Method and system for determining a representative value of temperature

Abstract

The application discloses a temperature representative value determination method and system, and relates to the field of statistics, which comprises the following steps: using a cubic interpolation method to fit and replace abnormal temperature data in a year, and using Gaussian filtering to perform smoothing processing; calculating a temperature difference and using an adaptive dynamic time warping (Adaptive DTW) and a mean Euclidean distance to obtain a difference distance matrix of temperature change characteristics of different months; using a weight parameter to combine the Adaptive DTW and the Euclidean distance, calculating a comprehensive distance matrix, performing a condensed hierarchical cluster analysis on the temperature change characteristics, and obtaining a plurality of temperature change modes; determining a typical summer temperature month and a typical winter temperature month according to the plurality of temperature change modes, and obtaining a daily extreme temperature difference therefrom; using a generalized extreme value distribution model to obtain a probability distribution of the daily extreme temperature difference in different return periods; and obtaining temperature action representative values in different return periods according to the probability distribution result. The application can accurately obtain temperature action representative values.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of statistics, in particular to a method and system for determining temperature representative value. BACKGROUND

[0002] In the time series analysis method of temperature data, most of the division of typical seasons still adopts a relatively subjective seasonal month division. However, the temperature characteristics and representative seasons of different regions differ significantly, and the existing season division method often lacks effectiveness and rationality. In addition, in the aspect of time series temperature data anomaly detection, traditional methods still mainly use extreme value elimination and box plot probability elimination mode. For large-scale time series temperature data with implicit trends, these methods are difficult to efficiently extract potential features in the data. In addition, for the research of temperature action representative value of steel-concrete composite structure, most of them are still based on the fitting of return period extreme value distribution model. There is a lack of reasonable division of typical temperature seasons and reliable data anomaly value detection. Research on temperature action representative value helps to accurately predict and evaluate the stress of steel-concrete composite structure under different temperature conditions, so as to ensure the safety and durability of the structure.

[0003] In summary, how to accurately obtain the temperature action representative value is an important problem to be solved. SUMMARY

[0004] The embodiment of the present application provides a method and system for determining temperature representative value, which can solve the problem of inaccurate temperature action representative value obtained in the prior art.

[0005] The embodiment of the present application provides a method for determining temperature representative value, comprising the following steps:

[0006] Obtain the abnormal temperature data of the steel-concrete composite structure within one year, use the adaptive dynamic time warping algorithm (Adaptive DTW) and the mean Euclidean distance to obtain the difference distance matrix of the temperature change characteristics of different months, and use the difference distance matrix of the Adaptive DTW and the difference distance matrix of the mean Euclidean distance to obtain the comprehensive distance matrix;

[0007] Perform agglomerative hierarchical clustering analysis on the comprehensive distance matrix, divide the months with similar temperature change characteristics within one year into the same class, obtain the seasonal pattern of temperature change, determine the typical summer temperature month and the typical winter temperature month, and obtain the comprehensive distance matrix;

[0008] Select a temperature measurement point on the steel-concrete composite structure, obtain the daily extreme temperature difference of the measurement point in the typical summer temperature month and the typical winter temperature month, and use the generalized extreme value distribution model to obtain the probability distribution of the daily extreme temperature difference; according to the probability distribution, use the exceedance probability to obtain the temperature action representative value of different return periods.

[0009] Furthermore, the method of obtaining abnormal temperature data of the steel-concrete composite structure within one year specifically includes the following steps: using a gradient detection method to identify data with sharp changes in the annual temperature data time series of the steel-concrete composite structure; wherein the annual temperature data time series is used to represent temperature change data at different time points within one year; using a box plot detection method to filter out temperature values ​​that abnormally deviate from the normal range from the annual temperature data time series; using an isolation forest algorithm to identify outliers from the annual temperature data time series; and obtaining abnormal temperature data within one year based on the identification and screening results.

[0010] Furthermore, the agglomerative hierarchical clustering analysis is performed on the comprehensive distance matrix to classify months with similar temperature change characteristics within a year into the same category to obtain the seasonal pattern of temperature change. The specific steps include:

[0011] Adaptive interpolation algorithm is used to fit and replace abnormal temperature data within one year, and Gaussian filtering is used to smooth the replacement results.

[0012] The smoothing results are classified by month, and the 12 months are paired using dynamic time warping (DTW). The similarity between the temperature series of the 12 months is obtained pair by pair, and a symmetrical DTW distance matrix is ​​generated. DTW _ dist :

[0013] ;

[0014] in, n Indicates the maximum month of the year is 12;

[0015] Get the mean of the data for each month and get the Euclidean distance matrix based on the mean ;

[0016] Normalize the DTW distance matrix and the Euclidean distance matrix, and weight the normalized results:

[0017] ;

[0018] in, represents weight;

[0019] Distance Matrix Agglomerative hierarchical clustering method was used to cluster the 12 months into k clusters, each cluster corresponding to a seasonal pattern and the seasonal pattern was taken as a temperature change pattern.

