Steel structure elevator shaft settlement prediction method based on LSTM model

Abstract

The invention relates to the technical field of structure health monitoring, in particular to a steel structure elevator shaft settlement prediction method based on an LSTM (Long Short Term Memory) model, which comprises the following steps: analyzing three-axis dip angle data, screening a dip angle change effective section, extracting three-dimensional displacement data to identify a dip angle linkage mode interval, identifying offset amplitude and constructing a sample weight distribution table; and detecting a jump point, setting a prediction path to obtain settlement prediction information, and comparing residual error identification time migration to generate structure prediction migration trend data. According to the method, the effective data segment is extracted by identifying the dip angle inflection point and combining the three-axis change amplitude difference, the direction accuracy of structural trend judgment is enhanced by combining the main shaft displacement direction identification, the weight of the training sample is adjusted by combining the data offset amplitude, the learning ability of the model for the key deformation time period is optimized, and by utilizing the jump density, the structural trend judgment accuracy is improved. A prediction path starting point is set, behavior responsiveness of trend prediction is improved, and interpretability of a prediction result is enhanced in combination with translation state residual fluctuation analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring, and particularly to a method for predicting the settlement of a steel structure elevator shaft based on an LSTM model. Background Art

[0002] The technical field of structural health monitoring includes real-time or periodic monitoring, evaluation, and early warning of the states of key load-bearing components such as building structures, bridges, tunnels, and elevator shafts. The core content is to collect physical change data of the structure during operation, such as inclination angle, displacement, strain, vibration, etc., by deploying sensor devices, and to judge and predict the structural health state through establishing mathematical models, signal analysis, and trend prediction methods. The systematicness of structural health monitoring covers multiple links such as data acquisition, state recognition, model construction, threshold judgment, and early warning release, aiming to provide a basis for structural maintenance and operation decision-making and avoid safety accidents caused by structural anomalies or damages.

[0003] Among them, the method for predicting the settlement of a steel structure elevator shaft based on an LSTM model refers to a method that, based on the characteristics of time series data processing, uses a long short-term memory network model to model and predict the inclination angle and settlement displacement in the structural state of the elevator shaft, involving the inclination and settlement changes that occur during the long-term use of the steel structure elevator shaft, especially the prediction problem of small deformations caused by environmental, load, and structural characteristics. Specifically, it includes collecting the inclination angle and displacement data of the elevator shaft by installing Beidou GNSS monitoring equipment, constructing a time series monitoring data set as the basis, inputting it into the LSTM neural network model, using the internal input gate, forget gate, and output gate structures to process historical data, forming a prediction output for the subsequent time node state, and completing model optimization through multiple rounds of model training and error calculation.

[0004] Traditional steel structure elevator shaft settlement prediction technologies adopt a fixed threshold judgment method in the screening logic of inclination and settlement data, lacking a fine-grained response judgment for continuous time evolution behaviors, resulting in the inability to effectively extract some small deformations in the preprocessing stage, ignoring the dynamic coordination relationship between the inclination angle and three-dimensional displacement in the spatial direction, causing the identified feature segments to have a drift risk in the spatial direction, and adjusting weights mainly based on the mean difference or the degree of outliers during the sample training process, being unable to effectively identify and strengthen the attention to multiple high-amplitude deformation processes, making the model lose sensitivity to key deformation periods of the structure during the training stage, the prediction path starts at a fixed time point, resulting in an offset of the path starting point, lacking a structural state evolution driving mechanism, and reducing the structural stability and interpretability of the overall prediction chain. Summary of the Invention

[0005] To solve the technical problems existing in the prior art, an embodiment of the present invention provides a method for predicting the settlement of a steel structure elevator shaft based on an LSTM model, which optimizes the learning ability of the model for key deformation periods, uses the jump density, sets the starting point of the prediction path, improves the behavioral responsiveness of trend prediction, and combines the translation state residual fluctuation analysis to enhance the interpretability of the prediction results.

[0006] To achieve the above object, the present invention adopts the following technical solutions. A method for predicting the settlement of a steel structure elevator shaft based on an LSTM model includes the following steps: S1: Obtain the inclination data of the X, Y, and Z axes. According to the slope change direction of the inclination data of each axis at consecutive time points, identify the inflection point positions, compare the inclination change amplitudes of each axis in the same time period, analyze the distribution state of the inclination differences in time, and screen the effective sections of inclination changes. S2: Based on the effective sections of inclination changes, obtain the three-dimensional displacement data of the structure, calculate the included angle and modulus length change of the displacement vectors between adjacent time points, judge the consistency of the displacement directions, analyze the stability of the main axis direction, identify the time periods when the displacement directions are consistent with the inclination change trends, and generate the inclination linkage mode intervals. S3: Invoke the inclination linkage mode intervals, extract the corresponding inclination and settlement data, calculate the change amplitude relative to the mean value of the sample library, judge the deviation degree of the data in each time period, and construct a sample weight distribution table. S4: According to the inclination linkage mode intervals and the sample weight distribution table, analyze the inclination change sequence and detect the micro jump points. According to the distribution density of the micro jump points, identify the peak sections of the change frequency, set the starting position of the trend prediction path, predict the settlement behavior of the elevator shaft, and generate the settlement prediction information.

