Neural network positioning method for optimizing optical power uniformity based on improved Harris eagle algorithm

Abstract

The invention relates to a neural network positioning method for optimizing optical power uniformity based on an improved Harris eagle algorithm, and belongs to the field of visible light communication. The method comprises the following steps: firstly, optimizing a topological structure of an LED layout and the power of an LED according to an indoor environment; and secondly, a modulation and demodulation scheme capable of realizing LED ID identification without coding is realized for the LEDs, so that intersymbol interference and a multipath effect of a channel caused by dense deployment of the LEDs can be avoided, and a positioning error is reduced. And finally, making the received light intensity signals into a fingerprint database, substituting the fingerprint database into the attention convolutional neural network for training, and finally realizing real-time indoor positioning.

Description

Technical Field

[0001] The invention belongs to the field of visible light positioning and relates to an indoor positioning method based on optical power uniformity of an improved Harris Eagle algorithm. Background Art

[0002] With the development of new technologies such as the Internet of Things (IOT), autonomous mobile robots (AMR), unmanned aerial vehicles (UAV) and smart devices, various location-based service applications have gradually entered our lives, such as intelligent navigation, unmanned driving, online taxi-hailing and virtual reality services. Nowadays, people have more and more affairs in large shopping malls, airports, hotels, underground parking lots and mines, and convenient services are inseparable from the basic technical support of indoor positioning. Indoor positioning technology is gradually becoming the focus of attention in the industry with its huge commercial potential. Since the global positioning system and Beidou satellite signals are blocked by the walls of buildings and attenuated, they cannot provide accurate positioning. It is necessary to achieve high reliability and high precision positioning and tracking of people and equipment in indoor areas. Therefore, indoor positioning has extremely high research value and huge application potential.

[0003] Visible light positioning technology is a positioning method based on visible light communication (VLC) as the basic technology. It uses LED light sources as signal transmitters and cooperates with light-sensitive smart devices as signal receivers for precise positioning. Commonly used are photodiodes and image sensors (IS). A major advantage of this technology is that the propagation of visible light signals in the air is not affected by radio spectrum congestion. At the same time, because light cannot penetrate obstacles such as walls, it provides users with a highly private positioning method. In addition, LED lights have the advantages of small size, low energy consumption and long service life, making them an ideal indoor positioning solution that can achieve precise positioning in indoor environments.

[0004] Compared with other indoor positioning methods, visible light positioning technology has the following advantages in many aspects: rich spectrum resources, excellent anti-electromagnetic interference capability, excellent security and confidentiality, easy implementation and high-precision positioning. However, there are still some problems in visible light positioning that need to be solved. For example, the strength of the received signal of indoor visible light positioning is prone to fluctuations, resulting in large positioning errors, which greatly hinders the commercialization of visible light positioning systems.

[0005] Therefore, the present invention improves the positioning error caused by insufficient light distribution in the visible light positioning technology to meet the requirements of effective positioning. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide an indoor positioning method based on optical power uniformity of an improved Harris Eagle algorithm, which improves the common problems existing in the current visible light positioning and increases practicality and applicability.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] An indoor positioning method based on the uniformity of optical power of the improved Harris Eagle algorithm. First, the algorithm is used to optimize the light source layout and power distribution of LEDs to achieve uniform distribution of optical signals in indoor visible light positioning systems. In order to better achieve three-dimensional positioning, the algorithm is used to optimize the optical power of the three-dimensional area, and the variance of the received power of the three planes is considered as the optimization target on average. The quality factor is used to evaluate the uniformity of the receiving plane of the system. The LED at the transmitting end is modulated to reduce the inter-symbol interference caused by the dense deployment of LEDs. The receiving end uses PD to receive the optical signal and realizes positioning through the attention convolutional neural network;

[0009] The method specifically comprises the following steps:

[0010] S1: For indoor positioning environment, the indoor environment is divided into smaller positioning units;

[0011] S2: Use the improved Harris Eagle optimization algorithm to optimize the LED power and LED position;

[0012] S3: Based on the optimization results, PWM modulation is performed on each LED to achieve power distribution; the receiving end processes the received signal;

[0013] S4: Collect received light intensity value signals and three-dimensional coordinates to create a positioning fingerprint library;

[0014] S5: Design a convolutional neural network to reduce the nonlinear attenuation of the channel. Use the database to train the attention convolutional neural network to achieve online positioning when inputting the light intensity signal.