[0020] Furthermore, the method of using a generalized extreme value distribution model to obtain a probability distribution of the daily extreme temperature difference comprises the following steps:

[0021] The daily temperature difference extreme value of the two detection points is used as a random variablex , using the generalized extreme value distribution for probability fitting , the formula is:

[0022] ;

[0023] in, is the position parameter, controlling the center of the data; is the scale parameter, which controls the width or variability of the data; is the shape parameter, which determines the tail behavior of the distribution;

[0024] Cumulative distribution function , the formula is:

[0025] ;

[0026] The distribution parameters are determined using maximum likelihood estimation, and the probability distribution of daily extreme temperature differences is determined based on the distribution parameters.

[0027] Furthermore, the method of using the exceedance probability to obtain representative values ​​of temperature effects at different return periods specifically includes the following steps:

[0028] According to the probability distribution of daily extreme temperature difference in different return periods, based on a 50-year return period, the exceedance probability is used to obtain the corresponding representative value of the temperature effect. The formula is:

[0029] ;

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] in, X is the daily temperature difference extreme value, is the typical number of days in summer with a 50-year return period; is the typical number of days in winter with a 50-year return period; is the random variable of daily temperature difference in summer; is the winter daily temperature difference random variable; is the exceedance probability with a return period of 50 years. During the 100-year design reference period, the expected number of times that the temperature threshold is exceeded for an extreme temperature event with a return period of 50 years is 2. =0.02; is the shape parameter for summer; is the shape parameter for winter; is the scale parameter for summer; is a scale parameter for winter; is a position parameter for summer; is a position parameter for winter;

[0035] is a temperature action representative value.

[0036] The embodiment of the present application provides a temperature representative value determination system, comprising:

[0037] The data acquisition module is used for acquiring abnormal temperature data of the steel-concrete composite structure in one year, and is used for acquiring difference distance matrices of temperature variation characteristics of different months by using an adaptive dynamic time warping algorithm (Adaptive DTW) and a mean Euclidean distance respectively, and is used for obtaining a comprehensive distance matrix by performing weighted processing on the difference distance matrix of the Adaptive DTW and the difference distance matrix of the mean Euclidean distance.

[0038] The typical seasonal temperature month acquisition module is used for performing a condensed hierarchical clustering analysis on the comprehensive distance matrix, and is used for dividing months with similar temperature variation characteristics into the same class to obtain a seasonal pattern of temperature variation, and is used for determining a typical summer temperature month and a typical winter temperature month.

[0039] The temperature action representative value acquisition module is used for selecting a temperature measurement point on the steel-concrete composite structure, acquiring daily extreme temperature differences of the measurement point in the typical summer temperature month and the typical winter temperature month, and obtaining a probability distribution of the daily extreme temperature differences by using a generalized extreme value distribution model; and is used for obtaining temperature action representative values of different return periods by using a beyond probability according to the probability distribution.

[0040] The embodiment of the present application provides a temperature representative value determination method and system, and has the following beneficial effects compared with the prior art:

[0041] The Adaptive DTW and the mean Euclidean distance are used to acquire difference distance matrices of temperature variation characteristics of different months respectively, and a condensed hierarchical clustering analysis is performed on the temperature variation characteristics, months with similar temperature variation characteristics are divided into the same class to obtain a seasonal pattern of temperature variation, a typical summer temperature month and a typical winter temperature month are determined, typical temperature seasons are reasonably divided, and finally, a temperature action representative value is obtained according to the reasonably divided result, so that the purpose of accurately obtaining the temperature action representative value is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The embodiment of the present application provides a temperature representative value determination method and system, and has the following beneficial effects compared with the prior art:

[0043] Figure 2 ​A flowchart of an isolation forest algorithm for determining a temperature representative value provided by an embodiment of the present invention;

[0044] Figure 3 A flow chart of an adaptive interpolation algorithm for interpolation fitting and Gaussian filtering processing of a method for determining a temperature representative value provided by an embodiment of the present invention;

[0045] Figure 4 A local Gaussian smoothing algorithm processing flow chart of an interpolation fitting and Gaussian filtering processing flow chart of a method for determining a temperature representative value provided by an embodiment of the present invention;

[0046] Figure 5 A flowchart of a hierarchical clustering algorithm based on Adaptive DTW and Euclidean distance for determining a temperature representative value provided by an embodiment of the present invention;

[0047] Figure 6 A hierarchical clustering dendrogram of a classification demonstration diagram of randomly generated data for a method for determining a temperature representative value provided by an embodiment of the present invention;

[0048] Figure 7 The hierarchical clustering results of adaptive DTW and Euclidean distance of a classification demonstration graph of randomly generated data of a method for determining a temperature representative value provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] See also Figure 1 , an embodiment of the present invention provides a method for determining a temperature representative value, comprising the following steps:

[0051] Step 1: Obtain abnormal temperature data for the steel-concrete composite structure over a year. Use the Adaptive Dynamic Time Warping algorithm (Adaptive DTW) and the mean Euclidean distance to obtain the difference distance matrix of the temperature change characteristics in different months. Weighted processing is performed on the difference distance matrix of Adaptive DTW and the difference distance matrix of mean Euclidean distance to obtain a comprehensive distance matrix.

