Equipment deployment method based on precoding cross entropy optimization in Internet of Things system

Abstract

The invention discloses an equipment deployment method based on precoding cross entropy optimization in an Internet of Things system, which comprises the following steps of: (1) defining a physical boundary and an initial parameter of an equipment deployment area, and initializing an equipment deployment space coordinate; (2) discretizing a continuous space into binary codes through a space pre-coding module and a position coding mapper, and compressing search dimensions; (3) generating candidate deployment position samples according to the probability distribution; (4) calculating the system performance of the candidate position by using a complex black box system and an evaluation function; (5) updating probability distribution based on an elite sample, and accelerating convergence to an optimal solution; and (6) judging an optimal deployment position. Compared with a traditional gradient optimization and linear enumeration method, the closed-loop global optimization of the equipment deployment position is realized through space coordinate discretization coding, probability-driven candidate sample generation, system evaluation and a cross entropy probability iteration updating mechanism; and the global optimality guarantee and the calculation efficiency are synchronously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of device positioning, and in particular relates to a device deployment method based on precoding cross entropy optimization in an Internet of Things system. Background Art

[0002] Optimizing device location deployment in IoT systems is a key step in improving system performance. Current mainstream technologies rely primarily on two approaches, both of which have significant limitations.

[0003] The first is the linear enumeration method, which ensures global optimality through a brute-force search through discretized grid points. This method divides the deployment area into a uniform grid, calculates performance metrics point by point, and selects the optimal solution. However, its drawbacks are significant. The fixed grid accuracy cannot adapt to different scenarios—oversampling in simple environments wastes resources, while insufficient accuracy in complex environments leads to missed optimal solutions. Furthermore, when deploying multiple devices, all possible positions must be enumerated, exponentially expanding the search space and completely eliminating its engineering practicality.

[0004] The second type of optimization method is based on mathematical models, such as gradient descent and mixed integer programming. These methods rely on explicit mathematical representations of the objective function. However, in practical applications, the dynamic and nonlinear characteristics of IoT systems (such as IoT node deployment and multi-robot collaborative localization) are difficult to accurately describe using simplified mathematical models. Furthermore, as the number of devices increases, the search space dimensionality expands exponentially, leading to the curse of dimensionality. More importantly, these methods are sensitive to the initial solution and are prone to falling into local optimal solutions in multimodal optimization scenarios, making it difficult to ensure global optimality.

[0005] Therefore, the main problem that needs to be solved is that mathematical model optimization methods cannot handle black box high-dimensional problems, and although linear enumeration methods can guarantee theoretical optimality, they are limited by computational feasibility. Summary of the Invention

[0006] To solve the above problems, the present invention discloses a device deployment method based on precoding cross-entropy optimization in an Internet of Things system. Through discretized coding of spatial coordinates, probability-driven candidate sample generation, system evaluation and cross-entropy probability iterative update mechanism, it breaks through the high-dimensional optimization bottleneck of the black box system, eliminates the dimensionality curse of discrete enumeration, and realizes closed-loop global optimization of device deployment locations.

[0007] To achieve the above object, the technical solution of the present invention is as follows:

[0008] A device deployment method based on precoding cross entropy optimization in an Internet of Things system comprises the following steps:

[0009] (1) Define the physical boundaries and initial parameters of the equipment deployment area and initialize the equipment deployment space coordinates;

[0010] (2) Through the spatial precoding module and the position coding mapper, the continuous space is discretized into binary code to compress the search dimension;

[0011] (3) Using a random location sample generator, a sample of candidate deployment locations is generated according to the probability distribution;

[0012] (4) Using a complex black-box system and evaluation function, calculate the system performance of the candidate position;

[0013] (5) Update the probability distribution based on elite samples to accelerate convergence to the optimal solution;

[0014] (6) Determine the optimal deployment location.

[0015] Wherein, step (1) is specifically as follows:

[0016] Set the physical boundaries of the device deployment area, with the center coordinates set to (x0, y0) and the area side length to R x ×R y , so the horizontal coordinate constraint of the device is [x0-R x / 2,x0+R x / 2], the vertical coordinate constraint is [y0-R y / 2,y0+R y / 2].

