High-speed rail line along the wind process division-based second-level extreme wind speed prediction method and system

Abstract

The application discloses a high-speed railway line along the wind process division-based second-level extreme wind speed prediction method and system, which comprises the following steps: a high-speed railway wind speed second-level granularity significant feature point extraction algorithm is used to extract initial significant feature points of a second-level wind speed observation sequence along the high-speed railway line and remove transitional peak values, so that a final significant feature point set is obtained; wind processes are classified according to the set, and a wind process database is established; a similarity optimization dynamic time adjustment algorithm is designed; the algorithm is used to match wind speed observation sequences with a similarity of more than 90% in the wind process database; the sequences are taken as a training set and input into a TCN-BiGRU prediction model; and the prediction model parameters are optimized through a Bayesian optimization algorithm to realize wind speed prediction. The application effectively processes complex and nonlinear modes in the second-level wind signal along the high-speed railway line, and improves the accuracy and adaptability of the prediction.

Description

Technical Field

[0001] This invention relates to the field of time series forecasting, and in particular to a method and system for predicting second-level extreme wind speeds along high-speed railway lines based on wind process division. Background Technology

[0002] Currently, the high-speed rail disaster prevention system has real-time wind monitoring and alarm functions, but it lacks early warning capabilities, resulting in insufficient disaster warning capabilities. If the system could predict wind speed values ​​for a period of time in the future based on current wind speed data, it would provide dispatchers and drivers with a longer window to respond to strong winds and make advance preparations for emergency response, thus better ensuring the safety of high-speed rail operations.

[0003] Although scholars both domestically and internationally have conducted extensive research on wind speed prediction, there is currently a lack of studies specifically addressing second-level granularity predictions along railway lines. Furthermore, the complex terrain and frequent strong winds along high-speed rail routes make it difficult to capture these characteristics. Therefore, the multi-step prediction lag issue is an inevitable challenge in multi-step time series prediction of wind speed along railway lines, and resolving this problem has become a bottleneck. Existing models are mostly based on the decomposition of wind signals and algorithmic improvements, rarely considering the inherent characteristics of second-level wind signals along the railway line. Simultaneously, existing predictions often focus on time series composed of single-point values, neglecting the correlation between adjacent data points and the fluctuation characteristics of a cohesive whole formed by multiple points over a period of time. Moreover, the research methods of scholars both domestically and internationally focus on historical data to infer future wind speeds, ignoring the compatibility issues between training and testing samples.

[0004] Therefore, how to develop a new second-level wind speed prediction method along high-speed railways by combining the characteristics of wind speed fluctuations, and how to match a more compatible training set with the data to be predicted to obtain more accurate prediction results is an important problem that needs to be solved. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method and system for predicting extreme wind speeds along high-speed railway lines at the second level based on wind process division. The wind fluctuation unit is divided according to the fluctuation characteristics, amplitude changes and trend transfer between different fluctuation segments, which greatly promotes the accurate matching of wind processes and the optimization of training set selection.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] The present invention proposes a method for predicting second-level extreme wind speeds along high-speed railway lines based on wind process division, comprising:

[0008] S1. The second-level wind signals along the high-speed railway line exhibit drastic and highly random variations, with numerous transitional peaks within a single fluctuation cycle, severely impacting the accuracy of wind process segmentation. The High-Speed ​​Railway Wind Speed ​​SFPE (Second-level significant feature point extraction algorithm) is used to extract initial significant feature points from the second-level wind speed observation sequence along the high-speed railway line and remove transitional peaks, obtaining the final set of significant feature points.

[0009] S2. Based on the final set of significant feature points in step S1, and considering the fluctuation characteristics, amplitude changes, and trend transitions between different fluctuation segments of the fluctuation unit, wind fluctuation processes with similar wind speed change trends and similar fluctuation profiles are classified into one category. The categories include continuous upwind processes, continuous downwind processes, multi-peak wind processes, continuous peak wind processes, and low wind speed wind processes. A wind process database is then established.

[0010] S3. Due to the frequent and significant fluctuations in second-level wind signals along high-speed railway lines, the classic DTW algorithm for similarity matching reveals that calculating the distance between the original points alone cannot capture the shape information of the two sub-time series, leading to erroneous matching due to inconsistent peak and trough trends. Therefore, the SO-DTW (Similarity Optimization-Dynamic Time Wrapping) algorithm is designed and used for wind speed observation sequence matching in the wind process database.

[0011] S4. The wind speed observation sequence obtained in step S3 is used as a training set and input into the TCN-BiGRU prediction model. The prediction model parameters are optimized by Bayesian optimization algorithm to achieve wind speed prediction.