[0052] Step 2: Perform agglomerative hierarchical cluster analysis on the comprehensive distance matrix, grouping months with similar temperature variation characteristics within a year into the same cluster. This identifies seasonal patterns of temperature variation and identifies typical summer and winter months. Temperature measurement points on the steel-concrete composite structure are selected, and the daily extreme temperature differences at these measurement points are obtained for the typical summer and winter months. A generalized extreme value distribution model is used to determine the probability distribution of these daily extreme temperature differences. Based on this probability distribution, exceedance probabilities are used to obtain representative values ​​of temperature effects at different return periods.

[0053] To address the shortcomings of existing technologies, this paper proposes an improved method that uses Adaptive DTW with mean Euclidean distance and agglomerative hierarchical clustering to classify months with similar temperature variation characteristics, thereby identifying typical summer and winter months in the annual temperature time series. This method effectively improves data analysis efficiency, is well-suited for analyzing data series of varying lengths, and minimizes the impact of regional temperature variations on analysis results.

[0054] For temperature data outlier detection, this paper combines multiple methods, including gradient detection, boxplot detection, and the isolation forest algorithm, to comprehensively detect temperature anomalies. By analyzing the time-course temperature trend, an adaptive interpolation algorithm is used to fit and replace outliers. Finally, a local Gaussian smoothing algorithm is used to eliminate the impact of short-term fluctuations, thereby improving data reliability and accuracy.

[0055] Furthermore, this paper analyzes the probability distribution of measured daily extreme temperatures based on the generalized extreme value distribution model (GEV), thereby proposing a method for calculating a representative value of temperature effects based on the return period. This method more accurately reflects the distribution characteristics of temperature extremes and provides a reliable basis for calculating representative values ​​of temperature effects.

[0056] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention includes the following steps:

[0057] S1: Batch read temperature data.

[0058] In this step, the MATLAB program is used to read the temperature data time history from different sensors or temperature measurement points. Through standardization, the data format is ensured to be consistent and ready for subsequent processing.

[0059] S2: Abnormal temperature detection and evaluation based on gradient detection and isolation forest algorithm.

[0060] This step detects outliers in the temperature data through gradient detection, box plot detection, and isolation forest algorithm. The gradient detection method is used to identify the parts of the temperature data that change sharply, the box plot detection is used to filter out temperature values ​​that deviate abnormally from the normal range, and the isolation forest algorithm is used to efficiently detect outliers in high-dimensional data. By combining multiple detection methods, the accuracy and robustness of outlier detection are improved. After detection, the proportion of each type of abnormal data is evaluated, and the data stability and reliability of the sensor measurement points are analyzed. Figure 2 shown.

[0061] S3: Abnormal temperature data interpolation fitting and Gaussian filtering processing.

[0062] The detected abnormal data is fitted and replaced by an adaptive interpolation algorithm. The adaptive interpolation algorithm uses a regional marking method to identify long and short NaN segments (data anomalies) and treat them differently. Small missing segments (less than 10 data points) are supplemented by cubic interpolation, while long missing segments are deleted to avoid the accumulation of interpolation errors and ensure interpolation accuracy. Cubic interpolation can infer reasonable temperature values ​​based on the local trend of the data, avoiding the interference of data fluctuations on the overall analysis results. Subsequently, Gaussian filtering is used to smooth the data. The local Gaussian smoothing algorithm only smooths short-term fluctuation data to avoid information loss caused by over-smoothing, ensuring that the long-term trend information of the time series is not affected, thereby improving the stability and reliability of the data. This step helps to ensure the accuracy of the final temperature data so that it conforms to the actual temperature change law. Figure 3 The figure shows the flow chart of the adaptive interpolation algorithm. Figure 4 This is the flowchart of the local Gaussian smoothing algorithm.

[0063] S4: Typical temperature seasons are divided based on Adaptive DTW+mean Euclidean distance+hierarchical clustering algorithm.

