Intelligent networked vehicle cooperative localization based on improved wolf pack algorithm optimized neural network

Abstract

The invention belongs to the field of intelligent vehicle positioning in an intelligent traffic system, and discloses an intelligent network connection vehicle cooperative positioning method and system based on an improved wolf pack algorithm optimization neural network, and the method comprises the steps: generating a target position based on a motion track, adding noise, and calculating an initial coordinate through employing a least square method; taking the normalized distance difference between the target and the road side unit as an input feature; optimizing a BP neural network weight matrix by adopting an improved wolf pack algorithm; and correcting the initial coordinates by using the optimized BP network, and outputting a high-precision 3D positioning result. According to the method, the global search efficiency of the wolf pack algorithm is improved by adopting chaos initialization and a dynamic step length strategy; a t distribution variation mechanism prevents the algorithm from falling into local optimum; a BP neural network structure is improved, and error compensation is realized in combination with IWPA. According to the method, the Z-axis error can be remarkably reduced, the global convergence speed is improved, the anti-noise capability is enhanced, and balanced optimization of precision and stability is realized.

Description

[0001] Method and system Technical Field

[0002] The present invention belongs to, but is not limited to, the field of intelligent vehicle positioning technology in intelligent transportation systems, and in particular relates to a collaborative positioning method and system for intelligent connected vehicles based on an improved wolf pack algorithm to optimize a neural network. Background Art

[0003] Existing vehicle positioning technology primarily relies on the Global Navigation Satellite System (GNSS), but it is susceptible to signal attenuation in complex urban conditions, resulting in reduced positioning accuracy. Geometric positioning methods (such as ToA and TDoA) are limited by vertical height differences between roadside units, leading to amplified Z-axis errors. While traditional BP neural networks can be used for error compensation, they are prone to falling into local optimality. Furthermore, the traditional wolf pack algorithm (WPA) suffers from slow convergence and insufficient global search capabilities when optimizing neural networks. Furthermore, existing methods experience significant increases in positioning errors in the presence of noise, making it difficult to meet high-precision requirements.

[0004] In view of the above analysis, the technical problems that need to be solved urgently in the existing technology are: (1) Z-axis positioning error amplification problem: In the field of vehicle-road collaborative positioning, the main challenge faced by existing methods is the amplification of vertical (Z-axis) positioning errors. This problem mainly stems from the nonlinear effect of the height difference of the roadside cooperative unit (RSU) on the three-dimensional signal propagation. When a vehicle communicates with roadside units at different heights, the vertical geometric constraints will cause the ranging noise to have a cumulative transmission effect on the Z-axis.

[0005] (2) Ordinary wolf pack algorithms are prone to falling into local optimality: The application of traditional BP neural networks in positioning error compensation has long been limited by the problem of local optimal traps. The fundamental reason is that the initialization and optimization strategies of network weight parameters have dual defects: on the one hand, the stochastic gradient descent method easily converges to suboptimal solutions in non-convex loss function spaces; on the other hand, the fixed step size strategy adopted by the traditional wolf pack algorithm (WPA) leads to an imbalance between global exploration and local exploitation.

[0006] (3) High noise sensitivity: The high sensitivity of existing positioning methods to noise is another technical challenge that needs to be addressed. The traditional least squares (LS) algorithm directly introduces noise into the positioning solution process, resulting in a nonlinear increase in positioning error with the noise level. Although a neural network is introduced into the BP-LS method for error compensation, its unoptimized network structure still shows obvious performance limitations in high-noise scenarios: the suppression of vertical errors often comes at the expense of amplifying planar (X / Y axis) errors. Summary of the Invention

[0007] In response to the problems existing in the prior art, the present invention provides a collaborative positioning method and system for intelligent connected vehicles based on an improved wolf pack algorithm to optimize the neural network.

[0008] The present invention is implemented as follows: a method for collaborative positioning of intelligent connected vehicles based on an improved wolf pack algorithm to optimize a neural network, characterized in that the method for collaborative positioning of intelligent connected vehicles based on an improved wolf pack algorithm to optimize a neural network specifically includes: S1: Generate the target position based on the motion trajectory, add noise and calculate the initial coordinates using the least squares (LS) method; S2: The normalized distance difference between the target and the roadside unit is used as the input feature; S3: Use the improved wolf pack algorithm (IWPA) to optimize the BP neural network weight matrix; S4: Use the optimized BP network to correct the initial coordinates and output high-precision 3D positioning results.