[0015] Further, the step S1 specifically includes:

[0016] In positioning systems, large spaces often result in uneven signal coverage and reduced positioning accuracy. Dividing positioning units can reduce the positioning area, making the signal in each unit more concentrated and stable, thereby improving the accuracy of positioning in each unit. By dividing a large space into small units, it is easier to monitor and maintain the positioning equipment in each unit, such as replacing or correcting faulty positioning lamps in a timely manner to ensure the normal operation of the positioning system. Through unitization, the deployment of non-positioning areas can be omitted, and resources can be concentrated in places where positioning is required, achieving more economical and efficient deployment.

[0017] Further, the step S2 specifically includes the following steps:

[0018] S21: The best point set strategy is used to generate the initial population. This method can generate individuals with more uniform distribution, reduce the repetition rate between individuals, and broaden the search range, thereby improving the solution accuracy of the algorithm. n is an n-dimensional Euclidean space. For any given S n Point r∈S n , then S n A point set in space can be represented as:

[0019] P m (k) = {(r 1 k),K,(r n k)},k=1,2,Km

[0020] Take point r k =2cos(2πk / p), 1≤k≤n, p is the smallest prime number satisfying (p-3) / 2≥n, or r k =e k ,1≤k≤n, then P m (k) is a set of good points, called r k is the best point. Deviation satisfy Where C(r,ε) is a constant that is only related to r,ε (ε>0).

[0021] S22: A nonlinear energy factor update strategy is introduced to improve the global optimization performance of the algorithm. The improved energy factor update formula can be expressed as:

[0022]

[0023] Where w is the adaptive parameter of the escape energy factor, which changes its effect with iteration. Its update rule can be expressed as:

[0024] w=((t+a) b / T)

[0025] S23: t distribution is also called Student distribution, which has parameter freedom m and its probability density function can be expressed as:

[0026]

[0027] Where Γ((m+1) / 2) is the second type Euler integral. The degree of freedom m affects the shape of the curve. When m=1, the t distribution is the Cauchy distribution; when m→∞, the t distribution is the Gaussian distribution.

[0028] By associating the degrees of freedom of the t distribution with the number of iterations of the algorithm, the Cauchy distribution approximation of random perturbations is achieved in the early stages of iteration. This adaptive adjustment strengthens the global exploration capability of the Harris Hawk Optimizer (HHO) in the early stages. As the iterations proceed, the t distribution gradually changes to a form closer to the Gaussian distribution, thereby enhancing the local search performance of the algorithm in the later stages and improving the optimization accuracy.

[0029] The adaptive t-distribution perturbation uses the update function to update the position:

[0030]

[0031] In the formula, t A (t) is an adaptive t-distribution function, which takes the number of population iterations as a parameter. Effectively based on the number of iterations, using the existing position information, the population position is disturbed and updated, and different disturbances can be introduced with the iteration update, which helps to prevent the algorithm from falling into a local optimal solution. As the number of iterations increases, this t-distribution function will gradually approach the Gaussian distribution, which helps the algorithm converge faster.

[0032] However, if this operator is used for all individuals in each iteration, it may increase the computation time and may conflict with the optimization characteristics of the algorithm itself. To solve this problem, the dynamic selection probability is used to adjust the application frequency of the adaptive t-distribution mutation operator to reduce its negative impact on the algorithm optimization algorithm. The dynamic selection probability strategy can be described as:

[0033] p=ω 1 -ω 2 ×(Tt) / T

[0034] S24: Optimize the position and power of the LED based on the three-dimensional positioning space. Use the quality factor Q of the received power Λ To evaluate the uniformity of the received power at the receiving end of the VLP system, it is proposed to use the LED light coverage R C To verify whether the light-deficient area has improved. Λ and R C Respectively expressed as:

[0035]

[0036]

[0037] In the formula, Λ=(R p P r ) 2 / σ 2 is the signal-to-noise ratio in a plane, R p is the photoelectric conversion efficiency, P r and σ are the received signal power and interference respectively, mean(·) represents the mean value, var(·) represents the variance, A FOV A is the indoor area directly covered by the LED field of view. total is the entire plane area. The variance of the plane received power is set as the optimization objective function, which can be expressed as:

[0038] f=var(Λ)

[0039] High Q Λ The value represents superior signal quality, which means that the uniformity of the light signal distribution on the receiving surface is better and the signal-to-noise ratio is higher. These characteristics enable the VLP system to have higher accuracy in providing positioning services. Λ When the value increases, it can be inferred that the overall performance of the system is improved, especially in terms of the quality of optical communication and positioning.