[0064] Adaptive DTW+mean Euclidean distance+condensed hierarchical clustering algorithm is used for clustering analysis of annual temperature data. Adaptive dynamic time warping (Adaptive DTW) and mean Euclidean distance are used to obtain the difference distance matrix of temperature variation characteristics of different months. A weight parameter is used to combine Adaptive DTW and Euclidean distance to calculate the comprehensive distance matrix. According to the comprehensive distance matrix, condensation hierarchical clustering analysis is performed, and months with similar temperature variation characteristics are divided into the same class. The hierarchical clustering algorithm continuously merges similar data points or divides different classes to obtain multiple temperature variation patterns. Through analysis of each cluster, typical summer temperature months and typical winter temperature months are determined, avoiding the subjectivity of traditional methods and achieving accurate clustering of unequal length time series. Adaptive DTW calculates the nonlinear alignment distance between time series, effectively solving the time axis misalignment problem, and significantly reducing the computational complexity of traditional DTW. Euclidean distance provides computational efficiency. After normalization, distance weighting fusion is performed based on the adjustable weight parameter to obtain an optimized clustering effect, which also reduces the impact of regional temperature changes. Figure 5

[0065] S5: Probability distribution analysis of daily extreme temperature difference.

[0066] By statistically analyzing the daily extreme temperature difference data, the probability distribution of the extreme value is fitted, and the distribution type of the extreme value is determined. The generalized extreme value distribution (GEV) model is used to fit the temperature extreme value data, the variation law of the daily temperature difference under different return periods is studied, and the occurrence probability of the extreme temperature difference is calculated. This analysis helps to predict the frequency of different extreme temperature events and provides data support for the calculation of subsequent temperature action representative values.

[0067] S6: Calculation of temperature action representative value based on return period and exceedance probability.

[0068] Based on the probability distribution results in S5, the temperature action representative values under different return periods are calculated based on the generalized extreme value distribution model. These representative values reflect the possible extreme temperature events that may occur under different return periods, providing a basis for the evaluation of the impact of temperature on structures. By calculating the exceedance probability, the occurrence risk of extreme temperature events can be further quantified, and scientific basis can be provided for the prediction of the impact of temperature changes on structural health.

[0069] When long-term temperature measurement of structures is performed, the detection equipment may be disturbed by factors such as construction and environmental changes, resulting in abnormal fluctuations in temperature data. In order to accurately identify these abnormal data, the gradient detection and isolated forest algorithm are used in S2 step, which can effectively detect and eliminate abnormal data.

[0070] ​Gradient detection identifies unusual fluctuations by calculating the rate of change (i.e., local slope) between adjacent data points. The basic idea behind this method is that large fluctuations between two adjacent data points may be due to data anomalies or sudden environmental changes. By calculating the differences between the data points, unusual fluctuations can be captured, marked, and processed.

[0071] Specifically, the steps of gradient detection are as follows:

[0072] Calculate the change between each pair of adjacent data points, that is, the difference value.

[0073] If the difference value exceeds the preset threshold, it is considered that the data corresponding to the point has undergone abnormal changes.

[0074] The data values ​​of these abnormal points are replaced with NaN for subsequent processing.

[0075] In gradient detection, the local slope (i.e., the rate of change of the data) can be expressed by calculating the difference between adjacent data points:

[0076] (1).

[0077] in, and is a data point in the time series, is the time step, that is, the time difference between adjacent data points.

[0078] Then, for each local slope, determine whether it exceeds the set slope threshold.

[0079] (2).

[0080] The above scheme has the beneficial effect of quickly and effectively identifying drastic changes in temperature data, thereby promptly marking and removing these abnormal data and ensuring data accuracy. The gradient detection method can quickly identify abnormal data with excessively drastic temperature changes and is simple to operate, making it suitable for preliminary anomaly detection in real-time monitoring.

[0081] Isolation Forest is an unsupervised machine learning algorithm widely used for outlier detection in high-dimensional data. This algorithm relies on the "isolation" of data points, and outliers are generally more easily isolated than normal data points. By constructing multiple random trees, the isolation forest splits the data into two subsets by randomly selecting features and split values. Normal data points are generally difficult to isolate, while outliers, due to their significantly different properties, are easily isolated during the tree construction process.

[0082] In the implementation of the present invention, the isolation forest scores the “degree of isolation” of each data point through multiple isolation trees, and points with higher scores are considered to be outliers.

[0083] The beneficial effects of the above scheme are: the algorithm is efficient and robust, and is particularly suitable for anomaly detection in large-scale data.

[0084] Data Preset: First, extract the temperature data from the data table, and then set the parameters of the Isolation Forest, including the number of trees and the subsample size of each tree.

[0085] Constructing Isolation Trees: For each tree, the algorithm selects a feature to split by randomly sampling subsamples from the dataset until the tree reaches a preset maximum depth or the number of data points to split is less than the specified sample size.

[0086] Calculate the path length of each data point: Calculate the path length for each data point through each tree. The path length represents the number of steps required to reach a leaf node from the root node. For outliers, since their properties are significantly different from the majority of the data, they are usually easier to isolate, so their path length is shorter.

[0087] (3).

[0088] in: is a data point In the tree The path length in ; is the number of subsamples of the tree.

[0089] Calculate anomaly score: The anomaly score of each data point is obtained by combining the path lengths of its path in all isolated trees. Data points with lower scores are considered anomalies, while data points with higher scores are considered normal points.