[0007] As a further solution of the present invention, the effective sections of inclination changes are specifically the multi-axis inclination joint change area, the time series slope inversion area, and the local amplitude difference aggregation area. The inclination linkage mode intervals include the main axis direction continuous section, the displacement inclination coordination section, and the spatial displacement stable section. The sample weight distribution table includes the data deviation grouping item, the sample priority label, and the training stage weight coefficient. The settlement prediction information is specifically the settlement trend path, the jump dense starting position, and the predicted target data sequence.

[0008] As a further solution of the present invention, the step of obtaining the inclination data of the X, Y, and Z axes, according to the slope change direction of the inclination data of each axis at consecutive time points, identifying the inflection point positions, comparing the inclination change amplitudes of each axis in the same time period, analyzing the distribution state of the inclination differences in time, and screening the effective sections of inclination changes is specifically as follows: S101: Obtain the inclination data of the X, Y, and Z axes, identify the slope directions at multiple time points in each axis inclination sequence, extract the inflection point positions in each inclination curve, calculate the time intervals between adjacent inflection points in each curve and record the inflection point indices, and generate an inclination change trend node group; S102: Based on the inclination change trend node group, call the inclination values of the X, Y, and Z axes in the same time period, analyze the distribution state in time of the differences in inclination change amplitudes by comparing the inclination change amplitudes of each axis in the same time period, and generate an inclination multi-axis difference change interval; S103: According to the inclination multi-axis difference change interval and the distribution state in time of the differences in inclination change amplitudes, identify the continuous change trend and screen the effective sections of inclination change.

[0009] As a further solution of the present invention, the steps of obtaining the three-dimensional displacement data of the structure based on the effective section of inclination change, calculating the included angle and modulus length change of the displacement vectors between adjacent time points, judging the consistency of the displacement directions, analyzing the stability of the main axis direction, and identifying the time periods in which the displacement directions are consistent with the inclination change trends to generate an inclination linkage mode interval are specifically as follows: S201: Based on the effective section of inclination change, obtain the X, Y, and Z three-dimensional displacement data in the corresponding time period, construct a spatial displacement vector group for each adjacent time point according to the time series order, calculate the included angle and modulus length change values of each group of vectors, and record the direction change trend and length change amplitude corresponding to each time point to generate a spatial displacement continuous change feature group; S202: Call the spatial displacement continuous change feature group, perform a consistency analysis on the change directions of the vector included angles in continuous time periods, analyze the stability of the displacement main axis in the corresponding time periods, extract the direction stable sections, and generate a set of main axis direction stable intervals; S203: According to the set of main axis direction stable intervals, obtain the inclination change trend sequence in the corresponding time period, identify the time periods in which the displacement directions are consistent with the inclination change trends, and generate an inclination linkage mode interval.

[0010] As a further solution of the present invention, the steps of calling the inclination linkage mode interval, extracting the corresponding inclination and settlement data, calculating the change amplitude relative to the mean value of the sample library, judging the deviation degree of the data in each time period, and constructing a sample weight distribution table are specifically as follows: S301: Call the inclination linkage mode interval, extract the inclination sequence and settlement sequence in the corresponding time period, calculate the difference between the inclination value at each time point and the corresponding inclination mean value in the sample library, calculate the difference between the settlement value and the settlement mean value in the sample library, and generate an inclination settlement deviation amplitude group; S302: Based on the inclination settlement offset amplitude group, sort the relative positions of the fluctuation amplitudes in each time period in the sample set, and generate sample offset level distribution labels according to the offset degree of the data in each time period; S303: According to the sample offset level distribution labels, adjust the training weights of the corresponding samples according to the offset degree, and establish a sample weight allocation table.

[0011] As a further solution of the present invention, according to the inclination linkage mode interval and the sample weight allocation table, analyze the inclination change sequence and detect micro amplitude jump points, and according to the distribution density of the micro amplitude jump points, identify the change frequency peak section, set the starting position of the trend prediction path, and predict the settlement behavior of the elevator shaft to generate settlement prediction information. The specific steps are as follows: S401: According to the inclination linkage mode interval and the sample weight allocation table, extract the inclination change sequence in the corresponding time period, detect micro amplitude jump points by comparing with the micro amplitude jump recognition threshold, and generate micro amplitude jump point information; S402: Based on the micro amplitude jump point information, analyze the distribution density of the micro amplitude jump points, identify the change frequency peak section according to the position of the jump points in the time series, and set it as the starting position of the trend prediction path to generate trend path starting position parameters; S403: According to the trend path starting position parameters, combine the inclination change directions of consecutive time points in the corresponding section to predict the settlement behavior of the elevator shaft and generate settlement prediction information.