[0012] Furthermore, in step S1, obtaining the final set of salient feature points includes the following sub-steps:

[0013] S101, Order This represents a wind speed observation sequence. ,in, This represents the i-th wind speed observation point. This represents the nth wind speed observation point, where n represents the total number of wind speed observation points.

[0014] Define the sliding window as starting from the first... A subsequence of length w starting from a given wind speed observation point, specifically expressed as:

[0015] ;

[0016] in, Indicates a sliding window. satisfy .

[0017] Based on the sliding window, define the set of local salient feature points, specifically expressed as:

[0018] ;

[0019] ;

[0020] in, This represents the k-th wind speed observation point. This represents the (k-1)th wind speed observation point. This represents the (k+1)th wind speed observation point. M represents the set of locally significant feature points; j This represents the j-th point in the set of local salient feature points.

[0021] S102, Calculation The independent index of each local salient feature point is used. The larger the independent index, the more important the point is in the overall signal. The specific formula is as follows:

[0022] ;

[0023] in, Independent index representing local salient feature points, This represents the absolute value of the distance between adjacent data. express The adjacent local salient feature point on the left, express The adjacent local salient feature point on the right.

[0024] S103, Calculation The weighted entropy of each local salient feature point in the wind speed observation sequence is calculated using the following formula:

[0025] ;

[0026] ;

[0027] in, This represents the sequence consisting of s wind speed observation points to the right of the wind speed observation sequence, where the local salient feature point is located. express Entropy; express At the r-th wind speed observation point, 0 <r<s; express Probability density estimation; This represents the difference between the s wind speed observation points to the right of the local salient feature point in the wind speed observation sequence. Diff represents difference calculation; This represents a sequence of s wind speed observation points to the left of the wind speed sequence, indicating a locally significant feature point. express Entropy; express The Middle Wind speed observation data, ; express Probability density estimation; This represents the difference between the s wind speed observation points to the left of the local salient feature point in the wind speed sequence. .

[0028] S104, Calculation The importance score for each salient feature point is calculated using the following formula:

[0029] ;

[0030] in, This indicates the importance score. When The larger the value, the more important the point is.

[0031] Will Locally significant feature points with a value greater than 1 are defined as initial maximal significant feature points.

[0032] S105. Redefine the set of local salient feature points, with the specific expression as follows:

[0033] ;

[0034] ;

[0035] in, This represents the redefined set of local salient feature points; This represents the j-th point in the redefined set of local salient feature points.

[0036] Repeat steps S102-S104 to... Locally significant feature points with a value less than or equal to 1 are defined as initial minimum significant feature points.

[0037] S106. Label the attributes of the initial maximal and minimum salient feature points. The attribute of the initial maximal salient feature points is 1, and the attribute of the initial minimum salient feature points is -1. When two consecutive points have an attribute of 1, remove the minimum value among the corresponding values ​​of these two points, and the remaining points are the final maximal salient feature points. When two consecutive points have an attribute of -1, remove the maximum value among the corresponding values ​​of these two points, and the remaining points are the final minimum salient feature points. This yields the final set of maximal and minimum salient feature points. This represents the final set of maximal salient feature points. ; This represents the final set of minimal salient feature points. ;in, Let a represent the a-th final maximal salient feature point. Let a represent the a-th final minimal salient feature point. Indicates the first The final maximal salient feature point, Indicates the first The final minimal salient feature point.

[0038] Therefore, the final set of salient feature points is obtained. , .

[0039] Furthermore, in step S2, establishing the wind process database includes the following sub-steps:

[0040] S201. Parameter definition: The attribute of the initial maximum salient feature point is 1, the attribute of the initial minimum salient feature point is -1, and when the attributes of two consecutive points in the final salient feature point are "1, -1" or "-1, 1", the point at this time is a fluctuation point.

[0041] The process from a minimum fluctuation point to a maximum fluctuation point, and then to the next minimum fluctuation point, is called a fluctuation unit.

[0042] The time values ​​corresponding to the final minimum significant feature points on both sides of the final maximum significant feature point are the wave widths of the wave element, specifically expressed as:

[0043] ;

[0044] in, Indicates the wave width of the wave unit. This represents the time value corresponding to the final minimum significant feature point to the right of the final maximum significant feature point. This represents the time value corresponding to the final minimum significant feature point to the left of the final maximum significant feature point.

