Parameter adjustment optimization system, method and equipment for map binding and medium

Abstract

The invention relates to the technical field of map road binding positioning, in particular to a parameter adjustment optimization system, method and device for map road binding and a medium, and the parameter adjustment optimization system for map road binding comprises a sampling space module which is at least used for receiving hyper-parameter configuration; the optimizer module is at least used for optimizing the hyper-parameter configuration based on a preset optimization strategy; the parameter optimization iteration module is at least used for iterating the hyper-parameter configuration; the black box objective function module is at least used for generating a binding success proportion value under new hyper-parameter configuration during each iteration; and the evaluation updating module is at least used for comparing the binding success proportion value under the new hyper-parameter configuration generated by the black box objective function module during each iteration so as to determine the optimal hyper-parameter configuration. According to the method, the problem of limitation of hyper-parameter setting in the field of map binding positioning and the problem of efficiency in a high-dimensional parameter space of an existing HMM (Hidden Markov Model) can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of map binding and route positioning, and in particular to a parameter adjustment and optimization system, method, device and medium for map binding and route positioning. Background Art

[0002] In the field of map-based road positioning technology, Hidden Markov Model (HMM) has been widely used as a key technology to achieve accurate matching of vehicle locations and road networks. Although HMM has powerful sequence modeling capabilities in theory, its actual application effect is highly dependent on the configuration of hyperparameters. The setting of hyperparameters is crucial to the accuracy and efficiency of the model, but due to the different sampling ranges of hyperparameter values ​​required by different application scenarios, the current setting method has certain limitations.

[0003] Currently, the setting of hyperparameters relies on manual experience. This type of method often requires a lot of experiments and professional knowledge, which is not only time-consuming and labor-intensive, but also difficult to ensure optimal results in changing application scenarios.

[0004] In the existing technology, there are some automatic parameter optimization methods such as grid search and random search. Although they can reduce manual intervention and improve search efficiency to a certain extent, these parameter optimization methods often have problems such as slow convergence speed and high computational cost when facing high-dimensional hyperparameter space. In addition, there are some parallel computing technologies that optimize calculations by expanding computing resources, resulting in high consumption of computing resources. Summary of the invention

[0005] In response to the above technical problems, the present invention provides a parameter optimization system, method, device and medium for map binding, which can solve the limitations of hyperparameter setting of existing HMM in the field of map binding positioning and the efficiency problem in high-dimensional parameter space, can adapt to different map scenarios, and improve the positioning accuracy of autonomous driving vehicles.

[0006] A first aspect of the present invention provides a parameter optimization method for map binding, comprising: A sampling space module, at least for receiving hyperparameter configurations; An optimizer module, at least used to optimize the hyperparameter configuration based on a preset optimization strategy; A parameter optimization iteration module, at least used to iterate the hyperparameter configuration; The black box objective function module is used to generate a new binding success ratio value under the hyperparameter configuration at least in each iteration; The evaluation update module is used to at least compare the successful path binding ratio value under the new hyperparameter configuration generated by the black box objective function module at each iteration to determine the optimal hyperparameter configuration.

[0007] In a possible implementation, the sampling space module includes at least a user-defined input data type submodule, a sampling space submodule, and a sampling configuration submodule; wherein: The custom input data type submodule is used at least to input the data type, add the parameter name, and set the upper and lower sampling limits of the reference; The sampling space submodule is at least used to record hyperparameters and their indexes, sample new hyperparameter configurations, obtain new hyperparameter configurations, and align hyperparameter information with configuration information; The sampling configuration submodule is used for at least normalization operation and / or serialization operation hyperparameter configuration and is connected with the optimizer module.

[0008] In a possible implementation, the optimizer module includes at least an optimizer initialization submodule, a proxy model training submodule, a collection function update submodule and a collection function optimizer; wherein: The optimizer initialization submodule is used to construct at least a proxy model, a collection function, and a collection function optimizer; The surrogate model training submodule is at least used to collect performance data of the objective function under the hyperparameter configuration based on the surrogate model; The acquisition function updating submodule is at least used to update the acquisition function based on the performance data, wherein the acquisition function is used to improve the hyperparameter configuration according to the maximized expectation; The acquisition function optimizer is used to at least search for an improved solution and randomly select points in the sampling space for evaluation, and obtain an optimized hyperparameter configuration based on the evaluation results and the acquisition function.

