Coarse-to-fine-grained time sequence data simplification method, system, equipment and medium

Abstract

The invention relates to the technical field of time series data simplification, in particular to a coarse-to-fine-grained time series data simplification method, system and device and a medium. The method comprises the following steps: identifying anchor points having key influence on a sequential sequence space structure, and generating a pre-simplified sequential sequence; on the basis of the generated pre-simplified sequential sequence, feature recognition is carried out, potential key areas are quantified, and a candidate point set is generated; and constructing a dynamic decision framework, and screening an optimal key point in the candidate set to gradually optimize the time sequence. According to the method, the simplification efficiency and precision are remarkably improved through three stages. According to the method, an improved Douglas-Peucker algorithm (RDP) is introduced to serve as initialization of coarse-grained simplification, an original time sequence is traversed at a time, a main salient point is dynamically selected, the problem of low-efficiency initialization of a traditional method is avoided through an adaptive threshold strategy, and time complexity can be effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of time series data simplification, and in particular to a method, system, device and medium for simplification of time series data from coarse granularity to fine granularity. Background Art

[0002] With the widespread adoption of mobile internet and various smart devices, time series data has become an indispensable component of modern data processing, widely used in fields such as traffic monitoring, logistics scheduling, environmental monitoring, and behavioral analysis. However, large-scale, high-dimensional time series data poses severe challenges to storage management and processing efficiency. Time series simplification, as a key technology, can effectively address these issues. It significantly reduces storage and computational burdens by strategically removing redundant points from the original time series data while retaining the data's key features and utility. Currently, it is one of the most common lossy compression methods, capable of achieving high compression while maintaining acceptable information loss.

[0003] Existing time series simplification algorithms fall into two main categories: error-bounded algorithms and error-driven algorithms. Our research focuses on the latter, which involves selecting a fixed number of key data points from the original time series data to construct sub-time series within a given "budget" constraint, and striving to minimize the overall error in the time series simplification. Currently, a variety of algorithms have been proposed to address this problem, including heuristic-based and learning-based methods. While progress has been made in some scenarios, the following challenges remain:

[0004] (1) Inefficient initialization strategy. Most existing algorithms construct an initial baseline by connecting the start and end points of the time series, which serves as the starting point for time series simplification. Although this strategy is conceptually simple and intuitive, it has significant shortcomings in computational efficiency. Since multiple iterations are required to gradually meet the preset storage budget, the algorithm's running time will increase significantly, thus limiting its applicability in large-scale time series data processing.

[0005] (2) Fixed error thresholds are not adaptable enough. They usually rely on a fixed error threshold and uniformly apply the same simplification criteria to all time segments. This rigid strategy is difficult to flexibly adapt to the complexity of the time segments, resulting in simplification results that are often lower than expected. In addition, it is extremely challenging for users to determine an appropriate threshold for different scenarios, which not only limits the universality of the algorithm but may also result in significant performance differences in different scenarios. More importantly, these manually set rules lack solid theoretical support and are difficult to provide sufficient guarantees for the effectiveness of the algorithm, which has raised questions about its reliability.

[0006] (3) Limitations of a single distance metric. Previous studies have primarily emphasized distance-based metrics, which are effective in preserving the geometric structure of time series data. However, this singular focus on distance often overlooks other key features of time series data, such as directionality, which is equally important for capturing the behavioral patterns of moving objects. This one-sided metric often results in poor performance of time series simplification algorithms in complex scenarios, making them difficult to cope with the diverse demands of practical applications. Summary of the Invention

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

[0008] In a first aspect, the present invention provides a method for simplifying time series data from coarse-grained to fine-grained, comprising:

[0009] Identify the anchor points that have a key impact on the spatial structure of the time series and generate pre-simplified time series;

[0010] Based on the generation of pre-simplified time series, feature recognition is performed, potential key areas are quantified, and candidate point sets are generated;

[0011] Build a dynamic decision-making framework to screen the optimal key points in the candidate set and gradually optimize the time series.

[0012] As a preferred solution of the method for simplifying time series data from coarse-grained to fine-grained according to the present invention, the step of identifying anchor points that have a key impact on the spatial structure of the time series sequence and generating a pre-simplified time series sequence includes:

[0013] Take the original time series data as input and extract key data points;

[0014] After receiving the original time series data and the threshold parameter, a first algorithm is executed to generate pre-simplified time series data.