[0090] (4).

[0091] in, is a constant that depends on the sample size:

[0092] (5).

[0093] in, is the number of subsamples in the dataset; is Euler's constant.

[0094] Identify outliers: By setting a threshold (e.g., 95th percentile), data points with high anomaly scores are marked as outliers.

[0095] if s (x )>threshold (6), the point is marked as an outlier.

[0096] Replace outliers: Replace detected outliers with NaN (missing values) for subsequent processing. Finally, update the data table and return the processed temperature data.

[0097] In data analysis, direct removal of outliers after detection can negatively impact time series data with inherent trends. This is particularly true when outliers represent short-term fluctuations or measurement errors, potentially destroying the underlying information in the data. In step S3, an adaptive interpolation algorithm infers reasonable temperature values ​​based on the data's local trends, preventing data fluctuations from interfering with the overall analysis results. Using an adaptive interpolation algorithm effectively restores data trends without disrupting the overall data pattern. This interpolation method infers reasonable values ​​based on local trends, thus avoiding the information loss associated with direct removal of outliers.

[0098] The beneficial effects of this approach are as follows: the adaptive interpolation algorithm distinguishes between short-term and long-term missing data through region labeling and adopts different processing strategies. For short-term missing data (fewer than 10 consecutive missing points), Hermite interpolation is used to maintain data continuity; for long-term missing data (10 or more consecutive missing points), NaNs are deleted to prevent the accumulation of interpolation errors and affect data authenticity. In addition, by deduplicating and sorting time data, the stability of the interpolation calculation is ensured and the adaptability of the interpolation method is improved. Hermite interpolation is an interpolation method based on given data points and local derivative information. Unlike traditional linear interpolation or Lagrange interpolation methods, Hermite interpolation not only considers the location of data points but also utilizes their derivative information, resulting in a smoother interpolation curve. It is particularly suitable for processing time series data containing noise or outliers.

[0099] (7).

[0100] in: is the cubic Hermite interpolation function. is the normalized parameter, and is the horizontal coordinate of the known data point. and is the function value corresponding to the data point. and is the derivative (tangent slope) value corresponding to the data point.

[0101] In order to further improve the reliability and stability of the data after interpolation fitting, Gaussian filtering is used to optimize the temperature data. Gaussian filtering is a technology widely used in signal processing and data smoothing. It smoothes the data by weighted averaging, which can effectively remove noise in the data. The local Gaussian smoothing algorithm uses a dynamic window calculation method to adaptively adjust the smoothing parameters according to the changes in short-term time intervals. To address the problem of short-term fluctuations, the algorithm calculates the median of the local time interval and constructs a smoothing weight based on the Gaussian kernel function. The weight value decreases as the distance between the data point and the center point increases. The core idea is to calculate the weight through a Gaussian function and then smooth the temperature data. The mathematical expression of the Gaussian function is:

[0102] (8).

[0103] in: For time point After smoothing, the data is the original time series data point; is the Gaussian weight, defined as:

[0104] (9).

[0105] in: For time point After smoothing, the data It is an adaptive window, dynamically calculated based on the median of short-term time intervals: ,in Represents the set of time intervals within the window around the current time point, ensuring that smoothing targets short-term fluctuations.

[0106] Since smoothing calculation may lack sufficient neighborhood data at the data endpoints, the present invention introduces endpoint mean constraints:

[0107] (10).

[0108] in, The number of points in the endpoint area is used to prevent abnormal deviations in endpoint data.

[0109] The beneficial effects of the above scheme are: in temperature data processing, the local Gaussian smoothing algorithm adopts a dynamic window calculation method, and the smoothing parameter is adaptively adjusted according to the short-term time interval change. For the short-term fluctuation problem, the algorithm calculates the median of the local time interval, and constructs the smoothing weight based on the Gaussian kernel function. Since the Gaussian weight decays exponentially with the increase of the time distance, the points far from the center have less influence on the current time point, so that the long-term trend is kept unchanged and only the short-term abnormal fluctuation is eliminated. In addition, the algorithm introduces a mean constraint mechanism at the data boundary to ensure the smoothness of the endpoint data while keeping the overall data trend unchanged. The local Gaussian smoothing algorithm is particularly suitable for processing data containing random noise, and it can effectively retain the main trend of the data and eliminate unnecessary details.

[0110] In S4, the Adaptive DTW+mean Euclidean distance+hierarchical clustering algorithm is used to divide the typical temperature season, and the measured temperature data is classified according to months. Using the adaptive dynamic time warping (Adaptive DTW) algorithm, the similarity between the temperature sequences of the 12 months is calculated pair by pair, and a symmetric DTW distance matrix D is generated:

[0111] .

[0112] The DTW distance matrix is a multi-dimensional symmetric matrix representing the similarity relationship between months.

[0113] The mean of the data of each month is also calculated, and the Euclidean distance is calculated based on the mean , the DTW distance matrix and the Euclidean distance matrix are normalized and then weighted, and the weighting formula is as follows:

[0114] .