[0009] In a possible implementation, the black box objective function module includes a data alignment submodule and a function encapsulation submodule; wherein: The data alignment submodule is at least used to obtain and compare the binding route ID comparison group with the hidden Markov model binding route ID; The function encapsulation submodule is at least used to input the new hyperparameter configuration into the black box objective function model, and output the binding success ratio value under the new hyperparameter configuration.

[0010] In a possible implementation, a hyperparameter configuration input module is further included, and the hyperparameter configuration input module is at least used to input hyperparameter information configuration into the sampling space module.

[0011] In a possible implementation, it also includes a historical observation recording module, which is at least used to record and visualize historical data, wherein the historical data at least includes configuration information corresponding to the observation results, the output value of the observation objective function, the test status and the calculation time.

[0012] In a possible implementation, an initialization module is also included, and the initialization module is at least used to selectively establish the sampling space module, the settings of the optimizer module and the basic settings in the historical observation record module.

[0013] A second aspect of the present invention provides a parameter optimization method for map binding, comprising: receiving a hyperparameter configuration in a sampling space; Optimizing the hyperparameter configuration based on a preset optimization strategy; The hyperparameter configuration is iterated, and the optimal hyperparameter configuration is determined by comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration.

[0014] In a possible implementation, optimizing the hyperparameter configuration based on a preset optimization strategy includes: Build surrogate models, acquisition functions, and acquisition function optimizers; Collect performance data of the objective function under hyperparameter configuration based on the surrogate model; updating an acquisition function based on the performance data, wherein the acquisition function is used to improve the hyperparameter configuration according to the maximized expectation; The acquisition function optimizer searches for improved solutions and randomly selects points in the sampling space for evaluation, and obtains the optimized hyperparameter configuration based on the evaluation results and the acquisition function.

[0015] In a possible implementation, the proxy model is a probabilistic random forest model, the acquisition function is an expected improvement method, and the acquisition function optimizer searches for an improved solution based on a random local search method.

[0016] In a possible implementation, after comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration to determine the optimal hyperparameter configuration, the method further includes: The historical data are recorded and visualized, wherein the historical data at least includes configuration information corresponding to the observation results, the output value of the observation objective function, the test status and the operation time.

[0017] In a possible implementation, the parameter optimization method for map binding is applicable to a hidden Markov model; the method of comparing the binding success ratio values ​​under the new hyperparameter configuration generated in each iteration to determine the optimal hyperparameter configuration also includes: Obtain and compare the binding route ID comparison group with the hidden Markov model binding route ID; Input the new hyperparameter configuration into the black box objective function model and output the successful binding ratio value under the new hyperparameter configuration; Get the ratio value within the preset ratio range as the optimal hyperparameter configuration.

[0018] According to a third aspect of the present invention, there is provided an electronic device, comprising: At least one processor; and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method described in the second aspect of the embodiment of the present invention.

[0019] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a computer, the method according to the second aspect of the embodiment of the present invention is executed.

[0020] The present invention improves the exploration efficiency of parameter space, utilizes the distribution information of historical parameters, and accelerates the parameter optimization process through intelligent sampling, so that the global optimal solution can be quickly found in the complex multi-dimensional parameter space of HMM, avoiding the common problems of inefficiency and parameter adjustment failure in traditional manual methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a block diagram of a parameter optimization system for map binding in an embodiment of the present invention.

[0022] Figure 2 The present invention is a block diagram of functional modules in a parameter optimization system for map binding in an embodiment of the present invention.

[0023] Figure 3 The present invention is a basic flow chart of a parameter optimization process for map binding in an embodiment of the present invention.

[0024] Figure 4 The present invention is a flow chart of an architecture for a parameter optimization process for map binding in an embodiment of the present invention.

[0025] Figure 5 The present invention is an interactive flow chart of a parameter optimization process for map binding in an embodiment of the present invention.