[0015] As a preferred solution of the method for simplifying time series data from coarse-grained to fine-grained according to the present invention, the method of generating a pre-simplified time series sequence, performing feature recognition, quantifying potential key areas, and generating a candidate point set includes:

[0016] Calculate the smoothness of each segment on the pre-simplified time series and divide the candidate regions according to the smoothness;

[0017] Compute directional difference measures;

[0018] The smoothness and direction error calculation results of the input time series segment are used to generate a candidate point set.

[0019] As a preferred solution of the method for simplifying time series data from coarse-grained to fine-grained according to the present invention, the step of constructing a dynamic decision framework includes:

[0020] Build a decision framework based on the inherent temporal dependencies of time series data and the sequential decision-making characteristics of simplified tasks;

[0021] Through the decision framework, the optimal key points are screened in the candidate set and the time series is gradually optimized.

[0022] As a preferred solution of the method for simplifying time series data from coarse-grained to fine-grained according to the present invention, wherein: after receiving the original time series data and the threshold parameter, executing the first algorithm to generate the pre-simplified time series data includes:

[0023] Initialize parameters and data structures to construct initial simplified time series data;

[0024] Start traversing from the starting point of the time series and check whether each point in the time series is a main significant point;

[0025] After all time series data points have been examined, determine whether they can be discarded or retained.

[0026] As a preferred solution of the method for simplifying time series data from coarse-grained to fine-grained according to the present invention, the decision framework includes:

[0027] Starting from the root node, iteratively select the node with the largest utility value among the child nodes until an incompletely expanded node is encountered; for the incompletely expanded node, randomly select an unexplored candidate point to generate a new child node; starting from the newly expanded node, randomly select candidate points for path simulation until the compression rate constraint is met and a leaf node is generated; the simulation results are backpropagated from the leaf node to the root node, and the benefits and access frequencies of all nodes on the path are updated; when the number of iterations reaches the preset value or the compression rate constraint is met, the search is stopped, and the timing state corresponding to the path with the highest benefit is selected as the final simplified result.

[0028] As a preferred solution of the method for simplifying time series data from coarse-grained to fine-grained according to the present invention, wherein: through the decision framework, the optimal key points are screened in the candidate set and the time series sequence is gradually optimized, including:

[0029] Through the objective function S v =-λ f f(X,X' C )+L reg (X' C ) calculation, where f(X,X' C ) To simplify the timing error, L reg (X' C )=λ uni L uni (X' C )+λ dir L dir (X' C) is a regularization loss that includes spatial uniformity and directional consistency;

[0030] Record the number of times a node is visited during the search process;

[0031] Calculated according to the upper confidence bound formula:

[0032]

[0033] Among them, λ is the balance factor, N(v f ) is the number of parent node visits.

[0034] In a second aspect, the present invention provides a system for reducing time series data from coarse-grained to fine-grained, comprising:

[0035] The coarse-grained simplification module adopts an improved screening strategy based on the DP algorithm. It dynamically identifies anchor points (such as turning points and high curvature points) that have a key impact on the spatial structure of the time series through adaptive thresholds, generates pre-simplified time series and transmits them to downstream modules, providing an adaptive, heuristic-driven starting point.

[0036] The candidate point set exploration module takes into account a wider range of features based on the coarse-grained simplification results, including temporal sequence smoothness and directional error, and constructs a weighted scoring model to quantify potential key areas and generate candidate point sets, thereby balancing the requirements of retaining geometric structure and directional features, and solving the limitations of a single distance metric.

[0037] The fine-grained optimization module constructs a dynamic decision-making framework based on the Monte Carlo tree to address the inherent temporal dependencies of time series data and the sequential decision-making characteristics of simplified tasks. It quantifies the potential gains of the candidate point set through an immediate reward function, and gradually optimizes the time series by screening the optimal key points in the candidate set.

[0038] In a third aspect, the present invention provides an electronic device, comprising:

[0039] memory and processor;

[0040] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for simplification of time series data from coarse-grained to fine-grained are implemented.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the coarse-grained to fine-grained time series data simplification method.