[0040] If the power distribution in three-dimensional space is not considered, the power distribution of receiving planes at other heights is improved compared to before optimization, but there is still a large difference and unevenness in the plane receiving power. The objective table function is improved, and the signal-to-noise ratio error of the highest layer, the lowest layer and the middle layer of the space is considered at the same time during optimization:

[0041]

[0042] In the formula, and They are respectively expressed as the variance values ​​of the plane heights at the top, bottom, and middle of the three-dimensional space.

[0043] Further, the step S3 specifically includes the following steps:

[0044] S31: In order to address the inter-symbol interference that may be caused by dense deployment, this study uses a non-coded LED identification scheme, and the transmitter and receiver do not need to go through a complex encoding and decoding process to directly and accurately identify the ID signal of each unit and its internal LED. This scheme requires the application of pulse width modulation to the LED lights in each positioning unit. By assigning different frequencies based on their spatial positions to each LED in the positioning unit, and in different units, different duty cycles based on the units need to be assigned to the LEDs. Assuming there is a square wave with a duty cycle of D, an amplitude of A, and a period of T, the function expression within one period can be expressed as:

[0045]

[0046] The Fourier expansion in this period can be expressed as:

[0047]

[0048] In the formula, ω = 2π / T is the phase. Therefore, in the Fourier expansion of the PWM waveform with different duty cycles D, the amplitudes and phases of different harmonics are different, which can be used to distinguish different LEDs.

[0049] This solution achieves the distinction of signals in the same frequency signal environment in multiple positioning units. This non-coded modulation scheme realizes LED frequency reuse, reduces the deployment cost of LEDs, and effectively overcomes the inter-code crosstalk and multipath interference problems caused by the increase in the number of LED lights, ensuring the performance of the VLP system.

[0050] S32: The FFT of the received signal γ(t) at the receiving end can be expressed as:

[0051]

[0052] Where N is the length of the fast Fourier transform, balancing the positioning accuracy with the signal sampling and processing time. In this paper, N=1024 is selected; ω=2πf is the analog frequency, where f is the frequency of the LED. According to Parseval's theorem, the energy of the signal is equal in the time domain and the frequency domain, so the power spectral density (PSD) of γ(t) can be expressed as:

[0053]

[0054] In the formula, the peak value of S(ω) can indicate the average power of the received signal at different frequencies. Therefore, the estimated RSS vector r can be obtained by capturing the peak value and phase of the LED fixed frequency in S(ω), that is:

[0055] r=[S(ω 1 ),S(ω 2),L,S(ω N )] T

[0056] In the formula, ω i =2πf i , f i is the frequency of the i-th LED.

[0057] Further, the step S4 specifically includes:

[0058] The RSS values ​​of the points collected at fixed intervals in each unit are used as data, and the coordinate information used as labels is used to construct the training data set of the neural network. At the same time, appropriate different points are selected from other points to make the test data set. In order to reduce the measurement error, each point will be repeatedly collected 20 times, and the average value will be calculated for further analysis.

[0059] Further, the step S5 specifically includes:

[0060] S51: Design a convolutional neural network to reduce the nonlinear attenuation of the channel. This paper chooses to use CNN to process the features of location fingerprints. By utilizing the deep learning ability of CNN, it can more accurately analyze and identify LED signals, thereby achieving precise positioning.

[0061] Under the premise of balancing positioning accuracy and computational efficiency, its network structure consists of two convolutional layers, two maximum pooling layers, and two fully connected layers. Each hidden layer in the network uses the rectified linear unit (ReLU) as the activation function. Furthermore, in order to keep the feature map dimension after the convolution operation matching the input signal size, the convolution kernel size is set to 3×3, the step size is set to 1, and a padding strategy of size 1 is adopted.

[0062] After the first convolution operation, 16 feature layers are generated, while the second convolution operation generates 32 feature layers. When performing the maximum pooling operation, a 2×2 window is used, so that the size of the output matrix after pooling is reduced to half of the original size. This design aims to reduce redundant data, extract key features of the signal, and achieve the best balance between efficiency and accuracy of the computational process. As for the fully connected layers, the first and second layers are configured with 64 and 32 neurons respectively, and the ReLU activation function is also used in these layers. The network output consists of three nodes, corresponding to the predicted three-dimensional coordinates In order to achieve accurate positioning prediction, the mean square error MSE is used as the loss function for network optimization:

[0063]

[0064] Where n is the number of positioning points, (x i ,y i ,zi ) are the actual coordinates.