[0026] Figure 6 The present invention is an interactive scenario flow chart of a parameter optimization process for map binding in an embodiment of the present invention.

[0027] Figure 7 The figure is a comparison chart of the effects of various parameter optimization methods for map binding in the embodiments of the present invention.

[0028] Figure 8 The present invention is a flowchart of a method for optimizing parameters of map binding in an embodiment of the present invention.

[0029] Fig. 9 The present invention is a flowchart of another method for optimizing parameters of map binding in an embodiment of the present invention.

[0030] Fig.10 The present invention is a flowchart of another method for optimizing parameters of map binding in an embodiment of the present invention.

[0031] Fig.11 The figure is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0033] It should be understood that the terms "first", "second", and "third" etc. in the claims, specifications, and drawings of the present disclosure are used to distinguish different objects rather than to describe a specific order. The terms "include" and "comprise" used in the specification and claims of the present disclosure indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections. It should also be understood that the terms used in this disclosure specification are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure.

[0034] Map binding is the process of matching GPS track points to digital maps; common map binding technologies include: geometric method: matching GPS points to the nearest road segment based on distance or shape similarity. This method is simple to calculate, but is easily affected by GPS errors and complex road networks; topological method: considering topological information such as the connectivity and turning restrictions of the road network for matching, which is more accurate than the geometric method, but has higher computational complexity; probabilistic method: using probabilistic models to deal with the uncertainty of GPS and map data, such as Kalman filtering, particle filtering, etc. This type of method can better handle noise, but requires a reasonable probabilistic model; advanced fusion method: combining multiple sensor data and map information, such as fusing GPS, inertial sensors, vehicle dynamics models, etc. for matching. This type of method has high accuracy, but the system is complex and the cost is high.

[0035] Hidden Markov Model (HMM) is a commonly used probabilistic map binding method. Its basic idea is to model the road network as a hidden state, with GPS observation points as the observation sequence; define the emission probability, which represents the probability of observing a specific GPS point at a certain road location; define the transition probability, which represents the probability of moving from one road location to another, and use methods such as the Viterbi algorithm to solve the most likely state sequence.

[0036] The specific implementation may involve candidate point generation: generating multiple possible matching locations as candidates for each GPS point; emission probability calculation: calculating the emission probability based on factors such as the distance and direction from the GPS point to the candidate location; transition probability calculation: calculating the transition probability based on factors such as road section length, driving direction, speed limit, etc.; observation sequence processing: preprocessing of GPS data, such as denoising and interpolation; parameter optimization: optimizing the various parameters of HMM through training data; post-processing: performing post-processing such as smoothing and anomaly detection on the HMM output results.

[0037] The HMM method can effectively handle GPS noise and road network complexity and performs well in many applications. However, its performance depends largely on the selection of model parameters, so parameter optimization is the key to improving the performance of HMM map binding.

[0038] In view of this, if Figure 1 and Figure 2 As shown, a first aspect of the present invention provides a parameter optimization system for map binding, and the parameter optimization system includes a functional module and an operation module.

[0039] in: The functional modules at least include a sampling space module, an optimizer module, a black box objective function module, and a historical observation recording module; The operation module at least includes a hyperparameter configuration input module, an initialization module, a parameter optimization iteration module and an evaluation update module; The functional module is combined with the operation module to provide a parameter optimization strategy for a map binding system using a Hidden Markov Model (HMM).

[0040] In one embodiment of the present invention, the sampling space module is at least used to receive hyperparameter configuration.

[0041] For details, please continue to refer to Figure 2 The sampling space module is a storage part of the core configuration information in the parameter optimization system for map binding shown in the present invention, which at least includes a custom input data type submodule, a sampling space submodule and a sampling configuration submodule; Specifically, the custom input data type submodule is at least used to input data types, add parameter names, set reference sampling upper and lower limits, and determine whether to take logarithms; the input data types include numeric types and Boolean types, the numeric types include real types and integers, and the Boolean types include true and false value selection. The number of adjacent points of the parameters that can be collected and the adjacent points can be obtained by setting the upper and lower sampling limits, thereby improving and standardizing the sampling range.