[0115] The weighted comprehensive distance matrix is obtained by using the condensation hierarchical clustering method. After clustering, the 12 months are divided into k clusters, and each cluster corresponds to a similar seasonal pattern. Among them, indicates the weight, which is generally set to 0.3 to 0.5, dtw can handle data of different sequence lengths, but if the month data is less, excessive stretching will affect the accuracy of the extracted distance feature matrix, and if there is more missing data in the month, the weight is lower.

[0116] The beneficial effects of the above scheme are: since DTW can handle time series of different lengths and calculate similarity through alignment, it is very suitable for capturing the changing trend in seasonal time series, and the adaptive algorithm significantly reduces the computational complexity of the DTW distance matrix. Euclidean distance provides efficient computation. After normalization, distance weighting fusion is performed based on the adjustable weight parameter to obtain an optimized clustering effect.

[0117] In S5, according to the cluster analysis results, different seasonal patterns are analyzed respectively. The extreme value of the daily temperature difference between the two measuring points is used as a random variable, and the generalized extreme value distribution is used for probability fitting. The probability density formula is as follows:

[0118] (13).

[0119] The cumulative distribution function formula is as follows:

[0120] (14).

[0121] in, : Position parameter, controlling the center of the data; : scale parameter, which controls the width or variability of the data; : shape parameter, which determines the tail behavior of the distribution: >0: has heavy tails (Frechet distribution); =0: Gumbel distribution, applicable to the extreme value of exponential distribution; <0: has bounded tails (Weibull distribution).

[0122] The maximum likelihood estimation is used to determine the distribution parameters, thereby determining the distribution form of the daily extreme temperature difference.

[0123] In S6, based on the fitted probability distribution form of S5, the corresponding representative value of temperature effect is proposed based on the 50-year return period and exceedance probability. The calculation formula is as follows:

[0124] (15).

[0125] (16).

[0126] (17).

[0127] (18).

[0128] (19).

[0129] in: X is the daily temperature difference extreme value, is the typical number of days in summer with a 50-year return period; is the typical number of days in winter with a 50-year return period; is the random variable of daily temperature difference in summer; is the winter daily temperature difference random variable; is the exceeding probability of 50-year return period, the expected number of times of exceeding the temperature threshold value of the extreme temperature event of 50-year return period within the design reference period of 100 years is 2, = 0.02; is the shape parameter of summer; is the shape parameter of winter; is the scale parameter of summer; is the scale parameter of winter; is the location parameter of summer; is the location parameter of winter. is the temperature action representative value.

[0130] The specific effects of the present application include:

[0131] Due to the time series discontinuity and time interval variability of temperature monitoring caused by site conditions, the adaptive interpolation algorithm adopts a region marking method to identify long and short NaN segments (data anomalies) and process them differently. Small missing data (less than 10 data points) are interpolated and completed, while long missing data are deleted to avoid interpolation error accumulation and ensure interpolation accuracy. The local Gaussian smoothing algorithm uses a dynamic window size calculation method to adjust the Gaussian weight according to the short-term time interval changes. The local time interval median is calculated to avoid information loss caused by excessive smoothing. Only short-term fluctuation data is smoothed to ensure that the long-term trend information of the time series is not affected. The adaptive dynamic time warping (Adaptive DTW) and mean Euclidean distance are used to obtain the difference distance matrix of temperature change characteristics of different months. A weight parameter is used to combine the Adaptive DTW and Euclidean distance to calculate the comprehensive distance matrix. According to the comprehensive distance matrix, the months with similar temperature change characteristics are divided into the same class through agglomerative hierarchical clustering analysis, and multiple temperature change patterns are obtained. The typical summer temperature months and typical winter temperature months are determined through multiple temperature change patterns, and the typical temperature seasons are reasonably divided. Finally, the temperature action representative value is obtained according to the reasonable division result, achieving the purpose of accurately obtaining the temperature action representative value.

[0132] The embodiment of the present application provides a temperature representative value determination system, comprising:

[0133] A data acquisition module is configured to acquire abnormal temperature data of a steel-concrete composite structure within one year, use an adaptive dynamic time warping algorithm (Adaptive DTW) and a mean Euclidean distance to respectively acquire a difference distance matrix of temperature change characteristics of different months, and perform weighted processing on the difference distance matrix of the Adaptive DTW and the difference distance matrix of the mean Euclidean distance to obtain a comprehensive distance matrix.

[0134] The typical seasonal temperature month acquisition module is used to perform agglomerative hierarchical clustering analysis on the comprehensive distance matrix, classify months with similar temperature change characteristics within a year into the same category, obtain the seasonal pattern of temperature change, and determine the typical summer temperature month and the typical winter temperature month.