[0042] Compared with existing technologies, the present invention significantly improves the efficiency and accuracy of simplification through three phases. This method introduces a modified Douglas-Peucker algorithm (RDP) as the initialization for coarse-grained simplification, traversing the original time series in a single pass to dynamically select key salient points. This method avoids the inefficient initialization problem of traditional methods through an adaptive thresholding strategy, effectively reducing time complexity. Furthermore, this method performs multi-feature analysis on the coarse-grained simplification results, combining the smoothness of the time series segments (a measure of path curvature) and directional error (a quantification of angular deviation). Key candidate points are selected through standardized and weighted scoring to establish a point set. This replenishes important feature points missed in the coarse-grained phase, overcomes the limitation of relying solely on geometric distance, and ensures the preservation of time series behavior patterns (such as turning features). Finally, in the fine-grained optimization phase, the method iteratively optimizes the candidate point set based on Monte Carlo tree search. The time series data is modeled as a state tree, and the optimal candidate points are gradually added to the simplified time series. This phase overcomes the constraints of fixed thresholds and combines multi-objective regularization (uniform spacing and directional consistency) to globally minimize information loss under compression rate constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is the basic framework of the algorithm of the coarse-grained to fine-grained time series data simplification method described in one embodiment of the present invention.

[0045] Figure 2 Schematic diagram of the comparison of simplification errors (under the PED indicator) of the coarse-grained to fine-grained time series data simplification method on two data sets according to an embodiment of the present invention.

[0046] Figure 3 This is a schematic diagram of the comparison of simplification errors (under the SED indicator) on two data sets of the coarse-grained to fine-grained time series data simplification method described in one embodiment of the present invention.

[0047] Figure 4 This is a schematic diagram of a specific embodiment of the method for simplification of time series data from coarse-grained to fine-grained according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0049] Example 1, reference Figure 1 , as one embodiment of the present invention, provides a method for simplifying time series data from coarse-grained to fine-grained, including:

[0050] In this paper, a two-stage time series data simplification framework is designed to overcome the above challenges. The basic framework of the algorithm is as follows: Figure 1 As shown in Figure 1, it mainly includes three modules: coarse-grained simplification, candidate point set exploration, and fine-grained optimization. This method is a fully adaptive temporal simplification algorithm that significantly improves simplification accuracy while ensuring computational efficiency through a three-stage progressive optimization architecture. Its core steps are as follows:

[0051] S1: Utilizes an improved screening strategy based on the DP algorithm. Using adaptive thresholds, it dynamically identifies anchor points (such as turning points and points of high curvature) that have a critical impact on the spatial structure of the time series. This strategy then generates a pre-simplified time series and transmits it to downstream modules. This screening strategy overcomes the limitations of fixed error thresholds and allows for flexible adjustments, adapting well to the complexity of time series segments.

[0052] S2: Based on the coarse-grained simplification results, the candidate point set exploration module considers a wider range of features, including temporal sequence smoothness and directional error, and constructs a weighted scoring model to quantify potential key areas and generate candidate point sets, thereby balancing the requirements of preserving geometric structure and directional features, and addressing the limitations of a single distance metric.

[0053] S3: To address the inherent temporal dependencies of time series data and the sequential decision-making characteristics of the simplification task, the fine-grained optimization module constructs a dynamic decision-making framework based on Monte Carlo trees. This framework quantifies the potential gain of a candidate point set using an immediate reward function, selects the optimal key points within the candidate set, and gradually optimizes the initial simplified time series. The pre-simplified time series provides an adaptive, heuristic-driven starting point for this step, addressing the inefficiencies associated with traditional initialization strategies.

[0054] The above steps form a progressive optimization chain: coarse-grained simplification provides an efficient starting point, candidate point exploration compensates for missed features, and fine-grained search improves global accuracy. These three stages work together to reduce storage and computational burdens while ensuring the integrity of the geometric, directional, and behavioral characteristics of time series data. Each module has a clear division of labor and orderly interactions.

[0055] Example 2, reference Figure 2-Figure 3 , is an embodiment of the present invention, based on the above embodiments, provides a method for simplifying time series data from coarse-grained to fine-grained.

[0056] In the embodiment of the present application, identifying anchor points that have a key impact on the spatial structure of the time series sequence in step S1 and generating a pre-simplified time series sequence specifically includes the following steps A1-A3:

[0057] A1: The identifying of anchor points that have a key impact on the spatial structure of the time series sequence and generating a pre-simplified time series sequence includes:

[0058] A2: Take the original time series data as input and extract key data points;

[0059] A3: After receiving the original time series data and the threshold parameter, execute the first algorithm to generate pre-simplified time series data.