[0012] As a further solution of the present invention, the specific formula for identifying the change frequency peak section is: ; Calculate the jump distribution density index; where, is the jump distribution density index in the th time window, is the index of the jump point in the current time window, is the index of the time window, is the original time index of the rd jump point, is the original time index of the th jump point, is the total time length of the current time window, is the training sample weight value corresponding to the rd jump point, is the normalized value of the inclination change amplitude of the th jump point, is the residual stability correction term constant to prevent division by zero operation, is the total number of micro jump points detected within the th time window.

[0013] As a further aspect of the present invention, it further includes: S5: Invoke the settlement prediction information, obtain the residual sequence between the predicted value and the actual value, analyze the prediction error at each time step, adjust the translation step size within the time window, calculate the error mean value under each translation state, and by comparing the residual means of each translation step, identify the optimal translation state and the corresponding time offset distance according to the stability of the error and the residual fluctuation, and generate the structural prediction offset trend data; The structural prediction offset trend data includes the model time response offset, the minimum residual value of translation matching, and the trend error classification label.

[0014] As a further aspect of the present invention, the steps of invoking the settlement prediction information, obtaining the residual sequence between the predicted value and the actual value, analyzing the prediction error at each time step, adjusting the translation step size within the time window, calculating the error mean value under each translation state, and by comparing the residual means of each translation step, identifying the optimal translation state and the corresponding time offset distance according to the stability of the error and the residual fluctuation, and generating the structural prediction offset trend data are specifically as follows: S501: Invoke the settlement prediction information, obtain the predicted value sequence and the actual value sequence, and construct the difference sequence between the data sequences to generate the prediction error residual sequence; S502: Invoke the prediction error residual sequence, set multiple consecutive time windows, adjust the translation step size of the predicted data sequence relative to the actual monitoring data sequence, and calculate the residual average value of the corresponding time window under each translation state to generate a list of translation state error mean values; S503: Invoke the list of translation state error mean values, identify the optimal translation state and the corresponding time offset distance by analyzing the stability of the error and the residual fluctuation of each translation step size, and mark it as the response offset amount predicted by the model to generate the structural prediction offset trend data.

[0015] As a further aspect of the present invention, the specific formula for analyzing the stability of the error and the residual fluctuation of each translation step size is: ; Calculate the residual fluctuation score value; Among them, is the normalized residual fluctuation score value with the translation step size number , is the arithmetic mean value of all prediction residual values within the time window with the translation step size number , is the translation step size number and the time point index is for a single predicted residual value, is the number of time points within the current time window, is for all translation step numbers the maximum value in the absolute mean of the residuals, is the constant correction term value, represents the current translation step number for analysis, represents the discrete time point index within the current time window, represents the traversal index number for all translation steps to be compared.

[0016] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include: By identifying the inflection point of the inclination angle and combining the differences in the variation amplitudes of the three axes, the extraction of effective data segments is realized. Combining the identification of the main axis displacement direction enhances the direction accuracy of the structural trend judgment. Adjusting the training sample weights in combination with the data offset amplitude optimizes the learning ability of the model for key deformation periods. Using the jump density to set the starting point of the prediction path improves the behavioral responsiveness of the trend prediction. Through the identification of the translation state residual fluctuations, the interpretability of the model prediction results and the tolerance identification ability are enhanced, the accuracy of the structural state identification is improved, and the training focus of the model and the rationality of the prediction error tolerance are enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is a schematic diagram of the working process of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following will describe the technical solutions in the present invention in conjunction with the drawings.

[0020] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two.

[0021] To make the technical problems to be solved, technical solutions and advantages of the present invention clearer, the following will be described in detail in conjunction with the drawings and specific embodiments.

[0022] Please refer to Figure 1 , the present invention provides a technical solution, a method for predicting the settlement of a steel structure elevator shaft based on an LSTM model, including the following steps: S1: Obtain the inclination data of the X, Y, and Z axes. According to the slope change direction of the inclination data of each axis at consecutive time points, identify the inflection point positions, compare the inclination change amplitudes of each axis in the same time period, analyze the distribution state of the inclination differences in time, and screen the effective sections of inclination changes; S2: Based on the effective sections of inclination changes, obtain the three-dimensional displacement data of the structure, calculate the included angle and modulus length change of the displacement vectors between adjacent time points, judge the consistency of the displacement directions, analyze the stability of the main axis direction, identify the time periods when the displacement directions and the inclination change trends are consistent, and generate the inclination linkage mode intervals; S3: Call the inclination linkage mode intervals, extract the corresponding inclination and settlement data, calculate the change amplitude relative to the mean value of the sample library, judge the deviation degree of the data in each time period, and construct a sample weight distribution table; S4: According to the inclination linkage mode intervals and the sample weight distribution table, analyze the inclination change sequence and detect the micro amplitude jump points. According to the distribution density of the micro amplitude jump points, identify the peak sections of the change frequency, set the starting position of the trend prediction path, predict the settlement behavior of the elevator shaft, and generate settlement prediction information; S5: Call the settlement prediction information, obtain the residual sequence between the predicted value and the actual value, analyze the prediction error at each time step, adjust the translation step size within the time window, calculate the mean error under each translation state, and by comparing the residual means of each translation step, according to the stability of the error and residual fluctuations, identify the optimal translation state and the corresponding time offset distance, and generate the structural prediction offset trend data.