[0009] Furthermore, the S1, positioning solution includes: Input: Noisy ranging data ; Output: initial coordinates (x, y, z); Implementation: Solve the overdetermined equations by the least squares (LS) method to minimize the sum of squared ranging errors.

[0010] Furthermore, the S3, improving the wolf pack algorithm, includes: (1) Chaotic initialization: Logistic chaotic mapping is used to generate the initial weight value formula (4), and the weight search space is evenly covered through nonlinear sequences; (4) (2) Dynamic step size strategy: Design an exponential decay step size formula (5), use a larger step size in the initial stage to enhance global exploration, and gradually reduce it in the later stage to fine-tune parameters, and combine 12 sinusoidal direction exploration; (5) (3) T-distribution mutation: Mutation formula (6) is applied to the 50% of individuals with the lowest fitness, and the heavy-tailed characteristics of the t-distribution are used to generate large deviation disturbances. The improved algorithm maintains diversity through a population update mechanism (eliminating the worst individuals and adding new solutions).

[0011] (6) Furthermore, the improved BP neural network in S3 is composed of an input layer (5 nodes), a hidden layer (12 nodes), and an output layer (3 nodes): The input layer fuses the initial 3D coordinates and the normalized distance difference between roadside units as features; the hidden layer uses the Sigmoid activation function to achieve nonlinear mapping; the output layer directly outputs the corrected high-precision coordinates through linear activation; the improved wolf pack algorithm (IWPA) is used to optimize the network weight matrix , and bias , IWPA covers the global parameter space through chaotic initialization, balances exploration and development with a dynamic step-size strategy, and enhances the ability to escape from local optimality through t-distribution variation, ultimately building an error compensation model. The training process is divided into two stages: global parameter optimization and fine-tuning.

[0012] Another object of the present invention is to provide an intelligent connected vehicle collaborative positioning system based on an improved wolf pack algorithm to optimize a neural network, the system specifically comprising: Initial positioning module, used to calculate the initial coordinates; A data preprocessing module is used to take the normalized distance difference between the target and the roadside unit as input features; Neural network module, which uses improved wolf pack algorithm to optimize the network weight matrix; The error compensation and positioning optimization module uses the optimized BP network to correct the initial coordinates and output high-precision 3D positioning results.

[0013] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows: First, this invention addresses the challenges of Z-axis error amplification caused by height differences in vehicle-road collaborative positioning, as well as the tendency of traditional algorithms to fall into local optimality. By proposing a collaborative positioning method based on an improved wolf pack algorithm to optimize a BP neural network, this method achieves improved three-dimensional positioning accuracy through a deep fusion of algorithms and models using chaotic initialization, dynamic step-size adjustment, and a t-distribution mutation strategy. Simulation experiments demonstrate that, in high-noise scenarios, the Z-axis error is reduced by 91.2% compared to the traditional least squares method (the error is stabilized within 1.15 meters), and the X / Y-axis error increments are controlled within 2.19 meters. The algorithm also converges rapidly within the first 200 iterations. This solves long-standing technical bottlenecks in the industry, such as insufficient vertical positioning accuracy and poor robustness to high noise levels in complex urban environments. This method fills a technological gap in the field of three-dimensional collaborative positioning for intelligent connected vehicles, providing a feasible solution for low-cost, high-precision positioning systems.

[0014] Second, the expected benefits and commercial value of this invention's technical solution after implementation are as follows: Once implemented, the technical solution is expected to bring significant benefits and commercial value to the fields of intelligent connected vehicles, autonomous driving, and smart transportation. By addressing the industry pain point of Z-axis error amplification in high-noise environments, this technology can significantly improve vehicle positioning accuracy, reduce reliance on high-density roadside unit (RSU) deployments, and save approximately [unclear context - likely a typo] in infrastructure costs. At the commercial application level, it can provide high-precision positioning solutions for autonomous driving companies, in-vehicle navigation system suppliers, and smart city projects. This technology can also be extended to scenarios such as drone logistics and industrial robotics, further expanding the market potential.