[0045] The absolute difference between adjacent final maxima and final minima represents the amplitude of the fluctuation unit, specifically expressed as:

[0046] ;

[0047] in, Indicates the amplitude of change of the fluctuation unit. Indicates the first The final maximal salient feature point, Indicates the first The final minimal salient feature point, ; , In the final set of salient feature points, they are adjacent salient feature points.

[0048] S202. Determine the degree of trend change of the fluctuation unit based on the slope of adjacent final maxima salient feature points to guide the division of wind processes. The specific formula is as follows:

[0049] ;

[0050] in, This value represents the degree of trend change within the fluctuation unit. This represents the summation function. Let m represent the m-th final maximal salient feature point. Let m represent the (m-1)th final maximal salient feature point. express At the corresponding time points in the wind speed observation sequence express The corresponding time point in the wind speed observation sequence.

[0051] according to The criteria for trend judgment of fluctuation units are as follows:

[0052] ;

[0053] in, Indicates a trend. Indicates an upward trend. This indicates a downward trend.

[0054] S203. If the continuous fluctuations of the unit remain within the discrimination threshold without any significant increase, this indicates a low-wind-speed process. The wind speed will not rise to the alarm point in a short period. The specific expression for a low-wind-speed process is:

[0055] ;

[0056] in, This indicates a low-speed wind process. This represents all the final salient features during a low-speed wind process. Indicates the discrimination threshold. This represents the final maximal salient feature points closely adjacent to the two ends of a low-wind-speed process. This represents the maximum value of the entire wind speed observation sequence. These represent the start and end points of a low-wind-speed process, respectively.

[0057] Due to wind speed fluctuations, the continuous rising units constitute a sustained upwind process. Within a certain timeframe, there is a possibility that the wind will rise to the alarm point. The specific expression for this sustained upwind process is:

[0058] ;

[0059] in, This indicates a continuous upwind process. The trend of continuous fluctuation units is upward.

[0060] Due to wind speed fluctuations, the continuous fluctuation units continue to decrease, which is a continuous descent wind process. Within a certain period, there is a possibility that the wind speed can decrease from the alarm point to normal. The specific expression for the continuous descent wind process is:

[0061] ;

[0062] in, This indicates a continuous descent of the wind. The trend of continuous fluctuation units is downward.

[0063] When wind speed fluctuates relatively violently over a period of time, with large fluctuation amplitudes and wide wave widths of the fluctuation units, this is a multi-peak wind process, specifically expressed as:

[0064] ;

[0065] in, Indicates a multi-peak wind process. This indicates the threshold value for the amplitude of the fluctuation unit. This indicates the wavelength threshold.

[0066] The wind fluctuates dramatically within a short period of time, which is a continuous peak wind process, specifically expressed as follows:

[0067] ;

[0068] in, This indicates a continuous peak wind process.

[0069] S204. Based on the division process in step S203, divide the wind speed observation sequence into wind processes and establish a wind process database.

[0070] Furthermore, in step S3, similarity matching includes the following sub-steps:

[0071] S301, Set two wind speed observation subsequences as follows and ,in, This represents the first wind speed observation subsequence. express The first wind speed observation point in the middle, express The p-th wind speed observation point, l represents The total number of wind speed observation points, where Y represents the second wind speed observation subsequence. Let q be the q-th wind speed observation point in Y. Let h represent the h-th wind speed observation point in Y, where h represents the total number of wind speed observation points in Y.

[0072] The distance function representing the distance between wind speed observation points in a wind speed observation subsequence. ,in This represents the slope of the error between adjacent points of the p-th wind speed observation point in X. The slope of the error between adjacent points of the q-th wind speed observation point in Y is represented by the following formula:

[0073] ;

[0074] ;

[0075] in, This represents the (p-1)th wind speed observation point in X. This represents the (p+1)th wind speed observation point in X. This represents the (q-1)th wind speed observation point in Y. Let Y represent the (q+1)th wind speed observation point.

[0076] The calculation of adjacent point error does not include the first and last data in the subsequence. Using the adjacent error slope instead of the direct distance between points is more representative and better reflects the trend of the sub-time series segment.

[0077] S302, The formula for calculating the penalty factor is:

[0078] ;

[0079] in, Indicates the penalty factor. This indicates the calculation of standard deviation. This indicates the calculation of JS divergence.

[0080] S303. Combining the penalty factor, the similarity calculation formula is optimized to obtain the SO-DTW algorithm, the specific formula of which is:

[0081] ;

[0082] in, This indicates the similarity between X and Y.

[0083] S304. Using the SO-DTW algorithm, the data to be predicted is matched in the wind process database to obtain wind speed observation sequences with a similarity greater than 90%.