[0135] The temperature action representative value acquisition module is used to select temperature measurement points on the steel-concrete composite structure, obtain the daily extreme temperature differences of the measurement points in typical summer temperature months and typical winter temperature months, and use the generalized extreme value distribution model to obtain the probability distribution of the daily extreme temperature differences; based on the probability distribution, the exceedance probability is used to obtain the representative values ​​of the temperature effects with different recurrence periods.

[0136] A specific embodiment is as follows:

[0137] This embodiment discloses a method for determining a temperature representative value, and the specific steps are as follows:

[0138] S1. First, extract the time-course temperature data and use the machine learning algorithm to clean and fit the data.

[0139] S2. Based on the results of data cleaning, temperature difference data is calculated with months as the basic unit. Adaptive dynamic time warping (DTW) and mean Euclidean distance are used to obtain the difference distance matrix of temperature change characteristics in different months. A weight parameter is used to combine the normalized Adaptive DTW distance matrix and the Euclidean distance matrix to calculate a weighted comprehensive distance matrix. Agglomerative hierarchical clustering analysis is performed based on the comprehensive distance matrix to divide the typical seasons.

[0140] S3. Secondly, the data of the typical seasons after division were analyzed separately. The daily extreme value of the temperature difference between the two measuring points was used as a random variable. The generalized extreme value distribution was used to fit its probability distribution, and the distribution parameters were estimated using maximum likelihood.

[0141] S4. Finally, using the fitted probability distribution results, based on the return period and exceedance probability, representative values ​​of temperature effects in typical seasons are proposed.

[0142] Finally, a classification demonstration diagram of a method for determining a temperature representative value, such as Figure 6 Shown is a dendrogram of hierarchical clustering. Figure 7 This is the hierarchical clustering result based on adaptive DTW+Euclidean distance.

[0143] The method of the present invention is a method for calculating representative values ​​of temperature effects of a combination of a machine learning algorithm and a generalized extreme value distribution, and is applicable to engineering structures with long-term measured temperature data. According to the isolation forest algorithm, in an unsupervised and labeled form, the outliers of deviations from potential trends in the measured data are analyzed more efficiently, thereby improving the effectiveness and reliability of outlier detection. At the same time, the problem of inaccurate seasonal division caused by seasonal differences in different regions is solved by the Adaptive DTW+mean Euclidean distance+hierarchical clustering algorithm. By extracting the characteristic values ​​of data from different months, the classification problem of typical seasons of temperature time series data is calculated more efficiently, which is applicable to the problem of typical seasonal division when analyzing seasonal differences in different regions.

[0144] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for determining a temperature representative value, characterized in that: The following steps are involved: Abnormal temperature data of steel-concrete composite structures over a year was obtained. The Adaptive Dynamic Time Warping algorithm (AdaptiveDTW) and the mean Euclidean distance (MED) were used to obtain the difference distance matrices of temperature variation characteristics for different months. The Adaptive DTW difference distance matrix and the mean Euclidean distance difference distance matrix were weighted to obtain a comprehensive distance matrix. Agglomerative hierarchical cluster analysis was performed on the comprehensive distance matrix to classify months with similar temperature change characteristics into the same cluster, thereby obtaining the seasonal pattern of temperature change and determining the typical summer temperature month and the typical winter temperature month. Temperature measurement points on the steel-concrete composite structure were selected, and the daily extreme temperature differences at the measurement points were obtained in typical summer and winter temperature months. The probability distribution of the daily extreme temperature differences was obtained using a generalized extreme value distribution model. Based on the probability distribution, the exceedance probability was used to obtain representative values ​​of temperature effects with different return periods. The method of using a generalized extreme value distribution model to obtain a probability distribution of the daily extreme temperature difference comprises the following steps: According to the extreme value of the daily temperature difference between the two detection points as the random variable x, the generalized extreme value distribution is used to perform probability fitting f(x; μ, σ, ξ), the formula is: Among them, μ is the location parameter, which controls the center of the data; σ is the scale parameter, which controls the width or variability of the data; ξ is the shape parameter, which determines the tail behavior of the distribution; The cumulative distribution function F(x; μ, σ, ξ) is: The distribution parameters are determined by using the maximum likelihood estimation, and the probability distribution of the daily extreme temperature difference is determined based on the distribution parameters. The method of using the exceedance probability to obtain representative values ​​of temperature effects with different return periods specifically comprises the following steps: According to the probability distribution of daily extreme temperature difference in different return periods, based on a 50-year return period, the exceedance probability is used to obtain the corresponding representative value of the temperature effect. The formula is: P=1-F Y (X); Among them, X is the extreme value of daily temperature difference, N S is the typical number of days in summer during the 50-year return period; N W is the typical number of days in winter with a 50-year return period; is the random variable of daily temperature difference in summer; is the winter daily temperature difference random variable; P is the exceedance probability with a return period of 50 years. During the 100-year design reference period, the expected number of times an extreme temperature event with a return period of 50 years exceeds the temperature threshold is 2 times, P = 0.02; ξ s is the shape parameter in summer; ξ W is the shape parameter for winter; σ s is the scale parameter for summer; σ W is the scale parameter in winter; μ s is the location parameter in summer; μ W is the location parameter for winter; Inversely solve the temperature difference extreme value x k , which is the representative value of temperature effect.