[0015] This invention's technical solution fills a technological gap in the industry, both domestically and internationally. Existing vehicle-road collaborative localization methods primarily focus on optimizing planar (X / Y axis) errors, but lack effective solutions for suppressing vertical (Z-axis) errors. This invention, for the first time, proposes a collaborative localization framework based on an IWPA-BP neural network. By analyzing the nonlinear propagation characteristics of vertical ranging noise and designing a targeted error compensation model, it fills a technological gap in 3D positioning, particularly in suppressing Z-axis errors, in high-noise environments.

[0016] The technical solution of this invention solves a long-standing but elusive technical challenge: in complex environments such as urban canyons and overpasses, traditional positioning algorithms suffer from a persistent problem of amplified Z-axis errors due to height differences, and in high-noise environments, these errors grow exponentially. By integrating an intelligent optimization algorithm with a neural network, this invention stabilizes the Z-axis error to below 1.15 meters even in high-noise environments, resolving the industry's challenge of insufficient vertical positioning accuracy in dynamic traffic scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a collaborative positioning method for intelligent connected vehicles based on an improved wolf pack algorithm to optimize a neural network, provided by an embodiment of the present invention; Figure 2 This is a flow chart of a wolf pack algorithm provided by an embodiment of the present invention; Figure 3 This is a diagram of the BP neural network structure provided by an embodiment of the present invention; Figure 4 This is a module diagram of an intelligent connected vehicle collaborative positioning system based on an improved wolf pack algorithm to optimize a neural network, provided by an embodiment of the present invention; Figure 5 This is a comparison diagram of the errors of various algorithms under different noise levels provided by an embodiment of the present invention; Figure 6 This is a comparison of the mean absolute errors of each axis under different noise levels provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] like Figure 1 As shown, an embodiment of the present invention provides a collaborative positioning method for intelligent connected vehicles based on an improved wolf pack algorithm to optimize a neural network. The method specifically includes: S1: Generate the target position based on the motion trajectory, add noise and calculate the initial coordinates using the least squares (LS) method; S2: The normalized distance difference between the target and the roadside unit is used as the input feature; S3: Use the improved wolf pack algorithm (IWPA) to optimize the BP neural network weight matrix; S4: Use the optimized BP network to correct the initial coordinates and output high-precision 3D positioning results.

[0020] Existing collaborative positioning solutions for intelligent connected vehicles are susceptible to reduced positioning accuracy in the presence of multipath interference, accumulated ranging errors, and dynamic environments. In particular, relying solely on the least squares (LS) method for initial coordinate estimation in complex roadside environments struggles to balance noise suppression with computational efficiency. Subsequent filtering or neural network corrections often lead to local optima, making them unable to meet the requirements of high-precision, real-time collaborative positioning. To address these technical challenges, the present invention integrates the advantages of the Improved Wolf Pack Algorithm (IWPA) and the BP neural network to globally optimize initial coordinates under noise perturbations. This approach overcomes the bottlenecks of traditional algorithms, which are prone to local optima and slow convergence, thereby achieving high-precision three-dimensional collaborative positioning.

[0021] During the target position generation phase, the time-continuous motion coordinates of the target node are first obtained by interpolating and filtering the vehicle trajectory data. To simulate the error characteristics of signal propagation and ranging systems in real environments, observation noise with a zero-mean Gaussian distribution is introduced into the trajectory point positions. The noisy coordinates are correlated with the roadside unit (RSU) ranging information using the least squares method to determine the initial spatial coordinates of the target. This step balances data processing efficiency while providing an initial estimate with representative noise interference for subsequent neural network corrections, helping to improve overall convergence accuracy.

[0022] The construction of a normalized distance difference feature effectively reveals the relative spatial geometric relationship between the target and multiple roadside units. Specifically, the ranging values ​​between the target and each RSU are normalized with respect to the maximum ranging range to obtain a normalized distance difference vector in the range [0, 1], which is used as the input feature of the neural network. This feature engineering not only leverages the robustness of this distance information to environmental changes, but also reduces gradient instability caused by numerical differences in the input dimensions during network training, thereby providing a robust feature foundation for subsequent network weight optimization.