[0042] The sampling space submodule is the sampling space, which is at least used to record hyperparameters and their indexes, sample new hyperparameter configuration information, obtain new hyperparameter configuration information, and align the hyperparameter information with the configuration information, thereby completing the overall management of HMM hyperparameters; The sampling configuration submodule is at least used for normalization operation and / or serialization operation hyperparameter configuration, and serves as a link between the sampling space module and the optimizer module, and is directly connected to the optimizer module.

[0043] Here, through this method, variable inputs of various data types can be automatically adapted, and the subsequent Bayesian optimization strategy can be assisted to perform efficient parameter search for the variables, overcoming the complexity of multiple data types in the parameter tuning process and ensuring the global optimal performance of the model.

[0044] In one embodiment of the present invention, the optimizer module is at least used to optimize the hyperparameter configuration based on a preset optimization strategy. Specifically, the preset optimization strategy may be a Bayesian optimization strategy.

[0045] For details, please continue to refer to Figure 2 The optimizer module is the core component module of the parameter optimization system for map binding shown in the present invention. Taking the Bayesian optimization strategy as an example, the optimizer module at least includes an optimizer initialization submodule, an agent model training submodule, an acquisition function update submodule and an acquisition function optimizer; wherein: The optimizer initialization submodule is at least used to construct a proxy model, a collection function and a collection function optimizer; wherein the proxy model is a probabilistic random forest model, which is responsible for modeling the objective function so as to predict its behavior without directly evaluating the objective function; the collection function is an expected improvement (EI) method, which updates the collection function according to the prediction results of the proxy model, and the collection function optimizer searches for parameter settings that can maximize the value of the collection function based on a random local search method.

[0046] The proxy model training submodule is at least used to collect performance data of the objective function under hyperparameter configuration based on the proxy model. Specifically, the proxy model training part related to Bayesian optimization can collect performance data of the objective function under different parameter settings by updating the data warehouse information of random numbers, thereby building an understanding of the objective function.

[0047] The acquisition function update submodule is at least used to update the acquisition function based on the performance data, and the acquisition function is used to improve the hyperparameter configuration according to the maximized expectation; specifically, the acquisition function update submodule guides the search process by evaluating the possible improvement of the objective function under the new parameter setting, and the acquisition function is changed according to the update of the proxy function tree.

[0048] The acquisition function optimizer is at least used to search for improved solutions and randomly select points in the sampling space for evaluation, and obtain the optimized hyperparameter configuration based on the evaluation results and the acquisition function; specifically, the acquisition function optimizer performs staggered local and random searches in the sampling space, searches for improved solutions through local neighborhood searches, and randomly selects points in the sampling space for evaluation to explore unknown areas and escape from local optimality, finds and outputs the parameter configuration that can maximize the expected improvement.

[0049] Here, in this way, the present invention improves the exploration efficiency of the parameter space, utilizes the distribution information of historical parameters, and accelerates the parameter optimization process through intelligent sampling, so that the global optimal solution can be quickly found in the complex multi-dimensional parameter space of HMM, avoiding the inefficiency and parameter adjustment failures common in traditional manual methods.

[0050] In one embodiment of the present invention, the black box objective function module is used at least to generate a new path binding success ratio value under a new hyperparameter configuration in each iteration.

[0051] For details, please continue to refer to Figure 2 The black box objective function module is the evaluation part of the parameter optimization system for map binding proposed in the present invention, which further includes: a data alignment submodule and a function encapsulation submodule; wherein: The data alignment submodule is at least used to obtain and compare the binding route ID comparison group with the hidden Markov model binding route ID to ensure that the corresponding evaluation information type and meaning are the same; The function encapsulation submodule is at least used to input the new hyperparameter configuration into the black box objective function model, and output the binding success ratio value under the new hyperparameter configuration.

[0052] Here, this method can assist the evaluation and update module of the process module to evaluate the accuracy of the hyperparameter configuration.