[0023] The effective sections of inclination changes are specifically the multi-axis inclination joint change area, the time series slope reversal area, and the local amplitude difference aggregation area. The inclination linkage mode intervals include the main axis direction continuous section, the displacement inclination coordination section, and the spatial displacement stable section. The sample weight distribution table includes data offset grouping items, sample priority labels, and training stage weight coefficients. The settlement prediction information is specifically the settlement trend path, the starting position of the jump dense area, and the predicted target data sequence. The structural prediction offset trend data includes the model time response offset amount, the minimum residual value of translation matching, and the trend error classification label.

[0024] The steps of obtaining the inclination data of the X, Y, and Z axes, according to the slope change direction of the inclination data of each axis at consecutive time points, identifying the inflection point positions, comparing the inclination change amplitudes of each axis in the same time period, analyzing the distribution state of the inclination differences in time, and screening the effective sections of inclination changes are specifically as follows: S101: Obtain the inclination angle data of the X, Y, and Z axes, identify the slope directions at multiple time points in each axis inclination angle sequence, extract the inflection point positions in each inclination angle curve, calculate the time intervals between adjacent inflection points in each curve, record the inflection point indices, and generate an inclination angle change trend node group; After obtaining the inclination angle data of the X, Y, and Z axes, the three-axis inclination angle sequences are collected and recorded at equal time intervals as 、 、 , where is the time point index, and the interval time is set to 30 seconds, forming an inclination angle time series array indexed by time. By performing the slope direction judgment operation on 、 、 sequences, that is, for each calculate the inclination angle difference between the current and the previous moment , where is the inclination angle change amount of the current axis at time , is the inclination angle value at the current moment, is the inclination angle value at the previous moment, . If then it is marked as an upward trend. If then it is marked as a downward trend. When the trend of three consecutive points changes from upward to downward or from downward to upward, it is identified as an inflection point. The inflection point time is recorded as , which forms an inflection point sequence 、 、 on each axis, and calculate the time interval between each adjacent inflection point , where is the time interval between adjacent inflection points, is the time of the latter inflection point, is the time of the previous inflection point. Record to form an inflection point time interval array , and each inflection point corresponds to a label index . Further integrate it into an inclination angle change trend node set. In actual tests, when the data at the inclination angle measurement point of the elevator shaft is like , then at towards the change point is a downward trend inflection point. At a sampling frequency of 30 seconds, the corresponding is 30 seconds. This judgment logic is executed by axis, forming a node list for each curve to form a complete trend node group.

[0025] S102: Based on the inclination change trend node group, call the inclination values of the three axes of X, Y, and Z within the same time period. By comparing the inclination change amplitudes of each axis within the same time period, analyze the distribution state of the differences in inclination change amplitudes over time, and generate the multi-axis difference change interval of inclination. Based on the inclination change trend node group, extract the inclination values of the nodes within the time period from the three axes of X, Y, and Z. For the inclination curve of each axis, extract a subsequence with a length of 5 time points within the corresponding time period, calculate the difference between the maximum and minimum inclination values in this section, where, is the inclination fluctuation amplitude at the th node of the axis, is the inclination value at the th moment in the sequence, . Calculate , , in sequence, and form them into an inclination change amplitude group . Judge the maximum difference among the change amplitudes of the three axes of this node, where, represents the maximum difference among the inclination change amplitudes of the three axes at the th time point, and are the maximum and minimum fluctuation amplitudes among the three axes respectively. If degrees, it is marked as having multi-axis differential changes. The system default difference threshold is degrees, which is jointly determined by the equipment error tolerance and the structural allowable deformation standard. For example, if , , , then , marked as a differential change segment. All time periods that satisfy constitute the multi-axis differential change interval of inclination.

[0026] S103: According to the multi-axis differential change interval of inclination, based on the distribution state of the differences in inclination change amplitudes over time, identify the continuous change trend and screen the effective area of inclination change; According to the multi-axis differential change interval of inclination, screen the time periods that continuously satisfy , judge whether the number of consecutive nodes that satisfy the conditions in each segment is greater than the consecutive determination criterion . If the number of consecutive nodes that satisfy the conditions is greater than or equal to 3, it is regarded as a continuous effective change trend segment. Compare the overlapping ranges of the three-axis differential sections, and take the time intersection covered by the three axes together, denoted as , where, Indicates the effective intersection segments of the X, Y, and Z axes on the time axis. Check each segment to see if there are at least two positive and negative flips in the inclination sequence, that is, the product of the slopes at consecutive adjacent time points satisfies , where represents the time point and is the slope of the axis represents the slope at time . If there are two or more pairs of points that satisfy this product condition, include this section in the set of effective inclination change sections. Example: In a section from 90 seconds to 150 seconds in an elevator shaft, 6 consecutive nodes are detected, and there are at least two direction reversals in both X and Y. This time period is recorded as an effective inclination change section.