[0023] During the weight optimization phase, the Improved Wolf Pack Algorithm (IWPA) was introduced as a global optimization tool for the BP neural network weight matrix. By integrating a multi-dimensional search strategy with a dynamic step-size adjustment mechanism during population initialization, IWPA empowers individual wolves with enhanced exploration capabilities in the high-dimensional weight space. During the iteration process, an adaptive update rule based on the historically optimal wolf pack positions and the local exploration neighborhood was employed to effectively balance global search with locally refined weight adjustments. Furthermore, by introducing position mapping constraints and a wolf-clustering factor, the algorithm rapidly converges to a near-global optimal solution, avoiding the drawbacks of traditional wolf pack algorithms that often fall into local optimality.

[0024] When correcting the initial coordinates using an IWPA-optimized BP network, the normalized distance difference features are input into the optimization network, and coordinate corrections containing environmental noise errors are obtained through multi-layer nonlinear mapping. During network feedforward and backpropagation, a mixed Sigmoid and ReLU activation unit is used layer by layer, enabling the network to both possess nonlinear fitting capabilities and accelerate gradient flow. The output layer uses a linear transformation to generate a three-dimensional offset vector, which is then added to the initial LS coordinates to form the final high-precision three-dimensional positioning result. This correction process leverages the advantages of the IWPA optimization strategy in the network weight space, significantly improving the accuracy and robustness of positioning estimates.

[0025] This invention effectively addresses the shortcomings of traditional collaborative positioning methods in terms of initial estimation accuracy and network correction convergence speed by integrating an improved wolf pack algorithm with a BP neural network. The improved global optimization capability of the wolf pack algorithm enables network weights to quickly approach optimal values. Combined with robust extraction of normalized features, this approach mitigates the effects of multipath noise and avoids local optimality traps, thereby achieving meter-level or even sub-meter-level high-precision three-dimensional positioning in dynamic, high-noise interference scenarios. This significantly enhances the practicality and reliability of collaborative positioning for intelligent connected vehicles.

[0026] The S1, positioning solution includes: Input: Noisy ranging data ; Output: initial coordinates (x, y, z); Implementation: Solve the overdetermined equations by the least squares (LS) method to minimize the sum of squared ranging errors.

[0027] The Wolf Pack Algorithm (WPA) achieves optimization by simulating the collaborative hunting behavior of wolves. Its core includes three roles: Alpha wolf (global optimal solution holder), scout wolf (multi-directional exploration of new solutions) and fierce wolf (local fine search). It iteratively updates the solution set based on three intelligent behaviors: reconnaissance (1), summoning (2) and siege (3). The specific process is shown in Figure 2 .

[0028] (1) (2) (3) Note: : The current position of the i-th wolf in the d-th dimension.

[0029] : Reconnaissance direction number (p=1,2,…,h).

[0030] : Total number of reconnaissance directions.

[0031] : Reconnaissance step length.

[0032] : The position of Alpha wolf (global optimal solution) in the dth dimension.

[0033] : Summoning step length, usually smaller than the detection step length.

[0034] : The current optimal solution position.

[0035] : Random perturbation factor.

[0036] : Siege step length, usually 1 / 2 of the reconnaissance step length.

[0037] However, the wolf pack algorithm has the defects of slow convergence speed and easy to fall into local optimality.

[0038] The S3, improved wolf pack algorithm, includes: (1) Chaotic initialization: Logistic chaotic mapping is used to generate the initial weight value formula (4), and the weight search space is evenly covered through nonlinear sequences to avoid uneven distribution and premature convergence caused by random initialization.

[0039] (4) Note: : The chaotic variable of the nth iteration, with a value range of (0,1).

[0040] : The chaotic variable of the n+1th iteration, calculated from the previous item.

[0041] and : The upper and lower bounds of the search space for weight parameters.

[0042] : Through chaotic variables The initial weight value obtained by mapping ensures global search coverage.

[0043] (2) Dynamic step size strategy: Design an exponential decay step size formula (5), use a larger step size in the initial stage to enhance global exploration, and gradually reduce it in the later stage to fine-tune parameters, and combine 12 sinusoidal direction exploration to improve search diversity.

[0044] (5) Note: : The step size at the tth iteration.

[0045] : Initial step size, controls the global exploration range.

[0046] : Current iteration number.

[0047] : Maximum number of iterations, controls the decay rate of the step size.