[0065] To optimize the CNN model in this paper, an Adam optimizer with an initial learning rate of 0.001 was selected. During the training process, the data was processed in batches, and each batch contained 12 samples. In addition, to alleviate the overfitting phenomenon, a dropout layer was introduced, and its ratio was set to 0.1, which means that the network randomly discarded 10% of neurons during training. This method helps to improve the generalization ability of the model and prevent overfitting of the training data.

[0066] The beneficial effect of the present invention is that the method of the present invention can effectively solve the problems encountered in the actual application of visible light positioning, including the problem of uneven light power caused by not considering LEDs, the error caused by inter-symbol interference caused by dense deployment of LEDs, and the problem caused by fluctuations in received signals. Dividing a large space into units also makes it possible to apply various positioning conditions in actual applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0068] Figure 1 It is a flow chart of the improved Harris Hawk optimization method described in the present invention.

[0069] Figure 2 This is a flow chart of the attention convolutional neural network described in the present invention. Figure 3 This is a framework diagram of the attention module described in the present invention. Figure 4 A block diagram of an indoor positioning system for optimizing optical power uniformity using the improved Harris Eagle algorithm described in the present invention. DETAILED DESCRIPTION

[0071] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0072] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0073] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0074] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0075] In order to solve the problem of uneven distribution of optical signals in indoor VLP systems, this paper proposes a uniformity optimization method based on the improved Harris Eagle optimization algorithm, which can optimize the light source layout and power distribution of indoor LEDs. The improved Harris Eagle optimization algorithm has the characteristics of fast convergence speed and high optimization accuracy, which can meet the requirements of wide search range and short optimization time in synchronous optimization.

[0076] The present invention effectively alleviates the problem of high volatility of received signals within a fixed position in the indoor visible light positioning technology based on received signal strength, and proposes an attention convolutional neural network algorithm to learn time domain features to reduce the impact of signal fluctuations.

[0077] like Figure 1 As shown, the improved Harris Hawk algorithm method of the present invention specifically includes the following steps:

[0078] 1. Optimize LED power and layout

[0079] 1.1. First, use the good point set method to initialize the Harris hawk population.

[0080] 1.2. Adjust and update the position of individuals in the population and the escape energy and position of prey.

[0081] First, the initial parameters of the population need to be set. As the algorithm iterates, the fitness value is recalculated. If this fitness value is better than the fitness value of the current prey (optimal solution), the prey's position is updated with the newly discovered position. The selection of the Harris Hawk's swarming strategy is based on the escape energy of the prey, which usually varies in the range of (-2,2) and gradually decreases with the increase in the number of iterations.

[0082] 1.3 Exploration Phase

[0083] During this phase, the Harris hawk will use two different strategies to randomly search for prey. This process corresponds to the initialization and exploration phases in the optimization algorithm, during which the algorithm will search the solution space in a random manner.

[0084] 1.4. Strategy Transition from Exploration to Encirclement

[0085] At this stage, when the Harris Hawk finds the optimal solution and determines the location of the prey, it will switch from the search phase to the siege phase. This process is simulated by a specific transition probability, which flexibly adjusts the algorithm's search strategy according to the prey's escape action, making the algorithm's behavior more targeted and adaptive.

[0086] The improved Harris Hawk algorithm uses nonlinear escape energy E to determine which strategy to choose, whether to use a search strategy or a siege behavior. It plays a vital role in the algorithm's iteration process and is the key decision basis for the conversion between search behavior and siege behavior. As the number of iterations increases, the escape energy of the prey will gradually decrease, which simulates the prey becoming tired over time, thus providing the Harris Hawk with an opportunity to capture the prey:

[0087]

[0088] Where w is the adaptive parameter of the escape energy factor, which changes its effect with iteration. Its update rule can be expressed as:

[0089] w=((t+a) b / T)

[0090] Where a and b are the weights of the control weight factors. It is possible to determine when a and b can achieve the best effect through experiments based on the current positioning system.