[0017] Step (2) is as follows:

[0018] The horizontal and vertical coordinates are encoded into N-bit binary sequences through the spatial precoding module to generate 2 N ×2 N Then, using binary coding rules, the horizontal coordinates are mapped into N-bit binary codes {a1, a2, ..., a N}, mapping the vertical coordinate to an N-bit binary code {a N+1 ,a N+2 ,…,a 2N}; For any 2N-bit coordinates {a1, a2, ..., a 2N}, the position encoding mapper follows the following formula for coordinate mapping.

[0019]

[0020] Step (3) is as follows:

[0021] The random position sample generator is located between the spatial precoding module and the sample generation parameter updater. Its function is to dynamically adjust the binary bit probability distribution θ n (initial value θ n=0.5), generates a candidate position sample set covering the entire hovering area, and provides input for subsequent system evaluation. In the specific implementation, the random position sample generator module receives the binary coding rule (i.e., the coordinate mapping relationship of the 2N-bit sequence) from the spatial precoding module, and calculates the probability parameter θ of the current iteration round. n , for each binary bit a n Perform Bernoulli sampling: with probability θ n will a n Set to 1, with 1-θ n Set to 0, and finally generate M groups of independent binary sequence samples {a m} 1≤m≤M Each set of samples corresponds to the physical coordinates (x m ,y m ), the binary sequence is converted into continuous spatial coordinate values through the discretization mapping rule defined by the position encoding mapper (i.e., Formula (1)). This process ensures that candidate positions are uniformly explored in the search space through probability-driven random sampling and discrete encoding mapping, while avoiding the redundant calculations of traditional continuous random search. The generated candidate position sample set is then passed to the complex black box system for system evaluation.

[0022] Step (4) is as follows:

[0023] The internal operating rules of the complex black box system are determined by the complex system factors of the actual deployment. In implementation, the black box system acts as an independent functional entity and receives the candidate hovering position coordinates (x m ,y m ) as input, and outputs a set of real-time performance indicators of the system at that location through a multi-domain coupling model of environmental simulation Since the complexity of the black box system is directly determined by the dynamics, nonlinearity and multi-objective conflicts of its deployment environment, its internal calculation process does not rely on any preset explicit mathematical model or simplified assumptions, and only interacts with the algorithm framework through the input and output interface; the evaluation function is generated by parsing the black box output. Extract key performance indicators directly related to the optimization goal and map them into a single scalar evaluation value

[0024]

[0025] The design of the evaluation function relies entirely on the output characteristics of the black-box system, without the need to know its internal logic in advance, thus ensuring that the method of the present invention can be adapted to black-box systems of different implementation forms, including systems based on measured data deployed in real networks or virtualized systems based on high-fidelity simulation.

[0026] Step (5) is as follows:

[0027] The sample generation parameter updater is the core optimization component of this patented method. Its function is to dynamically adjust the binary bit probability distribution θ n , guiding the candidate positions of the random position sample generator to converge efficiently to the low energy consumption area. The specific implementation includes the following steps:

[0028] a) Cross-entropy driven elite sample screening: From the M groups of candidate position samples generated in the current iteration round, based on the core idea of the cross-entropy method - importance sampling and distribution matching, the elite sample with the best system evaluation value is screened out. Group as an elite sample set Represents the deployment location candidate with the best performance under the current probability distribution;

[0029] b) Iterative update of cross-entropy probability distribution: According to the cross-entropy method’s minimization of the Kullback-Leibler divergence criterion, by maximizing the likelihood probability of the elite sample, based on the binary bit values of the elite sample, the probability of each binary bit of the random position sample generator is updated according to the following formula n

[0030]

[0031] The smoothing factor α is used to control the weight balance between the historical probability distribution and the elite sample statistics. Its value has been verified through experiments to effectively prevent the algorithm from converging to the local optimal solution too early, while ensuring the stability of the probability update. This formula is essentially a recursive expression in cross-entropy optimization (adaptive importance sampling distribution). By gradually compressing the entropy value of the probability distribution, the candidate positions output by the sample generator are concentrated in the elite area.