[0048] (3) T-distribution mutation: Mutation formula (6) is applied to the 50% of individuals with the lowest fitness, using the heavy-tailed nature of the t-distribution to generate large deviation disturbances, effectively escaping the local optimum. The improved algorithm maintains diversity through a population update mechanism, including eliminating the worst individuals and supplementing new solutions.

[0049] (6) Note: : Weight after mutation.

[0050] : Original weight value.

[0051] : Coefficient of variation intensity, controlling the disturbance amplitude.

[0052] : A t-distributed random number with v degrees of freedom, whose heavy-tailed nature can generate large deviations.

[0053] like Figure 3 As shown, the improved BP neural network S3 is composed of an input layer, a hidden layer, and an output layer. The input layer has 5 nodes, the hidden layer has 12 nodes, and the output layer has 3 nodes.

[0054] The input layer fuses the initial 3D coordinates and the normalized distance difference between roadside units as features; the hidden layer uses the Sigmoid activation function to achieve nonlinear mapping; the output layer directly outputs the corrected high-precision coordinates through linear activation. To overcome the defect of traditional BP network relying on gradient descent and easily falling into local optimality, the improved wolf pack algorithm (IWPA) is used to optimize the network weight matrix. , and bias , IWPA uses chaotic initialization to cover the global parameter space, a dynamic step-size strategy to balance exploration and exploitation, and t-distribution mutation to enhance the ability to escape local optima, ultimately building an error compensation model. The training process is divided into two phases: global parameter optimization and fine-tuning, ensuring the network maintains high robustness in complex noisy environments.

[0055] In order to address the problems in the existing collaborative positioning process of intelligent connected vehicles, such as the initial positioning accuracy being greatly disturbed by ranging noise, the algorithm convergence easily falling into local optimality, and insufficient online fine-tuning of the model, this paper proposes a hybrid positioning framework that combines least squares solution with an improved wolf pack algorithm to optimize the neural network, thereby significantly improving the 3D positioning accuracy and convergence stability.

[0056] The system uses trajectory data and multi-source ranging information fed back by roadside units to construct an overdetermined equation and solve it using the least squares method to obtain initial rough coordinates. This process utilizes noise modeling and residual constraints to effectively filter out Gaussian noise components in the ranging data, ensuring that the initial values ​​can be calculated quickly and with a certain degree of accuracy, laying the foundation for subsequent iterative optimization.

[0057] The normalized distance difference between the target vehicle and each roadside unit is used as a network input feature to achieve spatial scale uniformity and improve numerical stability. By introducing normalization, the interference of raw distance measurement differences on network training is suppressed, allowing the hidden layer to focus on capturing the key deviation distribution, effectively avoiding the vanishing or exploding gradient phenomenon caused by inconsistent input scales.

[0058] The Improved Wolf Pack Algorithm (IWPA) is used to globally optimize the weight matrix and bias of a BP neural network. During initialization, the algorithm utilizes a logistic chaotic map to uniformly sample the weight space. It coordinates global exploration with local exploitation through an exponentially decaying dynamic step size. It also utilizes a heavy-tailed t-distribution mutation mechanism to perturb the second half of the fitness solution, fostering sustained population diversity and the ability to escape from local optima. This overcomes the bottlenecks of traditional algorithms, which suffer from uneven granularity and premature convergence.

[0059] After IWPA optimization is complete, the optimal weights and biases are imported into the BP network, where initial coordinate errors are compensated using sigmoid and linear activation function structures. The network incorporates local gradient correction during online fine-tuning, rapidly adapting to environmental changes through short-range iterations. This achieves meter-level 3D positioning accuracy while maintaining high convergence speed.

[0060] The present invention integrates these modules into a collaborative positioning system, including an initial positioning module, a data preprocessing module, an IWPA-optimized neural network module, and an error compensation module. This system implements a closed-loop workflow from coarse to fine positioning. This system combines real-time performance with high precision, and is widely applicable in scenarios such as connected vehicles, autonomous driving, and vehicle-road collaborative navigation.