[0027] Based on the effective sections of inclination change, obtain the three-dimensional displacement data of the structure, calculate the included angle and modulus length change between adjacent time points, judge the consistency of the displacement direction, analyze the stability of the main axis direction, identify the time periods when the displacement direction and the inclination change trend are consistent, and generate the steps for the inclination linkage mode interval are as follows: S201: Based on the effective sections of inclination change, obtain the three-dimensional displacement data of X, Y, and Z within the corresponding time period. According to the time series order, construct a spatial displacement vector group for each adjacent time point, calculate the included angle and modulus length change value of each group of vectors, and record the direction change trend and length change amplitude corresponding to each time point to generate a continuous change feature group of spatial displacement; Based on the effective sections of inclination change, obtain the three-dimensional displacement data of X, Y, and Z within the corresponding time period, and represent the displacement of each time point in the form of a ternary coordinate vector , where represents the time index, respectively represent the spatial position values in the three-axis directions at this moment. The data is collected at 5-second time intervals. According to the time series order, construct a spatial displacement vector group for every two adjacent time points , that is, obtain it by subtracting the three coordinate components respectively . Record the modulus length change of each vector group, and calculate the amplitude of each spatial displacement through ; the direction change trend is obtained by calculating the included angle between adjacent vectors, and the included angle is calculated using the cosine formula . The result is between . If , it is judged that the direction is stable. If , it is judged that the direction has a sudden change. Record the modulus length and included angle change values of each time point as the spatial displacement characteristics of this point, and form a continuous change feature group of spatial displacement for the entire sequence. In the example, if there are three points , , , then and , the included angle between the two vectors is about 5.7 degrees, and it is judged that the direction changes stably.

[0028] S202: Invoke the continuously changing feature group of spatial displacement to perform consistency analysis on the changing direction of the included angle between vectors within a continuous time period, analyze the stability of the displacement main axis within the corresponding time period, extract the directionally stable sections, and generate a set of main axis directionally stable intervals; Invoke the continuously changing feature group of spatial displacement for all The included angle values between adjacent vectors represented are traversed and analyzed. The paragraphs in the included angle sequence with continuously changing less than the set angle difference are extracted as directionally consistent intervals. The specific set direction consistency threshold is , and the converted cosine value is . By continuously judging whether at multiple time points continuously meets this condition, if the direction consistency is met at more than 3 consecutive time points, it is considered that the direction of this section is stable, and its start and end times are marked. Further analyze whether the change in the vector modulus length within this directionally stable section maintains a positive increasing or decreasing trend. If there are three consecutive sections with a monotonic change trend in the modulus length, it is recorded as the main axis offset direction being stable, and its time period index and the main vector axial direction of the vector change direction are recorded. For example, if the cosine values of the included angles at five time points within the sampling period are , then the first three points meet the included angle consistency condition. If at the same time their modulus lengths are in sequence, then it is in an increasing state, forming a directionally stable section, which is included in the set of main axis directionally stable intervals.

[0029] S203: According to the set of main axis directionally stable intervals, obtain the sequence of tilt angle change trends within the corresponding time period, identify the time period when the displacement direction and the tilt angle change trend are consistent, and generate a tilt angle linkage mode interval; According to the set of main axis directionally stable intervals, extract the time index within each stable interval, and obtain the sequence of tilt angle change trends within the corresponding time period , where , and further compare it with the current main vector direction of spatial displacement. Calculate the included angle difference between the main axis direction vector of spatial displacement and the change direction of the tilt angle of each axis. If the included angle between the tilt angle change direction and the main axis direction on a certain axis within this section is less than the set included angle consistency threshold (for example, 15 degrees), then it is judged that it is a tilt angle linkage trend direction within this section. If at least two axes meet the condition that the tilt angle direction and the displacement direction are the same within the same section, then the time index of this section is recorded in the tilt angle linkage mode interval. In the example, if the main axis direction is the Z axis, and the tilt angle change sequence is continuously increasing positively on the Z axis, and the included angle with the main axis is 9 degrees, then this section can be recognized as the tilt angle and the displacement direction being linked, and mark the start and end times of this section into the set of tilt angle linkage mode intervals.

