0275] Therefore, the combination of the PICI model and the IGWCA-ELM model not only solves many technical difficulties in traditional methods, but also significantly improves the fusion accuracy, speed and stability, providing strong technical support for application scenarios such as smart greenhouses.

[0276] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0277] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0278] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A global fusion method of temperature data based on the PICI-IGWCA-ELM model, characterized by: include: A smart greenhouse multi-sensor system is set up, including edge sensing nodes and internal sensing nodes. The edge sensing nodes contain a large amount of independent information, while the internal sensing nodes have a large common sensing area and high data redundancy. Perform pairwise parallel fusion processing on internal perception nodes, and use the PICI model to calculate the fusion estimation value of each pair of internal perception nodes and their corresponding covariance matrix; After the local fusion of all internal sensing nodes is completed, the data of the edge sensing nodes are globally fused with the local fusion results of all internal sensing nodes; Then, based on the improved Great Wall construction algorithm, the extreme learning machine model is optimized and the labor population is initialized with chaos opposition. Adaptive T-distribution mutation operator is used to perturb the individual positions of workers; Optimize the global position of workers through vertical and horizontal cross-cutting strategies; The global optimization capability of the IGWCA algorithm was verified using the IEEE CEC 2005 benchmark suite, and the IGWCA-ELM model was established based on the verification results. The IGWCA algorithm is used to optimize the initial input weights and hidden layer biases of the ELM model; The fusion processing method includes: Distributed fusion with known cross-covariance: (1) Dual sensor fusion algorithm When there are dual sensors and , the cross-covariance matrix Known, the Bar-Shalom algorithm is: The corresponding fusion variance matrix for: in ; The fusion coefficient matrix is: The accuracy relationship satisfies: (2) Multi-sensor fusion algorithm In the smart greenhouse multi-source system, assuming there is Source-aware data , and its corresponding cross-covariance are precisely known, among which ; When the local node variance matrix When both are known, the fusion formula of the state mean and covariance matrix is: The variables in the above two formulas are: in represents the number of observers, Represents the dimension of the state vector.

2. The temperature data global fusion method based on the PICI-IGWCA-ELM model according to claim 1 is characterized in that: The method for initializing the chaotic opposition of the labor population is: Chaos Tent Mapping: Tent mapping has a fast iteration speed, and the chaotic sequence The uniform distribution is as follows: in For the Iteration chaos number; ,in is the maximum number of iterations; for Internal constant; Adversarial Learning: Assume that the position of a d-dimensional individual is: Assume that the lower and upper bounds of the individual positions are 、 , then the opposite side position can be expressed as: When the upper and lower bounds of the objective function are symmetrical, the opposite-side position calculation formula shows that the reverse solution will form an exact mirror image of the original solution. When considering the dual function, this mirror image solution has the same value as the initial objective solution, which makes it difficult to sort the two populations based on fitness, making it difficult to effectively improve the overall quality of the population. Labor initialization strategy integrating chaotic tent mapping and adversarial learning: Combining chaotic tent mapping with OBL, the specific mathematical model is: in , For the The first worker in the opposing position dimensional component, is the dot product operator, This strategy is equivalent to using the sum of the upper and lower limits of the objective function as the center, using the tent mapping to uniformly change, and then dynamically compressing the original initial worker positions; assuming the number of workers is n, the specific steps of this strategy are: first, generate n worker positions through random distribution; second, generate n chaotic opposing positions by improving the worker initialization strategy; finally, sort these 2n positions by fitness, and take the top n positions with the best fitness as the initial workers.

3. The temperature data global fusion method based on the PICI-IGWCA-ELM model according to claim 1 is characterized in that: The specific method of using the adaptive T-distribution mutation operator to perturb the individual positions of workers is as follows: The specific formula of the adaptive T-distribution mutation operator that perturbs the current optimal worker position is: in is the latest worker position after being perturbed by the adaptive T-distribution mutation operator; is the global optimal solution for the current iteration; is the current iteration number, is the total number of iterations; Generate a random probability value for the current iteration. If it is greater than 0.8, perform adaptive T-distribution mutation perturbation. Otherwise, use the initial optimal solution. is the dot product operator; The degrees of freedom are The Student's T distribution probability density function.

4. The temperature data global fusion method based on the PICI-IGWCA-ELM model according to claim 1 is characterized in that: The specific method of optimizing the global position of workers through vertical and horizontal cross-strategy is: Horizontal cross: The specific calculation formula is: in and They are and The d-th dimension of labor after horizontal cross processing; and yes Internal random number; and yes Internal random number; Vertical cross: in For workers 、 sub-generation laborers produced through vertical crossover; yes Internal random number.

5. The temperature data global fusion method based on the PICI-IGWCA-ELM model according to claim 1 is characterized in that: The establishment process of the IGWCA-ELM model specifically includes: Step A1: Set the initial parameters of the Great Wall construction algorithm and ELM, such as the initial population size, the number of hidden neural nodes, and the number of iterations; Step A2: Build the IGWCA optimized ELM network framework and form the corresponding initial labor based on the ELM input weights and hidden layer bias; Step A3: Calculate the fitness of the workers and determine the initial optimal workers, i.e. the emperor agent, based on the fitness; Step A4: Continuously iterate to update the global optimal position of the workers; Step A5: When the number of iterations is reached, stop the optimization, otherwise return to step A4; Step A6: Use the IGWCA global optimal output as the optimal input weight and hidden layer bias of ELM, and train the optimal ELM model based on the dataset to output the global optimal prediction.

6. The temperature data global fusion method based on the PICI-IGWCA-ELM model according to claim 2, characterized in that: The fusion processing method further includes: Distributed fusion with unknown cross-covariance: (1) Covariance cross fusion Assume there are two sensors in the system. 、 is a state variable Two unbiased estimates of , namely: in And the corresponding variance matrices are 、 ;when and The cross-covariance matrix When unknown, the CI fusion algorithm is as follows: in It is the CI fusion output; is the fusion variance matrix; fusion coefficient 、 satisfy , ; and The specific solution formula is: When the number of system sensors is greater than or equal to three, the CI fusion formula is: in is the fusion output; is the fusion variance; is the fusion coefficient, , , ; The specific solution formula is: (2) Inverse covariance cross fusion Assume there are two sensors in the system and , then the ICI fusion output is: in It is the ICI fusion output; is the fusion variance matrix; fusion coefficient ; and is the fusion coefficient matrix, which satisfies , specifically defined as: Fusion coefficient By minimizing The trace or determinant is used to solve, that is: 。

Images