[0053] In one embodiment of the present invention, the historical observation recording module is at least used to record and visualize historical data, wherein the historical data at least includes configuration information corresponding to the observation results, the output value of the observation objective function, the test status and the calculation time.

[0054] Specifically, the historical observation recording module is the recording and output part of the parameter optimization system for map binding shown in the present invention, which further includes a historical data recording submodule, an observation value updating submodule and a historical data visualization submodule.

[0055] Specifically, the observation value includes the configuration information corresponding to the observation result, the output value of the observation objective function, the test state and the operation time, etc. Among them, the configuration information and the configuration information described in the sampling space have the same meaning; the objective function output is consistent with the black box objective function described in the black box objective function module; the test state represents the result record information of the success and failure of the operation; the operation time is the time from the start to the end of the process, which is used to determine whether the system operation has timed out.

[0056] Here, through this method, the functional module can assist the running module to provide basic processing and calculation services for the parameter optimization of map binding. With the help of preset optimization strategies such as Bayesian optimization strategy, efficient parameter search services for hyperparameter configuration are realized, overcoming the complexity of multiple data types in the parameter tuning process and still maintaining stable binding accuracy.

[0057] In one embodiment of the present invention, the hyperparameter configuration input module is at least used to input necessary hyperparameter information configuration into the sampling space submodule in the sampling space module, which is defined as information conforming to the system format, waiting for subsequent parameter sampling.

[0058] In one embodiment of the present invention, the initialization module is at least used to selectively establish the basic settings of the sampling space module, the optimizer module and the historical observation record module. The initialization module can selectively perform basic settings for each sub-module within the above modules based on actual parameter optimization requirements.

[0059] In one embodiment of the present invention, the parameter optimization iteration module is at least used to iterate the hyperparameter configuration. Specifically, the parameter optimization iteration module is the main loop link of the parameter optimization system shown in the present invention. Based on the preset optimization strategy provided by the optimizer module, the HMM hyperparameter configuration is continuously adjusted and optimized through the iterative process. Through this module, the ability to process large-scale parameter quantities is improved, and the process with a high number of iterations is optimized.

[0060] In one embodiment of the present invention, the evaluation update module is at least used to compare the successful path binding ratio value under the new hyperparameter configuration generated by the black box objective function module in each iteration to determine the optimal hyperparameter configuration. Specifically, the evaluation update module can perform performance evaluation on the hyperparameter configuration generated in each iteration, and by comparing the black box function outputs under different configurations, the optimal configuration can be selected, and the evaluated optimal configuration is applied to the HMM to achieve effect optimization.

[0061] Here, the parameter optimization system for map binding proposed in the present invention can solve the limitations of the existing HMM in the field of map binding positioning, especially the shortcomings of relying on manual experience and the efficiency of the automatic parameter adjustment method in the high-dimensional parameter space. The present invention can achieve automatic and efficient hyperparameter configuration for better results, improve the positioning accuracy of autonomous driving vehicles, and adapt to different map changes.

[0062] In addition, the present invention can adapt to the needs of map-bound positioning models in various scenarios, and meet the requirements of map-bound models in terms of multiple hyperparameters, high number of iterations, fast convergence speed, and lightweight model evaluation criteria. At the same time, the present invention is developed based on C++, and this method has significant advantages in processing speed and cross-platform compatibility. Compared with the common Python-based solutions in the industry, the present invention utilizes the high performance characteristics of C++ and combines the Bayesian optimization strategy to ensure search efficiency while maintaining the high accuracy of the HMM model, so that it can also run efficiently in resource-constrained airborne industrial computer embedded systems, achieving faster data processing speeds and lower resource consumption, while ensuring efficient operation and deployment flexibility on different platforms.

[0063] like Figure 3 As shown, in an application scenario of the present invention, in the initial stage, all necessary hyperparameter configurations are input into the sampling space, waiting for subsequent parameter sampling; the parameter adjustment stage is initialized, including the historical observation records of the functional modules, the configuration space, and the optimizer modules; in the main loop stage of parameter adjustment and optimization, the hyperparameters of the HMM are continuously adjusted and optimized through an iterative process, and each round the system will generate a new set of hyperparameter configurations, and the new configurations are evaluated through the objective function. According to the evaluation results, the system updates or replaces the current optimal hyperparameter configuration to gradually approach the optimal solution; this process ends until the preset maximum number of cycles is reached.