[0030] Call the inclination linkage mode interval, extract the corresponding inclination and settlement data, calculate the change amplitude relative to the mean value of the sample library, judge the deviation degree of the data in each time period, and the steps of constructing the sample weight distribution table are specifically as follows: S301: Call the inclination linkage mode interval, extract the inclination sequence and settlement sequence in the corresponding time period, calculate the difference between the inclination value at each time point and the corresponding inclination mean value in the sample library, calculate the difference between the settlement value and the settlement mean value in the sample library, and generate an inclination settlement deviation amplitude group; Call the inclination linkage mode interval, extract the inclination value and settlement value at each moment in this interval, and record them as and respectively, where is the time point index, with the unit of second, the unit is degree, the unit is millimeter. Synchronously call the corresponding inclination and settlement reference mean values in the sample library during this period, and record them as and respectively. For each time point , calculate the inclination offset as , and the settlement offset as , where represents the change amplitude of the current inclination relative to the reference sample, represents the change amplitude of the current settlement relative to the reference sample. For all time points in the entire section, form offset difference sequences, which are and respectively. If it is set that in the reference sample , , and at a certain moment in the actual monitoring data, it is recorded as , , then the corresponding offset is , . For all time points, construct the offset amplitude set of this linkage section, and uniformly record it as the inclination settlement offset amplitude group.

[0031] S302: Based on the inclination settlement offset amplitude group, sort the relative positions of the fluctuation amplitudes of each time period in the sample set, and the deviation degree of the data in each time period, and generate a sample offset level distribution label; Based on the inclination settlement offset amplitude group, record the combined inclination and settlement offset of each time period as a two-dimensional feature vector , and sort the offset amplitudes of all labeled samples in this vector group in the overall sample set. The offset characteristics in the sample set are divided into levels according to a preset standard, and thresholds are set according to the amplitude combination interval. For example: the inclination offset is less than and the settlement offset is less than is the first level, and the inclination offset is in the range of and the settlement offset is within level 2, and if it exceeds this range, it is level 3. Based on this, the feature vectors corresponding to each time period are mapped to the above level labels, forming a sequence label group as the sample offset level distribution label for this time period. If the combined offset of five data points within a certain period is within the level 2 range, the entire segment label is set to "offset level 2".

[0032] S303: According to the sample offset level distribution label, adjust the training weights of the corresponding samples according to the degree of offset, and establish a sample weight distribution table; According to the sample offset level distribution label, different training weight values are assigned to the training samples corresponding to each level. The weight of level 1 samples is set to , level 2 is , level 3 is , where the weight values are taken from the recommended range of gradient adjustment factors in network training. During the entire training data construction phase, all time points in the linkage interval are mapped to their corresponding weight values according to their level labels, and paired with the original sample input data one by one. Finally, a mapping relationship matrix between each sample time and its training weight is formed, and the output is the sample weight distribution table. In the example, if the label sequence for a certain period is {2, 2, 3, 1, 2}, the corresponding weight sequence is {1.0, 1.0, 1.5, 0.5, 1.0}, and this sequence combined with the original sample index forms a complete training input template.

[0033] According to the inclination linkage mode interval and the sample weight distribution table, analyze the inclination change sequence and detect micro jump points. According to the distribution density of the micro jump points, identify the peak section of the change frequency, set the starting position of the trend prediction path, and predict the settlement behavior of the elevator shaft to generate settlement prediction information. The specific steps are as follows: S401: According to the inclination linkage mode interval and the sample weight distribution table, extract the inclination change sequence within the corresponding time period, and detect micro jump points by comparing with the micro jump recognition threshold, generating micro jump point information; According to the inclination linkage mode interval and the sample weight distribution table, extract the inclination change sequence within this interval. Let the inclination sequence be , where is the continuous time point index. Perform a first-order difference operation on the sequence to obtain the inclination change sequence , where represents the inclination change amount between two adjacent moments. Define the micro jump recognition threshold as , which is used as the occurrence criterion for judging jump behavior. Traverse all , if it satisfies and Less than the large abnormal fluctuation identification threshold (set as ), then this point is recorded as a micro amplitude jump point. Further, the sample weight allocation table is called to mark and distinguish the jump points within the high-weight section, forming the micro amplitude jump point triple information with the structure of , where is the training sample weight at the corresponding moment. In actual sampling, if , then , where the 1st, 2nd, and 4th items all meet the micro amplitude jump conditions. Record these time points and the corresponding weights to generate the micro amplitude jump point information.

[0034] S402: Based on the micro amplitude jump point information, analyze the distribution density of the micro amplitude jump points. According to the positions of the jump points in the time series, identify the peak section of the change frequency and set it as the starting position of the trend prediction path to generate the trend path starting position parameter; The specific formula for identifying the peak section of the change frequency is: ; Calculate the jump distribution density index; Among them, is the jump distribution density index within the th time window, is the index of the jump point in the current time window, is the index of the time window, is the original time index of the th jump point, is the original time index of the th jump point, is the total time length of the current time window, is the training sample weight value corresponding to the th jump point, is the normalized value of the inclination change amplitude of the th jump point, is the residual stability correction term constant used to prevent division by zero operation, is the th time window, and