[0084] Furthermore, this invention also proposes a second-level extreme wind speed prediction system along high-speed railway lines based on wind process division, including:

[0085] The final salient feature point set acquisition module is used to extract the initial salient feature points of the second-level salient feature point extraction algorithm of high-speed rail wind speed, remove the transitional peaks, and obtain the final salient feature point set.

[0086] The wind process database establishment module is used to obtain the final significant feature points in the module based on the final significant feature point set. Based on the fluctuation characteristics, amplitude changes, and trend transfer between different fluctuation segments of the fluctuation unit, wind fluctuation processes with similar wind speed change trends and similar fluctuation profiles are classified into one category, including continuous upwind process, continuous downwind process, multi-peak wind process, continuous peak wind process, and low wind speed wind process, and a wind process database is established.

[0087] The similarity matching module is used to design a similarity optimization dynamic time adjustment algorithm, which is used to match wind speed observation sequences with a similarity of more than 90% in the wind process database.

[0088] The wind speed prediction module uses the wind speed observation sequences obtained from the similarity matching module as a training set and inputs them into the TCN-BiGRU prediction model. The prediction model parameters are optimized through the Bayesian optimization algorithm to achieve wind speed prediction.

[0089] Furthermore, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the second-level extreme wind speed prediction method along high-speed railway lines based on wind process division described above.

[0090] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the second-level extreme wind speed prediction method along high-speed railway lines based on wind process division described above.

[0091] The present invention adopts the above technical solution, and its significant technical effects compared with the prior art are as follows:

[0092] 1. This invention proposes a second-level granularity significant feature point extraction algorithm for high-speed rail wind speed. Based on the changes in cross intervals and entropy, it removes transitional peaks and extracts significant extreme points.

[0093] 2. This invention classifies wind fluctuation units based on their fluctuation characteristics, amplitude changes, and trend transitions between different fluctuation segments. This classification system not only adapts to the short-period characteristics and strong fluctuations of wind signals, but also effectively clusters data segments with high similarity in fluctuation patterns, greatly promoting accurate matching of wind processes and optimization of training set selection, thus providing a solid foundation for accurate wind speed prediction.

[0094] 3. This invention identifies the fluctuation patterns in the wind speed data to be predicted, thereby dynamically selecting the optimal compatible training data based on the current prediction input. This allows the model to adapt to new data patterns and sudden events. It effectively handles complex and nonlinear patterns in second-level wind signals along high-speed railway lines, improving the accuracy and adaptability of predictions. Attached Figure Description

[0095] Figure 1 This is a flowchart illustrating the overall implementation of the present invention. Detailed Implementation

[0096] To illustrate the technical content, structural features, objectives, and effects of this invention in detail, the following description, in conjunction with accompanying drawings and examples, will provide a detailed explanation of the invention.

[0097] To achieve the above objectives, this invention proposes a second-level extreme wind speed prediction method along high-speed railway lines based on wind process division, such as... Figure 1 As shown, it includes:

[0098] S1. In this embodiment, wind speed monitoring data measured by an anemometer at a height of 4m above the ground at a train wind speed monitoring station along a high-speed railway in Jiangsu Province from January 1, 2018 to December 31, 2018 is selected. The time resolution is on the second level, and it includes strong wind monitoring data. Using the high-speed railway wind speed SFPE (Second-level significant feature point extraction algorithm), the initial significant feature points of the second-level wind speed observation sequence along the high-speed railway are extracted, and transitional peaks are removed to obtain the final set of significant feature points. Specifically:

[0099] S101, Order This represents a wind speed observation sequence. ,in, This represents the i-th wind speed observation point. This represents the nth wind speed observation point, where n represents the total number of wind speed observation points.

[0100] Define the sliding window as starting from the first... A subsequence of length w starting from a given wind speed observation point, specifically expressed as:

[0101] ;

[0102] in, Indicates a sliding window. satisfy .

[0103] In this embodiment, the window is set to slide in increments of 10.

[0104] Based on the sliding window, define the set of local salient feature points, specifically expressed as:

[0105] ;

[0106] ;

[0107] in, This represents the k-th wind speed observation point. This represents the (k-1)th wind speed observation point. This represents the (k+1)th wind speed observation point. M represents the set of locally significant feature points; j This represents the j-th point in the set of local salient feature points.

[0108] S102, Calculation The independent index of each local salient feature point is used. The larger the independent index, the more important the point is in the overall signal. The specific formula is as follows:

[0109] ;

[0110] in, Independent index representing local salient feature points, This represents the absolute value of the distance between adjacent data. express The adjacent local salient feature point on the left, express The adjacent local salient feature point on the right.