[0032] Step (6) is as follows:

[0033] Calculate the maximum change in probability distribution after update If it is less than the preset threshold δ, the probability distribution is determined to have converged, the iteration is terminated and the current optimal deployment position is output; otherwise, As the input parameter of the next round of iteration, the sample generation and update process is repeated. Through the above mechanism, the sample generation parameter updater gradually transforms the probability distribution θ n By concentrating on high-evaluation areas, the candidate positions generated by the random position sample generator gradually approach the global optimal solution, and finally achieve efficient optimization of the equipment deployment position; after the algorithm converges, the optimal position coordinates are calculated by inputting the randomly generated optimal binary sequence a into the position coding mapper.

[0034] The beneficial effects of the present invention are:

[0035] (1) Compared with traditional gradient optimization and linear enumeration methods, the core innovation of the present invention is the simultaneous improvement of global optimality and computational efficiency. The candidate sample generation mechanism based on probability model iteration gets rid of the dependence on the differentiability of the objective function, making it possible to robustly search for the optimal solution in complex black-box environments where analytical models cannot be constructed, such as wireless channel interference and multipath effects.

[0036] (2) To address the engineering pain points of multi-device collaborative optimization, the dynamic precision adaptation mechanism of the present invention allocates differentiated coding bit widths to different devices, while ensuring the performance of key devices and reducing the computational overhead compared to the unified precision solution.

[0037] (3) Supporting the simultaneous evaluation of multiple sample sets through a distributed parallel architecture improves the real-time response capability of ultra-large-scale scenarios and further reduces the time cost of system optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is the algorithm flow chart of the present invention.

[0039] Figure 2 A two-dimensional coordinate grid.

[0040] Figure 3 Flowchart of sample generation and parameter update based on parallel processing.

[0041] Figure 4 Flowchart for generating and mapping location codes for multiple devices. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0043] like Figure 1 As shown, the device deployment method based on precoding cross entropy optimization in an Internet of Things system according to the present invention includes the following steps:

[0044] (1) Define the physical boundaries and initial parameters of the equipment deployment area and initialize the equipment deployment space coordinates;

[0045] First, set the physical boundaries of the device deployment area. The center coordinates are set to (x0, y0) and the area side length is R. x ×R y , so the horizontal coordinate constraint of the device is [x0-R x / 2,x0+R x / 2], the vertical coordinate constraint is [y0-R y / 2,y0+R y / 2].

[0046] (2) Through the spatial precoding module and the position coding mapper, the continuous space is discretized into binary code to compress the search dimension;

[0047] The horizontal and vertical coordinates are encoded into N-bit binary sequences through the spatial precoding module to generate 2 N ×2 N Then, using binary coding rules, the horizontal coordinates are mapped into N-bit binary codes {a1, a2, ..., a N}, mapping the vertical coordinate to an N-bit binary code {a N+1 ,a N+2 ,…,a 2N For any 2N coordinates {a1, a2, ..., a 2N}, the position encoding mapper follows the following formula for coordinate mapping.

[0048]

[0049] like Figure 2 As shown in the figure, taking an 8×8 two-dimensional coordinate grid as an example, black, white, and gray are used to intuitively show the mapping rules between binary codes and spatial positions. The specific instructions are as follows:

[0050] Coordinate grid structure:

[0051] The horizontal and vertical axes are marked with three-bit binary numbers 000 to 111, respectively, forming a discrete logical space of 8 rows and 8 columns.

[0052] Each intersection corresponds to a six-bit binary code (three bits on the horizontal axis + three bits on the vertical axis). For example, the coordinate (row 011, column 010) corresponds to the code value 001101 (the first three bits are 011, the last three bits are 010), reflecting the discrete mapping rule of the spatial precoding module.

[0053] Highlight code point distribution:

[0054] Lower left area (row 011, column 010): code value 001101.