[0064] like Figure 4As shown, in an application scenario of the present invention, the input information type is limited by customizing Real, Integer and Boolen classes, the hyperparameter configuration stored in Configuration is defined, the HMM map binding model is encapsulated as Objfunc4Binder, the HMM configuration file information is read, the candidate heading change threshold, the emission matrix distance weight, the beta parameter of the transition probability, etc. are selected with high precision, the candidate road index number output by the model is compared with the reference true value of the configuration file, and the number of consistent index numbers of successful binding is calculated. This objective function is used to evaluate the parameter adjustment effect of the parameter adjustment model in HMM. The Optimizer optimizer is the core link, which coordinates the optimization process and defines parameters, manages the running process of the Advisor, and the Advisor includes the initialization, operation and update of the Bayesian model, the prediction of the target value through the tree structure training constructed by ProbRandForest, the calculation of adjacent advantages by EI sampling, the process of expanding the sampling of the collection optimizer to find the optimal value, the update of Configuration information, the acquisition of new observations, the storage in History, and the final output of the best result.

[0065] like Figure 2 and Figure 5 As shown, in an application scenario of the present invention, the optimizer module also includes a recommender. The core of the present invention is composed of the recommender in the optimizer module, which can respond to query requests for configurations and provide corresponding decision suggestions to generate new optimal configurations. In addition, the present invention also includes a visualization tool that allows intuitive understanding of data and analysis results through a graphical interface.

[0066] The relationship between the application scenario of the present invention and the algorithm used is as follows: Figure 6 As shown, the present invention is applicable to single-objective optimization tasks. When specific conditions exist in the input space, such as a sequential relationship between parameters, a large number of parameters, and a large number of test rounds, the probabilistic random forest is selected. When the data type of the sampling space is floating-point and integer parameters, an interleaved local and random search strategy is adopted.

[0067] In the prior art, there is a reference method 1: OpenBox algorithm. This algorithm can handle generalized black-box optimization (BBO) problems with different objectives and constraints, and the optimization process can be sequential or parallel, and implements variants of competitive methods such as Bayesian optimization, evolutionary algorithm, and CMA-ES; the core of OpenBox is Optimizer, which accepts the objective function and search space as input, executes the optimization process, and outputs the final result.

[0068] In the prior art, there is a reference algorithm 2: Bayesian Optimization and Hyperband (BOHB). This algorithm combines the prediction capability of Bayesian optimization and the resource allocation efficiency of the hyperband algorithm to achieve efficient search in large-scale hyperparameter space; BOHB uses Bayesian optimization to guide the sampling of different configurations in the hyperband algorithm, so as to quickly locate potential hyperparameter areas under limited resources, which can be applied in HMM parameter adjustment, but this method has some restrictions on the application platform and is more suitable for parallel computing.

[0069] The comparative effect of the present invention is as follows Figure 7 As shown, in the experimental background of single HMM target optimization and adjustment of all HMM hyperparameters, the efficiency of the HMM parameter adjustment effect is demonstrated compared with the reference algorithms Openbox and BOHB. In terms of HMM parameter adjustment effect, the convergence speed and stability of the present invention are superior to the above two reference methods.

[0070] like Figure 8 As shown, the second aspect of the present invention provides a parameter optimization method for map binding, comprising: Step S100: receiving a hyperparameter configuration in a sampling space; Step S200: Optimizing the hyperparameter configuration based on a preset optimization strategy; Step S300: iterate the hyperparameter configuration, and determine the optimal hyperparameter configuration by comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration.

[0071] like Fig. 9 As shown, in one embodiment of the present invention, step S200 includes: Step S210: constructing a proxy model, a collection function and a collection function optimizer; Step S220: collecting performance data of the objective function under the hyperparameter configuration based on the proxy model; Step S230: updating an acquisition function based on the performance data, wherein the acquisition function is used to improve the hyperparameter configuration according to the maximum expectation; Step S240: The acquisition function optimizer searches for an improved solution and randomly selects points in the sampling space for evaluation, and obtains an optimized hyperparameter configuration based on the evaluation results and the acquisition function.