[0111] S103, Calculation The weighted entropy of each local salient feature point in the wind speed observation sequence is calculated using the following formula:

[0112] ;

[0113] ;

[0114] in, This represents the sequence consisting of s wind speed observation points to the right of the wind speed observation sequence, where the local salient feature point is located. express Entropy; express At the r-th wind speed observation point, 0 <r<s; express Probability density estimation; This represents the difference between the s wind speed observation points to the right of the local salient feature point in the wind speed observation sequence. Diff represents difference calculation; This represents a sequence of s wind speed observation points to the left of the wind speed sequence, indicating a locally significant feature point. express Entropy; express The Middle Wind speed observation data, ; express Probability density estimation; This represents the difference between the s wind speed observation points to the left of the local salient feature point in the wind speed sequence. .

[0115] S104, Calculation The importance score for each salient feature point is calculated using the following formula:

[0116] ;

[0117] in, This indicates the importance score. When The larger the value, the more important the point is.

[0118] Will Locally significant feature points with a value greater than 1 are defined as initial maximal significant feature points.

[0119] S105. Redefine the set of local salient feature points, with the specific expression as follows:

[0120] ;

[0121] ;

[0122] in, This represents the redefined set of local salient feature points; This represents the j-th point in the redefined set of local salient feature points.

[0123] Repeat steps S102-S104 to... Locally significant feature points with a value less than or equal to 1 are defined as initial minimum significant feature points.

[0124] S106. Label the attributes of the initial maximal and minimum salient feature points. The attribute of the initial maximal salient feature points is 1, and the attribute of the initial minimum salient feature points is -1. When two consecutive points have an attribute of 1, remove the minimum value among the corresponding values ​​of these two points, and the remaining points are the final maximal salient feature points. When two consecutive points have an attribute of -1, remove the maximum value among the corresponding values ​​of these two points, and the remaining points are the final minimum salient feature points. This yields the final set of maximal and minimum salient feature points. This represents the final set of maximal salient feature points. ; This represents the final set of minimal salient feature points. ;in, Let a represent the a-th final maximal salient feature point. Let a represent the a-th final minimal salient feature point. Indicates the first The final maximal salient feature point, Indicates the first The final minimal salient feature point.

[0125] Therefore, the final set of salient feature points is obtained. , .

[0126] S2. Based on the final set of salient feature points in step S1, and considering the fluctuation characteristics, amplitude changes, and trend transitions between different fluctuation segments of the fluctuation units, wind fluctuation processes with similar wind speed change trends and similar fluctuation profiles are classified into one category. This category includes continuous upwind processes, continuous downwind processes, multi-peak wind processes, continuous peak wind processes, and low-speed wind processes. A wind process database is established, the specific content of which is as follows:

[0127] S201. Parameter definition: The attribute of the initial maximum salient feature point is 1, the attribute of the initial minimum salient feature point is -1, and when the attributes of two consecutive points in the final salient feature point are "1, -1" or "-1, 1", the point at this time is a fluctuation point.

[0128] The process from a minimum fluctuation point to a maximum fluctuation point, and then to the next minimum fluctuation point, is called a fluctuation unit.

[0129] The time values ​​corresponding to the final minimum significant feature points on both sides of the final maximum significant feature point are the wave widths of the wave element, specifically expressed as:

[0130] ;

[0131] in, Indicates the wave width of the wave unit. This represents the time value corresponding to the final minimum significant feature point to the right of the final maximum significant feature point. This represents the time value corresponding to the final minimum significant feature point to the left of the final maximum significant feature point.

[0132] The absolute difference between adjacent final maxima and final minima represents the amplitude of the fluctuation unit, specifically expressed as:

[0133] ;

[0134] in, Indicates the amplitude of change of the fluctuation unit. Indicates the first The final maximal salient feature point, Indicates the first The final minimal salient feature point, ; , In the final set of salient feature points, they are adjacent salient feature points.

[0135] S202. Determine the degree of trend change of the fluctuation unit based on the slope of adjacent final maxima salient feature points to guide the division of wind processes. The specific formula is as follows:

[0136] ;

[0137] in, This value represents the degree of trend change within the fluctuation unit. This represents the summation function. Let m represent the m-th final maximal salient feature point. Let m represent the (m-1)th final maximal salient feature point. express At the corresponding time points in the wind speed observation sequence express The corresponding time point in the wind speed observation sequence.

[0138] according to The criteria for trend judgment of fluctuation units are as follows:

[0139] ;

[0140] in, Indicates a trend. Indicates an upward trend. This indicates a downward trend.