[0035] Formula: ; Detailed explanation of the formula and the formula calculation derivation process: The formula is used to calculate the normalized density index of the micro amplitude jump points in the th time window. This index is used to identify the time position where the frequency peak of the elevator shaft structure is concentrated during the micro amplitude jump, and the result is used to set the starting time position of the trend prediction path; Parameter meaning and setting value: is the original time index corresponding to the th jump point, with the unit of second. The time stamp of the jump point is recorded by the Beidou GNSS high-frequency monitoring module, and the set sequence value is , that is ; is the total time span of the current window, with the unit of second, calculated by adding 1 to the difference between the maximum and minimum values of the time index, and set to ; is the inclination change amplitude of the th jump point, with the unit of degree, obtained from the differential inclination change curve, and set to degrees; is the maximum inclination change amplitude within the current time window, set to degrees; is the normalized inclination change amplitude, set to ; is the sample weight, set to , derived from the relationship between the inclination offset amplitude of each jump point and the training offset level; is the numerical stability correction term, set to ; Substitute the parameters into the formula for calculation: ; ; ; The result 0.5457 indicates that the distribution of jump points in the current time window is relatively sparse in time, while their amplitude intensity is relatively concentrated, indicating that the jump density of this paragraph is medium and tends to be a key section candidate in the predicted starting path. This value will be sorted to identify the interval index position corresponding to the minimum density index, as the starting point of the next settlement trend prediction path.

[0036] S403: According to the starting position parameters of the trend path, combined with the inclination change direction of consecutive time points within the corresponding section, predict the settlement behavior of the elevator shaft and generate settlement prediction information; According to the starting position parameters of the trend path, locate the corresponding time point , extract the inclination value sequence within subsequent consecutive time points , and perform direction trend judgment. Calculate the inclination direction difference symbol at each moment , count the lengths and directions of consecutive identical symbol segments, and determine whether the main change trend is upward or downward. Combining this directional trend, search for the behavioral characteristics of the historical settlement sequence within the same time period, and construct a mapping relationship model between the dip trend and the settlement trend. Input the trend path into the mapping logic, perform trend prediction calculations, obtain the sequence of settlement trend output values for the corresponding time period, and combine the outputs into settlement prediction information. If the direction sequence of the dip sequence is a continuously rising segment and the corresponding settlement in the mapping history shows an increasing trend, then the subsequent data in the settlement prediction sequence is output as a continuously increasing structure. Finally, generate the settlement prediction information.

[0037] Call the settlement prediction information, obtain the residual sequence between the predicted value and the actual value, analyze the prediction error at each time step, adjust the translation step size within the time window, calculate the mean error for each translation state, and by comparing the residual means of each translation step, identify the optimal translation state and the corresponding time offset distance according to the stability of the error and residual fluctuations. The steps to generate the structural prediction offset trend data are specifically as follows: S501: Call the settlement prediction information, obtain the predicted value sequence and the actual value sequence, and construct the difference sequence between the data sequences to generate the prediction error residual sequence; Call the settlement prediction information and extract the sequence of settlement predicted values output by the model within a specific time range , and at the same time retrieve the sequence of actual settlement monitoring values within the same time period from the structural health monitoring system , where represents the time point index at intervals of minutes or seconds, with the unit of millimeters. One-to-one correspondence is made between the two time-synchronized numerical sequences, and for each time point calculate the difference between the predicted value and the actual value to construct the residual sequence , and this sequence reflects the prediction error of the model for the structural settlement at different time points. For example, if the sampling frequency is set to once every 5 minutes at the bottom of the elevator shaft, and the predicted sequences for 4 consecutive time points are millimeters, and the corresponding actual monitoring sequence is millimeters, then the difference sequence is . Further store the residuals as structured data according to the time index and record it as , which is convenient for subsequent window analysis and trend identification. In actual engineering scenarios, this residual sequence will be used to judge problems such as systematic offset, trend delay, or error anomalies in model prediction. For example, continuous negative deviations within a certain time period may reflect the trend of the model's lagged prediction rather than structural mutations.

[0038] S502: Call the prediction error residual sequence, set multiple consecutive time windows, adjust the translation step size of the predicted data sequence relative to the actual monitoring data sequence, and calculate the mean residual for the corresponding time window in each translation state to generate a list of translation state error means; Call the predicted error residual sequence , set multiple sliding windows of fixed length for the entire time series, and set the length of each window to time points. Each time it slides backward by one unit time point, a new window is reconstructed to form a set of sliding windows . For each window , while keeping the actual value sequence unchanged, perform a translation operation on the predicted value sequence along the relative time axis. The translation step size ranges from time steps, indicating that the predicted value can be adjusted forward or backward by at most 3 time points. At each translation state, recalculate the residual sequence within the window, and then calculate the mean value of this translated residual to obtain the average error value under the current window and the current translation state . For example, in the window , if the translation step size is and the calculation result is , then the residual mean value is millimeters. Pair each with the corresponding to form a list of error mean mappings, which is used as the translation-error response data under the current window to support subsequent step size optimization and trend delay identification.