[0091] In the formula, E 0Represents the escape energy of the prey. It is a random value between (-1,1) and will be automatically updated according to the algorithm requirements at each iteration. Among them, t represents the current number of iterations, and T is the set maximum number of iterations. When |E|≥1, it indicates that the prey has a high escape ability. At this time, the algorithm enters the exploration phase and randomly searches for possible prey locations in the solution space. When |E|<1, it means that the escape energy of the prey is low, and the Harris Hawk is more likely to capture the prey. Therefore, the algorithm enters the capture phase, at which time it will focus on a more precise search in the area around the current best solution to find a solution that can be improved.

[0092] 1.5. Encirclement and capture phase

[0093] This phase mimics how Harris's hawks surround their prey and make surprise attacks. Harris's hawks dynamically adjust their position and speed relative to their prey, and this behavior is translated into a set of equations that guide the optimization process toward the best solution.

[0094] In this phase, the algorithm mimics the natural behavior of Harris's Hawk: how to circle and pounce on prey. This behavior is reflected in the algorithm by dynamically adjusting the position and speed of the pursuer relative to the target position. By introducing r as the probability of the prey successfully escaping, the random value range is (0,1).

[0095] The roundup phase can be further broken down into the following four strategies:

[0096] Soft Encirclement Strategy: When the prey has enough energy, the Harris Hawk will quietly and patiently encircle the prey.

[0097] Hard Encirclement Strategy: When prey is exhausted, Harris's Hawks will make rapid combined attacks.

[0098] Soft mobbing with a progressive rapid dive strategy: Harris's hawk executes a rapid dive to capture prey in an attempt to escape with low energy.

[0099] Hard Encirclement with Progressive Fast Dive Strategy: This is a faster, more aggressive way of hunting when the prey is very low in energy and the hawk has a high chance of success.

[0100] The HHO algorithm is particularly suitable for dealing with complex optimization problems through its adaptive switching mechanism between broad search (exploration phase) and centralized optimization (hunting phase). This ability to trade-off between global exploration and local hunting provides the HHO algorithm with the potential to avoid local optimal solutions while finding global optimal solutions.

[0101] The improved Harris Eagle algorithm is used to optimize the LED optical power in multiple height planes to adjust the light source layout and optimize the optical power distribution on the receiving plane, thereby reducing the positioning dead zone and improving the performance of the positioning system.

[0102] 2. Design of positioning algorithm.

[0103] 2.1. Design of the transmitter and receiver The positioning system applies pulse width modulation to the LED lights in each positioning unit. By assigning different frequencies based on their spatial positions to each LED in the positioning unit, and in different units, different duty cycles need to be assigned to the LEDs based on the units. The receiver continuously samples the received signal and performs fast Fourier transform to obtain the power spectrum to distinguish different LEDs.

[0104] 2.2. Signal preprocessing, the RSS value is preprocessed using fast Fourier transform, and then the obtained frequency domain signal is substituted into CNN for training.

[0105] 2.3. Attention CNN network design Figure 2 As shown in Figure 2, the network structure consists of two convolutional layers, two maximum pooling layers, and two fully connected layers. Each hidden layer in the network uses the rectified linear unit (ReLU) as the activation function. Furthermore, in order to keep the feature map dimension after the convolution operation matching the input signal size, the convolution kernel size is set to 3×3, the step size is set to 1, and a padding strategy of size 1 is adopted.

[0106] After the first convolution operation, 16 feature layers are generated, while the second convolution operation generates 32 feature layers. When performing the maximum pooling operation, a 2×2 window is used, so that the size of the output matrix after pooling is reduced to half of the original size. This design aims to reduce redundant data, extract key features of the signal, and achieve the best balance between efficiency and accuracy of the computational process. As for the fully connected layers, the first and second layers are configured with 64 and 32 neurons, respectively, and the ReLU activation function is also used in these layers. The network output consists of three nodes, corresponding to the predicted three-dimensional coordinates (x i ,y i ,z i ). In order to achieve accurate positioning prediction, the mean square error MSE shown in formula (3-7) is selected as the loss function for network optimization.

[0107]

[0108] Where n is the number of positioning points, (x i ,y i ,z i ) are the actual coordinates.