[0141] S203. If the continuous fluctuations of the unit remain within the discrimination threshold without any significant increase, this indicates a low-wind-speed process. The wind speed will not rise to the alarm point in a short period. The specific expression for a low-wind-speed process is:

[0142] ;

[0143] in, This indicates a low-speed wind process. This represents all the final salient features during a low-speed wind process. Indicates the discrimination threshold. This represents the final maximal salient feature points closely adjacent to the two ends of a low-wind-speed process. This represents the maximum value of the entire wind speed observation sequence. These represent the start and end points of a low-wind-speed process, respectively.

[0144] Due to wind speed fluctuations, the continuous rising units constitute a sustained upwind process. Within a certain timeframe, there is a possibility that the wind will rise to the alarm point. The specific expression for this sustained upwind process is:

[0145] ;

[0146] in, This indicates a continuous upwind process. The trend of continuous fluctuation units is upward.

[0147] Due to wind speed fluctuations, the continuous fluctuation units continue to decrease, which is a continuous descent wind process. Within a certain period, there is a possibility that the wind speed can decrease from the alarm point to normal. The specific expression for the continuous descent wind process is:

[0148] ;

[0149] in, This indicates a continuous descent of the wind. The trend of continuous fluctuation units is downward.

[0150] When wind speed fluctuates relatively violently over a period of time, with large fluctuation amplitudes and wide wave widths of the fluctuation units, this is a multi-peak wind process, specifically expressed as:

[0151] ;

[0152] in, Indicates a multi-peak wind process; This indicates the threshold value for the amplitude of the fluctuation unit. Indicates the wavelength threshold. .

[0153] The wind fluctuates dramatically within a short period of time, which is a continuous peak wind process, specifically expressed as follows:

[0154] ;

[0155] in, This indicates a continuous peak wind process.

[0156] S204. Based on the division process in step S203, divide the wind speed observation sequence into wind processes and establish a wind process database.

[0157] S3. Design the SO-DTW (Similarity Optimization-Dynamic Time Wrapping) algorithm. Use this algorithm to match wind speed observation sequences with a similarity of over 90% in the wind process database. Specifically:

[0158] S301, Set two wind speed observation subsequences as follows and ,in, This represents the first wind speed observation subsequence. express The first wind speed observation point in the middle, express The p-th wind speed observation point, l represents The total number of wind speed observation points, where Y represents the second wind speed observation subsequence. Let q be the q-th wind speed observation point in Y. Let h represent the h-th wind speed observation point in Y, where h represents the total number of wind speed observation points in Y.

[0159] The distance function representing the distance between wind speed observation points in a wind speed observation subsequence. ,in This represents the slope of the error between adjacent points of the p-th wind speed observation point in X. The slope of the error between adjacent points of the q-th wind speed observation point in Y is represented by the following formula:

[0160] ;

[0161] ;

[0162] in, This represents the (p-1)th wind speed observation point in X. This represents the (p+1)th wind speed observation point in X. This represents the (q-1)th wind speed observation point in Y. Let Y represent the (q+1)th wind speed observation point.

[0163] S302, The formula for calculating the penalty factor is:

[0164] ;

[0165] in, Indicates the penalty factor. This indicates the calculation of standard deviation. This indicates the calculation of JS divergence.

[0166] S303. Combining the penalty factor, the similarity calculation formula is optimized to obtain the SO-DTW algorithm, the specific formula of which is:

[0167] ;

[0168] in, This indicates the similarity between X and Y.

[0169] S304. Using the SO-DTW algorithm, the data to be predicted is matched in the wind process database to obtain wind speed observation sequences with a similarity greater than 90%.

[0170] S4. The wind speed observation sequence obtained in step S3 is used as a training set and input into the TCN-BiGRU prediction model. The prediction model parameters are optimized by Bayesian optimization algorithm to achieve wind speed prediction.