2. The method for determining a temperature representative value according to claim 1, wherein: The specific steps of obtaining abnormal temperature data of the steel-concrete composite structure within one year include: Using a gradient detection method to identify data with sharp changes in the annual temperature data time history of the steel-concrete composite structure; wherein the annual temperature data time history is used to represent temperature change data at different time points within a year; The box plot detection method is used to filter out abnormal temperature values ​​that deviate from the normal range from the annual temperature data time course; Identify outliers from the annual temperature data time history using the Isolation Forest algorithm; Based on the identification and screening results, abnormal temperature data within one year is obtained.

3. A method for determining a temperature representative value according to claim 2, characterized in that: The agglomerative hierarchical clustering analysis is performed on the comprehensive distance matrix to classify months with similar temperature change characteristics within a year into the same category, thereby obtaining a seasonal pattern of temperature change. The specific steps include: Adaptive interpolation algorithm is used to fit and replace abnormal temperature data within one year, and Gaussian filtering is used to smooth the replacement results. The smoothing results are classified by month, and dynamic time warping (DTW) is used to pair the 12 months. The similarity between the temperature series of the 12 months is obtained pair by pair, and a symmetrical DTW distance matrix DTW_dist is generated: Where n represents the maximum month of a year, 12; Get the mean of the data for each month and get the Euclidean distance matrix D based on the mean Euclidean ; Normalize the DTW distance matrix and the Euclidean distance matrix, and weight the normalized results: D combine =ω×DTW_dist+(1-ω)×D Euclidean ; Among them, ω represents the weight; Distance matrix D combine Agglomerative hierarchical clustering method was used to cluster the 12 months into k clusters, each cluster corresponding to a seasonal pattern and the seasonal pattern was regarded as a temperature change pattern.

4. A system for determining a temperature representative value, characterized in that: include: The data acquisition module is used to obtain abnormal temperature data of steel-concrete composite structures within a year. The Adaptive Dynamic Time Warping algorithm (Adaptive DTW) and the mean Euclidean distance are used to obtain the difference distance matrix of temperature change characteristics in different months. The difference distance matrix of Adaptive DTW and the difference distance matrix of mean Euclidean distance are weighted to obtain a comprehensive distance matrix. The module for obtaining typical seasonal temperature months is used to perform agglomerative hierarchical clustering analysis on the comprehensive distance matrix, classify months with similar temperature change characteristics within a year into the same category, obtain the seasonal pattern of temperature change, and determine the typical summer temperature month and the typical winter temperature month; The temperature action representative value acquisition module is used to select temperature measurement points on the steel-concrete composite structure, obtain the daily extreme temperature differences of the measurement points in typical summer temperature months and typical winter temperature months, and use the generalized extreme value distribution model to obtain the probability distribution of the daily extreme temperature differences. Based on the probability distribution, the exceedance probability is used to obtain the temperature action representative values ​​with different recurrence periods. The method of using a generalized extreme value distribution model to obtain a probability distribution of the daily extreme temperature difference comprises the following steps: According to the extreme value of the daily temperature difference between the two detection points as the random variable x, the generalized extreme value distribution is used to perform probability fitting f(x; μ, σ, ξ), the formula is: Among them, μ is the location parameter, which controls the center of the data; σ is the scale parameter, which controls the width or variability of the data; ξ is the shape parameter, which determines the tail behavior of the distribution; The cumulative distribution function F(x; μ, σ, ξ) is: The distribution parameters are determined by using the maximum likelihood estimation, and the probability distribution of the daily extreme temperature difference is determined based on the distribution parameters. The method of using the exceedance probability to obtain representative values ​​of temperature effects with different return periods specifically comprises the following steps: According to the probability distribution of daily extreme temperature difference in different return periods, based on a 50-year return period, the exceedance probability is used to obtain the corresponding representative value of the temperature effect. The formula is: P=1-F Y (X); Among them, X is the extreme value of daily temperature difference, N S is the typical number of days in summer during the 50-year return period; N W is the typical number of days in winter with a 50-year return period; is the random variable of daily temperature difference in summer; is the winter daily temperature difference random variable; P is the exceedance probability with a return period of 50 years. During the 100-year design reference period, the expected number of times an extreme temperature event with a return period of 50 years exceeds the temperature threshold is 2 times, P = 0.02; ξ s is the shape parameter in summer; ξ W is the shape parameter for winter; σ s is the scale parameter for summer; σ W is the scale parameter in winter; μ s is the location parameter in summer; μ W is the location parameter for winter; Inversely solve the temperature difference extreme value x k , which is the representative value of temperature effect.

Images