[0061] like Figure 4 As shown, an embodiment of the present invention provides an intelligent connected vehicle collaborative positioning system based on an improved wolf pack algorithm to optimize a neural network, specifically comprising: Initial positioning module, used to calculate the initial coordinates; A data preprocessing module is used to take the normalized distance difference between the target and the roadside unit as input features; Neural network module, which uses the improved wolf pack algorithm (IWPA) to optimize the network weight matrix; The error compensation and positioning optimization module uses the optimized BP network to correct the initial coordinates and output high-precision 3D positioning results.

[0062] The simulation experiment was carried out in MATLAB2024b environment, setting the asymmetric base station layout (A(0,0,0), B(0,200,0), C(200,0,0), D(0,0,10)) and the composite motion trajectory. The experimental results are shown in Figure 2. Figure 5 、 6 show: Z-axis error is significantly reduced: Experiments show that the IWPA-BP algorithm reduces the Z-axis error by 91.2% compared to the traditional LS method at a noise level of 1.5. Improved global convergence speed: The improved wolf pack algorithm converges quickly within 350 iterations, increasing optimization efficiency by 40%; Improved noise immunity: At a noise level of 1.3, the overall positioning error is stabilized within 1.15 meters, a reduction of 89.8% compared to the LS method. Multi-dimensional error balance: The X / Y axis error increment is controlled within 2.19 meters, achieving balanced optimization of accuracy and stability.

[0063] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0064] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A collaborative positioning method for intelligent connected vehicles, characterized in that include: Step 1: Receive the target vehicle motion trajectory data and roadside unit ranging data, construct an overdetermined equation system and solve it by least squares method to obtain the initial three-dimensional coordinates of the target vehicle; Step 2: normalize the distance measurement values ​​between the target vehicle and each roadside unit to generate a normalized distance difference vector; Step 3: Globally optimize the BP neural network weight matrix and bias using an improved wolf pack algorithm, wherein the improved wolf pack algorithm includes chaos initialization, exponential decay step size strategy and t distribution variation; In step 4, the optimized neural network is applied to the error compensation of the initial three-dimensional coordinates to output a high-precision three-dimensional positioning result.

2. The method according to claim 1, characterized in that In step 1, the least squares method obtains the initial coordinates by constructing a ranging residual matrix with a weighted noise model to minimize the sum of squares of ranging errors.

3. The method according to claim 1, characterized in that In step 3, chaotic initialization uses logistic mapping to generate the initial value of the network weight to achieve global uniform coverage. The exponential decay step size strategy dynamically adjusts the step size according to the current number of iterations to balance exploration and development. The t-distribution mutation imposes heavy-tail perturbations on the second half of the fitness solution to maintain population diversity.

4. The method according to claim 1, characterized in that The BP neural network consists of an input layer, a hidden layer, and an output layer. The input layer has five nodes for fusing the initial coordinates and the normalized distance difference. The hidden layer has twelve nodes and uses the Sigmoid activation function. The output layer has three nodes and uses a linear activation function.

5. An intelligent connected vehicle collaborative positioning system, characterized in that include: The initial positioning module is used to receive multi-source ranging data and calculate the initial three-dimensional coordinates of the target vehicle using the least squares method; The data preprocessing module is used to normalize the ranging data and output the normalized distance difference vector; The optimized neural network module is used to globally optimize the weights and biases of the BP neural network using the improved wolf pack algorithm; The error compensation module is used to apply the optimized neural network output to the initial three-dimensional coordinates to generate high-precision three-dimensional positioning results.

6. The system according to claim 5, characterized in that The initial positioning module is equipped with a weighted residual matrix construction unit to eliminate the influence of ranging noise and improve the accuracy of initial coordinates.

7. The system according to claim 5, characterized in that The optimization neural network module further includes an online fine-tuning unit for performing short-range gradient iteration based on the real-time positioning error to adapt to environmental changes.

8. A computer-readable storage medium, characterized in that Instructions are stored, and the instructions enable the processor to execute the intelligent connected vehicle collaborative positioning method described in claim 1.

9. The storage medium according to claim 8, wherein: The instructions further cause the processor to construct a ranging residual matrix according to a least squares method and solve for the initial coordinates.

10. The storage medium according to claim 8, wherein: The instructions further cause the processor to perform steps such as chaos initialization, dynamic step size adjustment, and t-distribution mutation of the improved wolf pack algorithm to optimize neural network weights and biases.

Images