[0171] This invention also proposes a second-level extreme wind speed prediction system along high-speed railways based on wind process segmentation. The system includes a module for obtaining the final salient feature point set, a wind process database establishment module, a similarity matching module, a wind speed prediction module, and a computer program that can run on a processor. It should be noted that each module in the above system corresponds to a specific step of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0172] This invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0173] This invention also proposes a computer-readable storage medium storing a computer program. It should be noted that when the computer program is executed by a processor, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0174] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for predicting second-level extreme wind speeds along high-speed railway lines based on wind process division, characterized in that: include: S1. Using the high-speed rail wind speed second-level salient feature point extraction algorithm, the initial salient feature points of the second-level wind speed observation sequence along the high-speed rail line are extracted and the transitional peaks are removed to obtain the final set of salient feature points. Specifically: S101, Order This represents a wind speed observation sequence. ,in, This represents the i-th wind speed observation point. This represents the nth wind speed observation point, where n represents the total number of wind speed observation points. Define the sliding window as starting from the first... A subsequence of length w starting from a given wind speed observation point, specifically expressed as: ； in, Indicates a sliding window. satisfy ; Based on the sliding window, define the set of local salient feature points, specifically expressed as: ； ； in, This represents the k-th wind speed observation point. This represents the (k-1)th wind speed observation point. This represents the (k+1)th wind speed observation point. M represents the set of locally significant feature points; j This represents the j-th point in the set of local salient feature points; S102, Calculation The independent index for each local salient feature point is given by the following formula: ； in, Independent index representing local salient feature points, This represents the absolute value of the distance between adjacent data. express The adjacent local salient feature point on the left, express The adjacent local salient feature point on the right; S103, Calculation The weighted entropy of each local salient feature point in the wind speed observation sequence is calculated using the following formula: ； ； in, This represents the sequence consisting of s wind speed observation points to the right of the wind speed observation sequence, where the local salient feature point is located. express Entropy; express At the r-th wind speed observation point, 0 <r<s; express Probability density estimation; This represents the difference between the s wind speed observation points to the right of the local salient feature point in the wind speed observation sequence. Diff represents difference calculation; This represents a sequence of s wind speed observation points to the left of the wind speed sequence, indicating a locally significant feature point. express Entropy; express The Middle Wind speed observation data, ; express Probability density estimation; This represents the difference between the s wind speed observation points to the left of the local salient feature point in the wind speed sequence. ; S104, Calculation The importance score for each salient feature point is calculated using the following formula: ； in, Indicates importance score; Will Locally significant feature points with a value greater than 1 are defined as initial maximum significant feature points; S105. Redefine the set of local salient feature points, with the specific expression as follows: ； ； in, This represents the redefined set of local salient feature points; This represents the j-th point in the redefined set of local salient feature points; Repeat steps S102-S104 to... Locally significant feature points less than or equal to 1 are defined as initial minimum significant feature points; S106. Label the attributes of the initial maximal and minimum salient feature points. The attribute of the initial maximal salient feature points is 1, and the attribute of the initial minimum salient feature points is -1. When two consecutive points have an attribute of 1, remove the minimum value among the corresponding values ​​of these two points, and the remaining points are the final maximal salient feature points. When two consecutive points have an attribute of -1, remove the maximum value among the corresponding values ​​of these two points, and the remaining points are the final minimum salient feature points. This yields the final set of maximal and minimum salient feature points. This represents the final set of maximal salient feature points. ; This represents the final set of minimal salient feature points. ;in, Let a represent the a-th final maximal salient feature point. Let a represent the a-th final minimal salient feature point. Indicates the first The final maximal salient feature point, Indicates the first The final minimal salient feature point; Therefore, the final set of salient feature points is obtained. , ; S2. Based on the final set of significant feature points in step S1, classify the wind processes, including continuous upwind processes, continuous downwind processes, multi-peak wind processes, continuous peak wind processes, and low-speed wind processes, and establish a wind process database. S3. Design a similarity optimization dynamic time adjustment algorithm and use the algorithm to match wind speed observation sequences in the wind process database; S4. The wind speed observation sequence obtained in step S3 is used as a training set and input into the TCN-BiGRU prediction model. The prediction model parameters are optimized by Bayesian optimization algorithm to achieve wind speed prediction.