[0060] In the embodiment of the present application, the first algorithm adopts an improved method based on the Douglas-Peucker compression algorithm, namely the RDP algorithm.

[0061] The specific implementation process of steps A1-A3 is: using the original time series data as input, the aim is to efficiently and preliminarily extract some key data points from the original time series, providing an efficient and accurate starting point for the subsequent fine optimization stage. By defining the concept of main salient points, an improved method based on the Douglas-Peucker compression algorithm, the RDP algorithm, is proposed, which can quickly identify points that have an important impact on the geometric structure of the time series data, avoiding the problem of low computational efficiency caused by multiple iterations in traditional methods. After receiving the original time series data and threshold parameters, the module executes the RDP algorithm, which only requires a single traversal to quickly generate pre-simplified time series data, significantly improving the computational efficiency, and sending the results to the other two modules. The main calculation steps of the first algorithm are as follows:

[0062] 1.1 Initialization parameters and data structure. Input original time series X = {x i}, define the vertical distance threshold δ, and initialize an empty primary salient point list PrimaryList.

[0063] 1.2 Construct the initial simplified time series data. Set the starting point x of the time series X s and the end point x e Connect them into a straight line segment as the initial baseline.

[0064] 1.3 Detection of main salient points. Starting from the starting point of the time series, check whether each point on the time series X is a main salient point: for each x i , calculate the connection line from the beginning to the end of the current time segment If d>δ, perform the following steps:

[0065] 1.3.1 Mark the point as the primary salient point and add it to the list PrimaryList.

[0066] 1.3.2 Update the starting point of the original curve to the main significant point, and form a new curve with the end point of the original curve, and then search whether there is a main significant point in the new curve.

[0067] If there is no main significant point in the curve, it is considered that all points except the starting point and the end point in the curve can be discarded.

[0068] 1.4 Output pre-simplified time series. Repeat the process of 1.3 until all time series data points have been checked, and then determine whether they can be discarded or retained. The points stored in the PrimaryList list are the key points extracted by the main salient point detection module. Connect these points in sequence to form the pre-simplified time series data X. C , as input to the subsequent fine optimization stage.

[0069] In another optional embodiment, the first algorithm may also be replaced by a traditional DP algorithm.

[0070] In another optional embodiment, the first algorithm may also be replaced by a traditional DP-hull algorithm.

[0071] Given time series data and an error threshold, these two algorithms construct a discrete state space and narrow the search range based on the convex hull principle. Although computational efficiency is lower than that of the improved algorithm, they can still obtain suboptimal solutions that meet the constraints by traversing the state transition path. Both algorithms must be used in conjunction with the candidate point set exploration module and back-end optimization of the present invention to complete the complete generation process of time series simplified time series.

[0072] In the embodiment of the present application, the recognition in step S2 is based on generating a pre-simplified time series, performing feature recognition, quantifying potential key areas, and generating a candidate point set, which specifically includes the following steps B1-B3:

[0073] B1: Calculate the smoothness of each segment in the pre-simplified time series and divide the candidate regions according to the smoothness;

[0074] B2: Calculate the direction difference measure;

[0075] B3: Input the smoothness and direction error calculation results of the time series segment to generate a candidate point set.

[0076] The specific implementation of steps B1-B3 is:

[0077] The candidate point set exploration module aims to identify X CThe relatively important points not included in the (output of the coarse-grained simplification module) are further screened for candidate points with significant geometric and directional features, providing an accurate candidate set for subsequent fine-grained optimization. After receiving the pre-simplified time series, this module dynamically evaluates the potential key areas in the time series data by integrating the dual indicators of time series smoothness and directional error, and generates a candidate point set, providing the results to the fine-grained optimization module for supplementary optimization. The main calculation steps of this module are as follows:

[0078] Calculate the pre-simplified timing X C Each sequence on Smoothness:

[0079]

[0080] The numerator is the cumulative Euclidean distance between adjacent points in the original time series data segment, and the denominator is X C upper section A smoothness value close to 1 indicates that the segment is approximately a straight line and does not need to be optimized; a value significantly greater than 1 indicates that there are sharp turns or complex paths and needs to be marked as a candidate area.