[0109] Since the input power spectrum signal matrix may contain redundant and useless information, the attention mechanism is introduced to improve the model's ability to learn detailed features. Figure 3As shown. There is a channel attention module (CAM) in the network structure. The channel weight adjusted by this module is multiplied by the input feature layer to obtain the fused attention feature. Then, the output of the channel attention module is flattened into a one-dimensional array so that it can be connected to the subsequent fully connected layer. The input 32×4×8 feature layer is globally averaged pooled by the channel attention module, and the feature size after pooling is 1×1×32. Next, one-dimensional convolution is used for processing, where the size of the convolution kernel needs to be calculated by the input feature layer. The processed result is normalized to between 0 and 1 by the sigmoid function. This weight is the weight of each channel of the input feature layer. Multiply this weight by the original input feature layer to obtain the channel attention feature, which can be expressed as:

[0110] M c (F) = σ(conv1d(AvgPool(F)))

[0111] Where F is the input feature, σ(·) is the sigmoid function, conv1d(·) is the 1D convolution operation, and AvgPool is the global average pooling (GAP).

[0112] 2.4. Online sampling to achieve three-dimensional positioning.

[0113] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. An indoor positioning method based on optical power uniformity of improved Harris Eagle algorithm, characterized in that: Firstly, the improved Harris Eagle algorithm is used to optimize the layout and power of LEDs on the indoor plane, and the power uniformity of the receiving plane is used as an indicator; The method specifically comprises the following steps: S1: First, determine the three-dimensional positioning space according to the environment and divide it into small positioning units; S2: Optimize the layout and power of LEDs using the improved Harris Eagle algorithm; S3: modulate the LED at the transmitting end, and the receiving end processes the received signal; S4: Collect received light intensity value signals and three-dimensional coordinates to create a positioning fingerprint library; S5: Design a convolutional neural network to reduce the nonlinear attenuation of the channel. Use the database to train the attention convolutional neural network to achieve online positioning when inputting the light intensity signal.

2. The positioning method for optimizing optical power uniformity using the improved Harris Eagle algorithm according to claim 1, characterized in that: The step S1 specifically includes: Deploy i LEDs on the indoor ceiling with coordinates A single-chip microcomputer is used for pulse width modulation (PWM) driving as the transmitting end, and a photodiode (Pd) is used as the receiving end, and its coordinates are R Pd (x,y,z). And determine the three-dimensional space that needs to be positioned indoors.