[0072] In one embodiment of the present invention, the proxy model is a probabilistic random forest model, the acquisition function is an expected improvement (EI) method, and the acquisition function optimizer searches for an improved solution based on a random local search method.

[0073] In one embodiment of the present invention, after comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration to determine the optimal hyperparameter configuration, the method further includes: The historical data are recorded and visualized, wherein the historical data at least includes configuration information corresponding to the observation results, the output value of the observation objective function, the test status and the operation time.

[0074] In one embodiment of the present invention, the parameter optimization method for map binding is applicable to a hidden Markov model.

[0075] like Fig.10 As shown, in step S300, the method of comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration to determine the optimal hyperparameter configuration also includes: Step S310: Obtain and compare the binding route ID comparison group with the hidden Markov model binding route ID; Step S320: input the new hyperparameter configuration into the black box objective function model, and output the successful binding ratio value under the new hyperparameter configuration; Step S330: Obtain a ratio value within a preset ratio range as the optimal hyperparameter configuration.

[0076] The present invention adjusts the HMM hyperparameter configuration, improves the accuracy of matching the vehicle position with the road network, and enhances the stability of the HMM when facing different maps.

[0077] The present invention significantly enhances the processing capability of different types of data, such as Boolean quantities, character quantities, real numbers, etc., ensuring the stability and accuracy of the model under various data inputs; The present invention improves the speed of relevant parameter adjustment and data processing, ensures the best performance of HMM in various scenarios, and significantly improves the real-time performance of HMM parameter adjustment and the efficiency of engineering application.

[0078] Compared with traditional grid search or random search, the present invention reduces the mismatch rate and effectively improves robustness and positioning accuracy.

[0079] Based on the same idea as the system in the above embodiment, the parameter optimization method for map binding provided in the present application is implemented by the system in the above embodiment. The specific limitations of each step can be referred to the description of the system and will not be repeated here.

[0080] like Fig.11 As shown, the third aspect of the present invention provides an electronic device, including: At least one processor; and at least one memory in communication with the processor, wherein: the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method described in any of the above embodiments.

[0081] A fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any one of the above embodiments is implemented.

[0082] It can be understood that computer-readable storage media may include: any entity or device capable of carrying a computer program, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc. A computer program includes a computer program code. The computer program code may be in source code form, object code form, an executable file, or some intermediate form, etc. A computer-readable storage medium may include: any entity or device capable of carrying a computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc.

[0083] In certain embodiments of the present invention, the electronic device may include a controller or a processor, and the controller is a single-chip microcomputer chip that integrates a processor, a memory, a communication module, etc. The processor may refer to a processor included in the controller. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.

[0084] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present invention belong.

[0085] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0086] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A parameter optimization system for map binding, characterized in that: include: A sampling space module, at least for receiving hyperparameter configurations; An optimizer module, at least used to optimize the hyperparameter configuration based on a preset optimization strategy; A parameter optimization iteration module, at least used to iterate the hyperparameter configuration; The black box objective function module is used to generate a successful path binding ratio value under a new hyperparameter configuration at least in each iteration; The evaluation update module is used to at least compare the successful path binding ratio value under the new hyperparameter configuration generated by the black box objective function module at each iteration to determine the optimal hyperparameter configuration.

2. The parameter optimization system for map binding according to claim 1 is characterized in that: The sampling space module at least includes a custom input data type submodule, a sampling space submodule and a sampling configuration submodule; wherein: The custom input data type submodule is used at least to input the data type, add the parameter name, and set the upper and lower sampling limits of the reference; The sampling space submodule is at least used to record hyperparameters and their indexes, sample new hyperparameter configurations, obtain new hyperparameter configurations, and align hyperparameter information with configuration information; The sampling configuration submodule is used for at least normalization operation and / or serialization operation hyperparameter configuration and is connected with the optimizer module.