[0081] Calculate timing segments Directional error on .

[0082] Direction difference metric. Define the minimum rotation angle difference between two direction angles α and β:

[0083] diff(α,β)=min(|α-β|,2π-|α-β|)

[0084] Direction definition. For each adjacent point x in the original time series data segment k ,x k+1 , calculate its direction angle dir(x k x k+1 ), that is, vector Counterclockwise angle with the positive X axis

[0085] Calculate each segment of the pre-simplified timing First, calculate the direction angle dir(x k x k+1 ) and the segment direction angle dir(x i x j ) between the direction differences. Then the mean of all the direction differences in the segment is the direction error of the segment:

[0086]

[0087] The larger the direction error, the more significant the deviation of the direction of the pre-simplified segment from the original time series data.

[0088] The smoothness and direction error calculation results of each time segment are input to generate a candidate point set.

[0089] Feature normalization. For smoothness and direction error∈(x i ,x j ) performs Z-score normalization to eliminate dimensional differences and obtain S′ and D′.

[0090] Weighted scoring. Define the comprehensive scoring formula: Score = W s ×S'+W d ×D', where W s With W d are preset weights, reflecting the importance of smoothness and direction error respectively.

[0091] Key segment selection: Sort by score from high to low, select the first K time series data segments (K = |T| × CR, CR is the compression ratio), and add the corresponding original time series data points that are not selected by the RDP algorithm to the candidate set C.

[0092] The obtained candidate point set C is sent to the fine-grained optimization module.

[0093] In the implementation of this application, the dynamic decision framework is constructed in S3, and the optimal key points are screened in the candidate set and the timing sequence is gradually optimized, including the following specific steps C1-C2.

[0094] C1: Build a decision framework based on the inherent temporal dependencies of time series data and the sequential decision-making characteristics of simplified tasks;

[0095] C2: Through the decision framework, the optimal key points are selected from the candidate set to gradually optimize the time series.

[0096] In an embodiment of the present application, the decision framework dynamically evaluates candidate points through Monte Carlo Tree Search (MCTS) and selects the optimal subset to minimize the simplification error.

[0097] The specific implementation process of steps C1-C2 is as follows:

[0098] The fine-grained optimization module performs fine-grained optimization on the pre-simplified sub-sequences output by the previous two modules, dynamically evaluates candidate points through Monte Carlo Tree Search (MCTS), and selects the optimal subset to minimize the simplification error. During the optimization process, the module first models the pre-simplified sub-sequence as a state tree, where each node represents a time series state containing the currently selected key point, and the entire tree reflects the dynamic search process from the initial state to the final optimization result. In order to further improve the optimization efficiency, it gradually selects key points from the point set C generated by the candidate point exploration module, so that the key geometric and directional features of the time series can be retained. The main calculation steps of this module are as follows:

[0099] The time series data state is modeled as a tree structure. The root node represents the pre-simplified time series X filtered out in the granularity stage. C The .subsections represent the new temporal states generated by gradually adding candidate points.

[0100] Define node properties, where each node contains the following key properties:

[0101] Node benefit. Through the objective function S v =-λ f f(X,X' C )+L reg (X' C ) calculation, reflecting the simplified performance of the current time series state. C ) To simplify the timing error, L reg (X' C )=λ uni L uni (X' C )+λ dir L dir (X' C ) is a regularization loss that includes spatial uniformity and directional consistency.

[0102] Visit frequency. Records the number of times a node is visited during the search process, used to balance exploration and utilization.

[0103] Node utility. Combining revenue and access frequency can effectively explore the search space while achieving the global optimal solution. Calculated according to the upper confidence bound (UCB) formula: Where λ is the balance factor, N(v f ) is the number of parent node visits.

[0104] The Monte Carlo tree search begins in four steps:

[0105] Starting from the root node, iteratively select the node with the largest utility value among the child nodes until an incompletely expanded node is encountered.

[0106] For nodes that are not fully expanded, an unexplored candidate point is randomly selected to generate a new child node.

[0107] Starting from the newly expanded node, candidate points are randomly selected for path simulation until the compression rate constraint is met and a leaf node is generated.

[0108] The simulation results (such as simplified error and regularization loss) are back-propagated from the leaf node to the root node to update the benefits and visit frequencies of all nodes on the path.