3. The positioning method for optimizing optical power uniformity using the improved Harris Eagle algorithm according to claim 1, characterized in that: The step S2 specifically includes: The Harris Hawk Optimization (HHO) algorithm exhibits slow convergence and is prone to falling into local optimal solutions in the late stage of evolution. To overcome these limitations, an improved Harris Hawk Optimization algorithm is introduced to accelerate the convergence process, enhance the global search effect, and improve the performance of the algorithm for the optimization problem of spatial power reception in positioning systems. S21: The best point set strategy is used to generate the initial population. This method can generate individuals with more uniform distribution, reduce the repetition rate between individuals, and broaden the search range, thereby improving the solution accuracy of the algorithm. n is an n-dimensional Euclidean space. For any given S n Point r∈S n , then S n A point set in space can be represented as: P m (k)={(r1k),K,(r n k)},k=1,2,Km Take point r k =2cos(2πk / p), 1≤k≤n, p is the smallest prime number satisfying (p-3) / 2≥n, or r k =e k ,1≤k≤n, then P m (k) is a set of good points, called r k is the best point. Deviation satisfy Where C(r,ε) is a constant that is only related to r,ε (ε>0). S22: A nonlinear energy factor update strategy is introduced to improve the global optimization performance of the algorithm. The improved energy factor update formula can be expressed as: Where w is the adaptive parameter of the escape energy factor, which changes its effect with iteration. Its update rule can be expressed as: w=((t+a) b / T) S23: t distribution is also called Student distribution, which has parameter freedom m and its probability density function can be expressed as: Where Γ((m+1) / 2) is the second type Euler integral. The degree of freedom m affects the shape of the curve. When m=1, the t distribution is the Cauchy distribution; when m→∞, the t distribution is the Gaussian distribution. By associating the degrees of freedom of the t distribution with the number of iterations of the algorithm, the Cauchy distribution approximation of random perturbations is achieved in the early stages of iteration. This adaptive adjustment strengthens the global exploration capability of the Harris Hawk Optimizer (HHO) in the early stages. As the iterations proceed, the t distribution gradually changes to a form closer to the Gaussian distribution, thereby enhancing the local search performance of the algorithm in the later stages and improving the optimization accuracy. The adaptive t-distribution perturbation uses the update function to update the position: In the formula, t A (t) is an adaptive t-distribution function, which takes the number of population iterations as a parameter. Effectively based on the number of iterations, using the existing position information, the population position is disturbed and updated, and different disturbances can be introduced with the iteration update, which helps to prevent the algorithm from falling into a local optimal solution. As the number of iterations increases, this t-distribution function will gradually approach the Gaussian distribution, which helps the algorithm converge faster. However, if this operator is used for all individuals in each iteration, it may increase the computation time and may conflict with the optimization characteristics of the algorithm itself. To solve this problem, the dynamic selection probability is used to adjust the application frequency of the adaptive t-distribution mutation operator to reduce its negative impact on the algorithm optimization algorithm. The dynamic selection probability strategy can be described as: p=ω1-ω2×(Tt) / T S24: Optimize the position and power of the LED based on the three-dimensional positioning space. Use the quality factor Q of the received power Λ To evaluate the uniformity of the received power at the receiving end of the VLP system, it is proposed to use the LED light coverage R C To verify whether the light-deficient area has improved. Λ and R C Respectively expressed as: In the formula, Λ=(R p P r ) 2 / σ 2 is the signal-to-noise ratio in a plane, R p is the photoelectric conversion efficiency, P r and σ are the received signal power and interference respectively, mean(·) represents the mean value, var(·) represents the variance, A FOV A is the indoor area directly covered by the LED field of view. total is the entire plane area. The variance of the plane received power is set as the optimization objective function, which can be expressed as: f=var(Λ) High Q Λ The value represents superior signal quality, which means that the uniformity of the light signal distribution on the receiving surface is better and the signal-to-noise ratio is higher. These characteristics enable the VLP system to have higher accuracy in providing positioning services. Λ When the value increases, it can be inferred that the overall performance of the system is improved, especially in terms of the quality of optical communication and positioning. If the power distribution in three-dimensional space is not considered, the power distribution of receiving planes at other heights is improved compared to before optimization, but there is still a large difference and unevenness in the plane receiving power. The objective table function is improved, and the signal-to-noise ratio error of the highest layer, the lowest layer and the middle layer of the space is considered at the same time during optimization: In the formula, and They are respectively expressed as the variance values ​​of the plane heights above, below, and in the middle of the three-dimensional space.

4. The positioning method for optimizing optical power uniformity using the improved Harris Eagle algorithm according to claim 1, characterized in that: The step S3 specifically includes: S31: In order to address the inter-symbol interference that may be caused by dense deployment, this study uses a non-coded LED identification scheme, and the transmitter and receiver do not need to go through a complex encoding and decoding process to directly and accurately identify the ID signal of each unit and its internal LED. This scheme requires the application of pulse width modulation to the LED lights in each positioning unit. By assigning different frequencies based on their spatial positions to each LED in the positioning unit, and in different units, different duty cycles based on the units need to be assigned to the LEDs. Assuming there is a square wave with a duty cycle of D, an amplitude of A, and a period of T, the function expression within one period can be expressed as: The Fourier expansion in this period can be expressed as: In the formula, ω = 2π / T is the phase. Therefore, in the Fourier expansion of PWM waveforms with different duty cycles D, the amplitudes and phases of different harmonics are different, which can be used to distinguish different LEDs. This solution achieves the distinction of signals in the same frequency signal environment in multiple positioning units. This non-coded modulation scheme realizes LED frequency reuse, reduces the deployment cost of LEDs, and effectively overcomes the inter-code crosstalk and multipath interference problems caused by the increase in the number of LED lights, ensuring the performance of the VLP system. S32: The Fast Fourier Transform (FFT) of the received signal γ(t) at the receiving end can be expressed as: Where N is the length of the fast Fourier transform, balancing the positioning accuracy and signal sampling and processing time. In this paper, N=1024 is selected; ω=2πf is the analog frequency, where f is the frequency of the LED. According to Parseval's theorem, the energy of the signal is equal in the time domain and the frequency domain, so the power spectral density (PSD) estimate of γ(t) can be expressed as: In the formula, the peak value of S(ω) can indicate the average power of the received signal at different frequencies. Therefore, the estimated RSS vector r can be obtained by capturing the peak value and phase of the LED fixed frequency in S(ω), that is: r=[S(ω1),S(ω2),L,S(ω N )] T In the formula, ω i =2πf i , f i is the frequency of the i-th LED.

Images