[0055] Upper right area (row 101, column 110): code value 100110.

[0056] Middle right area (row 101, column 010): code value 101010.

[0057] (3) Using a random location sample generator, a sample of candidate deployment locations is generated according to the probability distribution;

[0058] The random position sample generator is located between the spatial precoding module and the sample generation parameter updater. Its function is to dynamically adjust the binary bit probability distribution θ n (initial value θn =0.5), generates a candidate position sample set covering the entire hovering area, and provides input for subsequent system evaluation. In the specific implementation, this module receives the binary coding rule (i.e., the coordinate mapping relationship of the 2N-bit sequence) from the spatial precoding module, and calculates the probability parameter θ of the current iteration round. n , for each binary bit a n Perform Bernoulli sampling: with probability θ n will a n Set to 1, with 1-θ n Set to 0, and finally generate M groups of independent binary sequence samples {a m} 1≤m≤M Each set of samples corresponds to the physical coordinates (x m ,y m ), the binary sequence is converted into continuous spatial coordinate values through the discretization mapping rule defined by the position encoding mapper (i.e., Formula (1)). This process ensures that candidate locations are uniformly explored in the search space through probability-driven random sampling and discrete encoding mapping, while avoiding the redundant calculations of traditional continuous random search. The generated candidate location sample set is then passed to the complex black box system for system evaluation.

[0059] (4) Using a complex black-box system and evaluation function, calculate the system performance of the candidate position;

[0060] The internal operating rules of the complex black box system are determined by the complex system factors of the actual deployment. In implementation, the black box system acts as an independent functional entity and receives the candidate hovering position coordinates (x m ,y m ) as input, and outputs a set of real-time performance indicators of the system at that location through a multi-domain coupling model of environmental simulation Since the complexity of the black box system is directly determined by the dynamic, nonlinear and multi-objective conflict of its deployment environment, its internal calculation process does not rely on any preset explicit mathematical model or simplified assumptions, and only interacts with the algorithm framework through the input and output interface. Extract key performance indicators directly related to the optimization goal and map them into a single scalar evaluation value

[0061]

[0062] The design of the evaluation function relies entirely on the output characteristics of the black-box system, without the need to know its internal logic in advance, thus ensuring that the method of the present invention can be adapted to black-box systems of different implementation forms, including systems based on measured data deployed in real networks or virtualized systems based on high-fidelity simulation.

[0063] (5) Update the probability distribution based on elite samples to accelerate convergence to the optimal solution.

[0064] The sample generation parameter updater is the core optimization component of this patented method. Its function is to dynamically adjust the binary bit probability distribution θ n , guiding the candidate positions of the random position sample generator to converge efficiently to the low energy consumption area. The specific implementation includes the following steps:

[0065] a) Cross-entropy driven elite sample screening: From the M groups of candidate position samples generated in the current iteration round, based on the core idea of the cross-entropy method - importance sampling and distribution matching, the elite sample with the best system evaluation value is screened out. Group as an elite sample set Represents the deployment location candidate with the best performance under the current probability distribution.

[0066] b) Iterative update of cross-entropy probability distribution: According to the cross-entropy method’s minimization of the Kullback-Leibler divergence criterion, by maximizing the likelihood probability of the elite sample, based on the binary bit values of the elite sample, the probability of each binary bit of the random position sample generator is updated according to the following formula n

[0067]

[0068] The smoothing factor α is used to control the weight balance between the historical probability distribution and the elite sample statistics. Its value has been verified through experiments to effectively prevent the algorithm from prematurely converging to a local optimum while ensuring the stability of probability updates. This formula is essentially a recursive expression for cross-entropy optimization (adaptive importance sampling distribution). By gradually compressing the entropy of the probability distribution, the candidate locations output by the sample generator are concentrated in the elite region.

[0069] (6) Determine the optimal deployment location.