[0109] The search stops when the number of iterations reaches a preset value (e.g., 1,000) or the compression rate constraint is met. The timing state corresponding to the path with the highest gain is selected as the final simplified result X'.

[0110] In another optional embodiment, the decision framework can be implemented alternatively using an agent based on reinforcement learning. The agent learns the mapping relationship between environmental state characteristics and simplified decisions through an offline training phase. During online reasoning, it dynamically outputs the decision action of whether to retain or delete the node based on the curvature characteristics, direction characteristics and preset simplification threshold of the current time series node. Specifically, the agent's policy network takes the local time series data fragment composed of the previous and next three-order nodes as input, extracts the geometric constraint features, and combines the reward function to perform multi-objective optimization on the time series smoothness error, simplification rate and energy consumption indicators, thereby generating the final simplified time series sequence.

[0111] DP-Hull, Error-Search and RLTS are also three effective time series data simplification algorithms. Therefore, the present invention compares the simplification quality ( Figure 2 and Figure 3 ) and simplified efficiency (Appendix Table 1) were experimentally compared, and the results proved the effectiveness and reliability of the present invention.

[0112] Appendix 1 is:

[0113]

[0114] Among them, PED and DAD are two commonly used evaluation indicators, which are used to measure the degree of error between the simplified sequence and the original sequence.

[0115] In Example 3, the above is a schematic scheme of a method for simplification of time series data from coarse-grained to fine-grained. It should be noted that the technical scheme of the system for simplification of time series data from coarse-grained to fine-grained is the same as the technical scheme of the method for simplification of time series data from coarse-grained to fine-grained. For details not described in detail in the technical scheme of the system for simplification of time series data from coarse-grained to fine-grained in this embodiment, please refer to the description of the technical scheme of the method for simplification of time series data from coarse-grained to fine-grained.

[0116] This embodiment also provides a system for reducing time series data from coarse-grained to fine-grained, including:

[0117] The coarse-grained simplification module adopts an improved screening strategy based on the DP algorithm. It dynamically identifies anchor points (such as turning points and high curvature points) that have a key impact on the spatial structure of the time series through adaptive thresholds, generates pre-simplified time series and transmits them to downstream modules, providing an adaptive, heuristic-driven starting point.

[0118] The candidate point set exploration module takes into account a wider range of features based on the coarse-grained simplification results, including temporal sequence smoothness and directional error, and constructs a weighted scoring model to quantify potential key areas and generate candidate point sets, thereby balancing the requirements of retaining geometric structure and directional features, and solving the limitations of a single distance metric.

[0119] The fine-grained optimization module constructs a dynamic decision-making framework based on the Monte Carlo tree to address the inherent temporal dependencies of time series data and the sequential decision-making characteristics of simplified tasks. It quantifies the potential gains of the candidate point set through an immediate reward function, and gradually optimizes the time series by screening the optimal key points in the candidate set.

[0120] This embodiment also provides an electronic device suitable for simplifying time series data from coarse-grained to fine-grained, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method for simplifying time series data from coarse-grained to fine-grained as proposed in the above embodiment.

[0121] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the method for simplification of time series data from coarse-grained to fine-grained as proposed in the above embodiment.

[0122] The storage medium proposed in this embodiment and the method for implementing coarse-grained to fine-grained time series data simplification proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0123] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general hardware, and of course can also be implemented by hardware. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0124] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0125] Example 4, as Figure 4 As shown, the present invention provides an efficient simplification method for time series data, which is divided into three ordered stages.

[0126] Next, we will combine a specific embodiment and Figure 4 The present invention is further described in detail in Table 2. It includes a time series data X={x1, x2, ... x8} with 8 data records. The feature space of each data is two-dimensional and is expressed as (lon i ,lat i ), representing its geographical location.

[0127] As shown in Appendix 2:

[0128]

[0129] 1. In the coarse-grained stage, input the time series data X and the distance threshold δ = 1. The RDP algorithm first connects the starting point x1 and the end point x8 into a straight line segment As the initial baseline. Traverse each point on the time series data X and calculate the vertical distance from each point to the initial baseline. The vertical distance of point x4 is 1.13, which is greater than the threshold δ=1, so x4 is identified as the primary salient point. Add x4 to PrimaryList and use x4 as the new starting point to form a new baseline with the end point x8. Repeat the above process until the traversal is completed, and finally generate the pre-simplified time series X C={x1,x4,x8}.