2. The method for predicting second-level extreme wind speeds along high-speed railway lines based on wind process division as described in claim 1, characterized in that, Step S2, establishing the wind process database, includes the following sub-steps: S201. Parameter definition: The attribute of the initial maximum salient feature point is 1, the attribute of the initial minimum salient feature point is -1, and when the attributes of two consecutive points in the final salient feature point are "1, -1" or "-1, 1", the point is a fluctuation point. The process from a minimum fluctuation point to a maximum fluctuation point, and then to the next minimum fluctuation point, is called a fluctuation unit; The time values ​​corresponding to the final minimum significant feature points on both sides of the final maximum significant feature point are the wave widths of the wave element, specifically expressed as: ； in, Indicates the wave width of the wave unit. This represents the time value corresponding to the final minimum significant feature point to the right of the final maximum significant feature point. This represents the time value corresponding to the final minimum significant feature point to the left of the final maximum significant feature point; The absolute difference between adjacent final maxima and final minima represents the amplitude of the fluctuation unit, specifically expressed as: ； in, Indicates the amplitude of change of the fluctuation unit. Indicates the first The final maximal salient feature point, Indicates the first The final minimal salient feature point, ; , In the final set of salient feature points, they are adjacent salient feature points; S202. Determine the degree of trend change of the fluctuation unit based on the slope of adjacent final maxima salient feature points to guide the division of wind processes. The specific formula is as follows: ； in, This value represents the degree of trend change within the fluctuation unit. This represents the summation function. Let m represent the m-th final maximal salient feature point. Let m represent the (m-1)th final maximal salient feature point. express At the corresponding time points in the wind speed observation sequence express The corresponding time point in the wind speed observation sequence; according to The criteria for trend judgment of fluctuation units are as follows: ； in, Indicates a trend. Indicates an upward trend. Indicates a downward trend; S203, The specific expression for the low-wind-speed process is: ； in, This indicates a low-speed wind process. This represents all the final salient features during a low-speed wind process. Indicates the discrimination threshold. This represents the final maximal salient feature points closely adjacent to the two ends of a low-wind-speed process. This represents the maximum value of the entire wind speed observation sequence. These represent the start and end points of a low-wind-speed process, respectively. This represents the final set of maximal salient feature points. This represents the final set of minimal salient feature points; The specific expression for a sustained upwind process is: ； in, This indicates a continuous upwind process. The trend of continuous fluctuation units is upward; The specific expression for the continuous descent process is: ； in, This indicates a continuous descent of the wind. The trend of continuous fluctuation units is downward; The specific expression for a multi-peak wind process is: ； in, Indicates a multi-peak wind process. This indicates the threshold value for the amplitude of the fluctuation unit. Indicates the wavelength threshold; The specific expression for the continuous peak wind process is: ； in, Indicates a continuous peak wind process; S204. Based on the division process in step S203, divide the wind speed observation sequence into wind processes and establish a wind process database.

3. The method for predicting second-level extreme wind speeds along high-speed railway lines based on wind process division as described in claim 1, characterized in that, Step S3, similarity matching, includes the following sub-steps: S301, Set two wind speed observation subsequences as follows and ,in, This represents the first wind speed observation subsequence. express The first wind speed observation point in the middle, express The p-th wind speed observation point, l represents The total number of wind speed observation points, where Y represents the second wind speed observation subsequence. Let q be the q-th wind speed observation point in Y. Let h represent the h-th wind speed observation point in Y, where h represents the total number of wind speed observation points in Y; The distance function representing the distance between wind speed observation points in a wind speed observation subsequence. ,in This represents the slope of the error between adjacent points of the p-th wind speed observation point in X. The slope of the error between adjacent points of the q-th wind speed observation point in Y is represented by the following formula: ； ； in, This represents the (p-1)th wind speed observation point in X. This represents the (p+1)th wind speed observation point in X. This represents the (q-1)th wind speed observation point in Y. This represents the (q+1)th wind speed observation point in Y; S302, The formula for calculating the penalty factor is: ； in, Indicates the penalty factor. This indicates the calculation of standard deviation. Indicates the calculation of JS divergence; S303. Combining the penalty factor, optimize the similarity calculation formula to obtain the similarity optimization dynamic time adjustment algorithm. The specific formula is as follows: ； in, Indicates the similarity between X and Y; S304. Utilize the similarity optimization dynamic time adjustment algorithm to match the data to be predicted in the wind process database to obtain wind speed observation sequences with a similarity greater than 90%.

4. A system applied to the second-level extreme wind speed prediction method along high-speed railways based on wind process division as described in claim 1, characterized in that, include: The final salient feature point set acquisition module is used to extract the initial salient feature points of the second-level salient feature point extraction algorithm of high-speed rail wind speed, remove the transitional peaks, and obtain the final salient feature point set. The wind process database establishment module is used to classify wind processes based on the final significant feature points obtained from the module, including continuous upwind processes, continuous downwind processes, multi-peak wind processes, continuous peak wind processes, and low wind speed wind processes, and to establish a wind process database. The similarity matching module is used to design a similarity optimization dynamic time adjustment algorithm, which is used to match wind speed observation sequences with a similarity of more than 90% in the wind process database. The wind speed prediction module uses the wind speed observation sequences obtained from the similarity matching module as a training set and inputs them into the TCN-BiGRU prediction model. The prediction model parameters are optimized through the Bayesian optimization algorithm to achieve wind speed prediction.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to perform the method of any one of claims 1 to 3.

Images