[0070] Calculate the maximum change in probability distribution after update If it is less than the preset threshold δ, the probability distribution is determined to have converged, the iteration is terminated and the current optimal deployment position is output; otherwise, As the input parameter of the next round of iteration, the sample generation and update process is repeated. Through the above mechanism, the sample generation parameter updater gradually transforms the probability distribution θ n By concentrating on high-evaluation areas, the candidate locations generated by the random location sample generator gradually approach the global optimal solution, ultimately achieving efficient optimization of the device deployment location. After the algorithm converges, the optimal location coordinates are calculated by inputting the randomly generated optimal binary sequence a into the location code mapper.

[0071] In step (2), the grid precision parameter N can be dynamically adjusted to construct a hash mapping table of binary codes and actual coordinates, and the codes and performance parameters of non-empty grid points can be compressed and stored. Figure 2 The two-dimensional structure shown here can exponentially improve the accuracy of device deployment location searches by adjusting the binary encoding bit parameter N (i.e., the bit width of the horizontal and vertical coordinates). For example, when N increases from 3 bits to 4 bits, the number of grid cells increases from 64 to 256. At the same time, the hash mapping mechanism based on formula (1) ensures the storage and computational efficiency of the expanded encoding.

[0072] like Figure 3 As shown in the figure, in the algorithm proposed by the present invention, the three parts of random position sample generator, position code mapper and evaluation function can be processed synchronously by parallel processors. The parallel processing system is composed of M independently running parallel processing units, each of which performs random position sample generation (such as generating binary sequence 10011010), position code mapping (converting to physical coordinates (x m ,y m )), black box system evaluation step, and finally the sample generation parameter updater aggregates the evaluation results of all units, screens elite samples and updates the probability distribution parameters. This architecture is achieved by parallelizing generation and evaluation (such as unit 1 processing (x 1 ,y 1 ) while unit 2 processes (x 2 ,y 2 )), reducing the single-round iteration time from O(M) to O(1), significantly improving the real-time performance of deployment location optimization while ensuring the accuracy of binary coding and the strictness of multi-objective constraint verification.

[0073] like Figure 4 As shown in FIG, the multi-device global optimal deployment method of the present invention is implemented through a differential coding bit width cascade mechanism and a joint probability iterative optimization framework. The implementation process is as follows.

[0074] 1. Multi-device encoding initialization and differentiated bit width setting

[0075] Assign an independent binary encoding bit width N to each device to be deployed i (i is the device number), for example, device 1 sets N1=4, device 2 sets N1=2, and device 3 sets N1=5. Total code length (I is the total number of devices), forming a unified optimization variable a=[a1,a2,…,a L ].

[0076] 2. Code concatenation and joint position mapping

[0077] The binary sequences of each device are concatenated into a long code in the order of device number. For example, the 8-bit code 10011010 of device 1, the 4-bit code 0100 of device 2, and the 10-bit code 1100101101 of device 3 are concatenated into a 22-bit code 1001101001001100101101. The total code is decomposed into independent coordinates of each device through a shared position code mapper:

[0078]

[0079] where s i The encoding start bit of device i (such as device 2's i =8+1=9), ensuring that each device code segment is independently mapped to its deployment space.

[0080] Based on processes 1 and 2, the system evaluation and sample generation parameter updater are processed. This process and the overall iterative process are consistent with the overall algorithm.

[0081] It should be noted that the above content merely illustrates the technical idea of the present invention and cannot be used to limit the scope of protection of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications all fall within the scope of protection of the claims of the present invention.

Claims

1. A device deployment method based on precoding cross entropy optimization in an Internet of Things system, characterized by: The following steps are involved: (1) Define the physical boundaries and initial parameters of the equipment deployment area and initialize the equipment deployment space coordinates; (2) Through the spatial precoding module and the position coding mapper, the continuous space is discretized into binary code to compress the search dimension; (3) Using a random location sample generator, a sample of candidate deployment locations is generated according to the probability distribution; (4) Using a complex black-box system and evaluation function, calculate the system performance of the candidate position; (5) Update the probability distribution based on elite samples to accelerate convergence to the optimal solution; (6) Determine the optimal deployment location.

2. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: Step (1) is as follows: Set the physical boundaries of the device deployment area, with the center coordinates set to (x0, y0) and the area side length to R x ×R y , so the horizontal coordinate constraint of the device is [x0-R x / 2,x0+R x / 2], the vertical coordinate constraint is [y0-R y / 2,y0+R y / 2].

3. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: Step (2) is as follows: The horizontal and vertical coordinates are encoded into N-bit binary sequences through the spatial precoding module to generate 2 N ×2 N Then, using binary coding rules, the horizontal coordinates are mapped into N-bit binary codes {a1, a2, ..., a N }, mapping the vertical coordinate to an N-bit binary code {a N+1 ,a N+2 ,…,a 2N }; For any 2N-bit coordinates {a1, a2, ..., a 2N }, the position encoding mapper follows the following formula for coordinate mapping; 4. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: Step (3) is as follows: The random position sample generator module receives the binary coding rule from the spatial precoding module and generates the random position sample according to the probability parameter θ of the current iteration round. n , for each binary bit a n Perform Bernoulli sampling: with probability θ n will a n Set to 1, with 1-θ n Set to 0, and finally generate M groups of independent binary sequence samples {a m } 1≤m≤M ; Each set of samples corresponds to the physical coordinates of a device deployment location (x m ,y m ), the binary sequence is converted into continuous spatial coordinate values through the discretization mapping rules defined by the position encoding mapper; this process uses probability-driven random sampling and discrete encoding mapping to generate a set of candidate position samples that are passed to a complex black box system for system evaluation.

5. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: Step (4) is as follows: The black box system, as an independent functional entity, receives the candidate hovering position coordinates (x m ,y m ) as input, and outputs a set of real-time performance indicators of the system at that location through a multi-domain coupling model of environmental simulation The evaluation function is generated by parsing the black box output Extract key performance indicators directly related to the optimization goal and map them into a single scalar evaluation value 6. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: Step (5) is as follows: a) Cross-entropy driven elite sample screening: From the M groups of candidate position samples generated in the current iteration round, based on importance sampling and distribution matching, screen out the ones with the best system evaluation value. Group as an elite sample set Represents the deployment location candidate with the best performance under the current probability distribution; b) Iterative update of cross-entropy probability distribution: According to the cross-entropy method’s minimization of the Kullback-Leibler divergence criterion, by maximizing the likelihood probability of the elite sample, based on the binary bit values of the elite sample, the probability of each binary bit of the random position sample generator is updated according to the following formula n The smoothing factor α is used to control the weight balance between the historical probability distribution and the elite sample statistics.

7. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: Step (6) is as follows: Calculate the maximum change in probability distribution after update If it is less than the preset threshold δ, the probability distribution is determined to have converged, the iteration is terminated and the current optimal deployment position is output; otherwise, As the input parameter of the next round of iteration, the sample generation and update process is repeated; after the algorithm converges, the optimal position coordinates are calculated by inputting the randomly generated optimal binary sequence a into the position coding mapper.

8. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 3, characterized in that: By dynamically adjusting the grid precision parameter N, a hash mapping table of binary codes and actual coordinates is constructed, and the codes and performance parameters of non-empty grid points are compressed and stored to ensure the storage and computing efficiency after code expansion.

9. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: The random position sample generator, position code mapper, and evaluation function are processed synchronously by a parallel processor; the parallel processor consists of M independently running parallel processing units, each of which performs the random position sample generation, position code mapping, and black box system evaluation steps in sequence. Finally, the sample generation parameter updater aggregates the evaluation results of all units, screens elite samples, and updates the probability distribution parameters.

10. The device deployment method based on precoding cross entropy optimization in an Internet of Things system according to claim 1, characterized in that: The location parameters of multiple devices are encoded into multi-dimensional binary vectors, and a unified optimization variable is formed by concatenating the target codes. The optimal positions of all targets are updated synchronously and iteratively to generate a globally optimal deployment plan for multiple devices. The code lengths of different devices are adaptively adjusted to achieve differentiated location accuracy.

Images