[0130] 2. Candidate point exploration stage. C Smoothness and direction error are calculated for each time segment of . The smoothness is 1.11 and the direction error is 0.42. The smoothness is 1.25 and the direction error is 0.58. Based on the comprehensive score of smoothness and direction error, select the key points x2 and x6 that are not selected by RDP from the original time series X. Add them to the candidate point set

[0131] 3. Fine-grained stage, pre-simplify the timing X C As the root node of the state tree, the Monte Carlo tree search is started. After multiple iterations (until the storage operation w = 4 is satisfied), MCTS gradually optimizes the timing and finally selects the optimal simplified timing: X' = {x1, x4, x6, x8}

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for simplifying time series data from coarse-grained to fine-grained, characterized in that: include: identifying anchor points that have a key impact on the spatial structure of the time series, and generating a pre-simplified time series by a first algorithm; Based on the generated pre-simplified time series, feature recognition is performed, potential key areas are quantified, and candidate point sets are generated; Build a dynamic decision-making framework to screen the optimal key points in the candidate set and gradually optimize the time series.

2. The method for simplifying time series data from coarse-grained to fine-grained according to claim 1, wherein: The identifying of anchor points having a key impact on the spatial structure of the time series sequence and generating a pre-simplified time series sequence by a first algorithm includes: Take the original time series data as input and extract key data points; After receiving the original time series data and the threshold parameter, a first algorithm is executed to generate pre-simplified time series data.

3. The method for simplifying time series data from coarse-grained to fine-grained according to claim 2, wherein: The method of generating a pre-simplified time series, performing feature recognition, quantifying potential key areas, and generating a candidate point set includes: Calculate the smoothness of each segment on the pre-simplified time series and divide the candidate regions according to the smoothness; Compute directional difference measures; The smoothness and direction error calculation results of the input time series segment are used to generate a candidate point set.

4. The method for simplifying time series data from coarse-grained to fine-grained according to claim 3, wherein: The construction of a dynamic decision-making framework includes: Build a decision framework based on the inherent temporal dependencies of time series data and the sequential decision-making characteristics of simplified tasks; Through the decision framework, the optimal key points are screened in the candidate set and the time series is gradually optimized.

5. The method for simplifying time series data from coarse-grained to fine-grained according to claim 4, wherein: The pre-simplified timing sequence includes: Initialize parameters and data structures to construct initial simplified time series data; Start traversing from the starting point of the time series and check whether each point in the time series is a main significant point; After all time series data points have been examined, determine whether they can be discarded or retained.

6. The method for simplifying time series data from coarse-grained to fine-grained according to claim 5, wherein: The decision-making framework includes, Starting from the root node, iteratively select the node with the largest utility value among the child nodes until an incompletely expanded node is encountered; for the incompletely expanded node, randomly select an unexplored candidate point to generate a new child node; Starting from the newly expanded node, candidate points are randomly selected for path simulation until the compression rate constraint is met and a leaf node is generated. The simulation results are backpropagated from the leaf node to the root node, and the benefits and access frequencies of all nodes on the path are updated. The search is stopped when the number of iterations reaches the preset value or the compression rate constraint is met, and the timing state corresponding to the path with the highest benefit is selected as the final simplified result.

7. The method for simplifying time series data from coarse-grained to fine-grained according to claim 6, wherein: Through the decision framework, the optimal key points are screened in the candidate set and the time series is gradually optimized, including: Through the objective function S v =-λ f f(X,X' C )+L reg (X' C ) calculation, where f(X,X' C ) To simplify the timing error, L reg (X' C )=λ uni L uni (X' C )+λ dir L dir (X' C ) is a regularization loss that includes spatial uniformity and directional consistency; Record the number of times a node is visited during the search process; Calculated according to the upper confidence bound formula: Among them, λ is the balance factor, N(v f ) is the number of parent node visits.

8. A system for reducing time series data from coarse-grained to fine-grained, applying the method according to any one of claims 1 to 7, characterized in that: include: Coarse-grained simplification module, candidate point set exploration module, and fine-grained optimization module.

9. An electronic device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the coarse-grained to fine-grained time series data simplification method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the coarse-grained to fine-grained time series data simplification method according to any one of claims 1 to 7.

Images