3. The parameter adjustment and optimization system for map binding according to claim 1 is characterized in that: The optimizer module at least includes an optimizer initialization submodule, a proxy model training submodule, a collection function update submodule and a collection function optimizer; wherein: The optimizer initialization submodule is used to construct at least a proxy model, a collection function, and a collection function optimizer; The surrogate model training submodule is at least used to collect performance data of the objective function under the hyperparameter configuration based on the surrogate model; The acquisition function updating submodule is at least used to update the acquisition function based on the performance data, wherein the acquisition function is used to improve the hyperparameter configuration according to the maximized expectation; The acquisition function optimizer is used to at least search for an improved solution and randomly select points in the sampling space for evaluation, and obtain an optimized hyperparameter configuration based on the evaluation results and the acquisition function.

4. The parameter adjustment and optimization system for map binding according to claim 1 is characterized in that: The black box objective function module includes a data alignment submodule and a function encapsulation submodule; wherein: The data alignment submodule is at least used to obtain and compare the binding route ID comparison group with the hidden Markov model binding route ID; The function encapsulation submodule is at least used to input the new hyperparameter configuration into the black box objective function model, and output the binding success ratio value under the new hyperparameter configuration.

5. The parameter adjustment and optimization system for map binding according to claim 1 is characterized in that: It also includes a hyperparameter configuration input module, which is at least used to input hyperparameter information configuration into the sampling space module.

6. The parameter adjustment and optimization system for map binding according to claim 1 is characterized in that: It also includes a historical observation recording module, which is at least used to record and visualize historical data, wherein the historical data at least includes configuration information corresponding to the observation results, the output value of the observation objective function, the test status and the operation time.

7. The parameter adjustment and optimization system for map binding according to claim 6 is characterized in that: It also includes an initialization module, which is used at least to selectively establish the sampling space module, the settings of the optimizer module and the basic settings in the historical observation record module.

8. A parameter optimization method for map binding, characterized in that: include: receiving a hyperparameter configuration in a sampling space; Optimizing the hyperparameter configuration based on a preset optimization strategy; The hyperparameter configuration is iterated, and the optimal hyperparameter configuration is determined by comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration.

9. The parameter optimization method for map binding according to claim 8 is characterized in that: Optimizing the hyperparameter configuration based on a preset optimization strategy includes: Build proxy models, acquisition functions, and acquisition function optimizers; Collect performance data of the objective function under hyperparameter configuration based on the surrogate model; updating an acquisition function based on the performance data, wherein the acquisition function is used to improve the hyperparameter configuration according to the maximized expectation; The acquisition function optimizer searches for improved solutions and randomly selects points in the sampling space for evaluation, and obtains the optimized hyperparameter configuration based on the evaluation results and the acquisition function.

10. The parameter optimization method for map binding according to claim 9 is characterized in that: The proxy model is a probabilistic random forest model, the acquisition function is an expected improvement method, and the acquisition function optimizer searches for an improved solution based on a random local search method.

11. The parameter optimization method for map binding according to claim 8, characterized in that: After determining the optimal hyperparameter configuration by comparing the successful path binding ratio values ​​under the new hyperparameter configuration generated in each iteration, the following is also included: The historical data are recorded and visualized, wherein the historical data at least includes configuration information corresponding to the observation results, the output value of the observation objective function, the test status and the operation time.

12. The parameter optimization method for map binding according to claim 8, characterized in that: The method for optimizing the parameters of map binding is applicable to a hidden Markov model; the method determines the optimal hyperparameter configuration by comparing the successful ratio of binding under the new hyperparameter configuration generated in each iteration, and further includes: Obtain and compare the binding route ID comparison group with the hidden Markov model binding route ID; Input the new hyperparameter configuration into the black box objective function model and output the successful binding ratio value under the new hyperparameter configuration; Get the ratio value within the preset ratio range as the optimal hyperparameter configuration.

13. An electronic device, characterized in that: include: at least one processor; And at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the parameter optimization method for map binding as described in any one of claims 8 to 12.

14. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is run by a computer, the parameter adjustment and optimization method for map binding as described in any one of claims 8 to 12 is executed.

Images