[0039] S503: Call the list of translation state error means, identify the optimal translation state and the corresponding time offset distance by analyzing the stability of the error and residual fluctuations of each translation step size, and mark it as the response offset amount predicted by the model to generate the structural prediction offset trend data; The specific formula for analyzing the stability of the error and residual fluctuations of each translation step size is: ; Calculate the residual fluctuation score value; Among them, is the normalized residual fluctuation score value with the translation step size number , is the arithmetic mean value of all predicted residual values within the time window with the translation step size number , is the single predicted residual value with the translation step size number and the time point index , is the number of time points within the current time window, is the maximum value among the absolute values of the residual means for all translation step sizes with the number , is the constant correction term value, Represents the current translation step number for analysis, represents the discrete time point index within the current time window, represents the traversal index number for all translation steps to be compared.

[0040] Formula: ; Detailed explanation of the formula and the derivation process of formula calculation: The formula is used to calculate the normalized residual fluctuation score value for each predicted translation step , and this score value reflects the error level and stability of the model prediction results under different time offset states. The result is used to identify the optimal predicted time offset step as the basis for model response adjustment; Meaning and setting values of parameters: is the mean value of all predicted residual values in the time window under the translation step , in millimeters, obtained by subtracting the predicted value from the measured value hour by hour and taking the average. The setting value is millimeters; is the predicted residual value at the time index under the translation step , in millimeters, from the difference sequence between the settlement monitoring data and the model predicted value. Set the current window length , is millimeters; is the number of time points within the current time window, determined by the monitoring frequency and the sliding window length. The current setting is 5; is the maximum value of the absolute mean of residuals among all translation steps, set to millimeters; is the numerical stability correction term to ensure that the denominator is non - zero. The setting value is ; Substitute the parameters into the formula for calculation: ; ; ; ; The results show that the relative error intensity of the model prediction residual level at the current translation step among all candidate states is 0.6876. The lower the value, the smaller the error and the higher the stability. This scoring value will be used to sort and compare with the scoring values of other translation states, and the translation step corresponding to the minimum value will be taken as the time correction amount for the model response offset, thereby generating the structural prediction offset trend data. This process directly serves the model output timing alignment mechanism and is the core calculation basis in the error attribution feedback link.

[0041] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more sets of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0042] It should be understood that the term “and / or” in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character “ / ” in this article generally represents an “or” relationship between the associated objects before and after, but it may also represent an “and / or” relationship, which can be specifically understood with reference to the context before and after.

[0043] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or a similar expression means any combination of these items, including any combination of single item(s) or plural item(s). For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0044] It should be understood that in various embodiments of the present invention, the magnitudes of the serial numbers of the above - mentioned processes do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0045] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0046] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above - described devices, apparatuses, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0047] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0048] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0049] In addition, in each embodiment of the present invention, each functional unit can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0050] If the above-mentioned function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0051] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for predicting the settlement of a steel structure elevator shaft based on an LSTM model, characterized in that, Including: S1: Obtain the inclination angle data of the X, Y, and Z axes. Identify the inflection point positions according to the slope change directions of the inclination angle data of each axis at consecutive time points. Compare the inclination angle change amplitudes of each axis in the same time period, analyze the distribution state of the inclination angle differences over time, and screen the effective sections of the inclination angle changes; S2: Based on the effective sections of the inclination angle changes, obtain the three-dimensional displacement data of the structure. Calculate the included angle and modulus length changes of the displacement vectors between adjacent time points, judge the consistency of the displacement directions, analyze the stability of the main axis direction, identify the time periods when the displacement directions are consistent with the inclination angle change trends, and generate the inclination angle linkage mode intervals; S3: Invoke the inclination angle linkage mode intervals, extract the corresponding inclination angle and settlement data, calculate the change amplitudes relative to the mean value of the sample library, judge the deviation degree of the data in each time period, and construct a sample weight distribution table; S4: According to the inclination angle linkage mode intervals and the sample weight distribution table, analyze the inclination angle change sequence and detect the micro amplitude jump points. According to the distribution density of the micro amplitude jump points, identify the peak sections of the change frequency, set the starting position of the trend prediction path, predict the settlement behavior of the elevator shaft, and generate the settlement prediction information.

2. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 1, wherein: The effective sections of the inclination angle changes are specifically the multi-axis inclination angle joint change area, the time series slope reversal area, and the local amplitude difference aggregation area. The inclination angle linkage mode intervals include the main axis direction continuous section, the displacement inclination angle coordination section, and the spatial displacement stable section. The sample weight distribution table includes data deviation grouping items, sample priority labels, and training stage weight coefficients. The settlement prediction information is specifically the settlement trend path, the starting position of the jump concentration, and the predicted target data sequence.

3. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 1, characterized in that: The steps of obtaining the inclination angle data of the X, Y, and Z axes, identifying the inflection point positions according to the slope change directions of the inclination angle data of each axis at consecutive time points, comparing the inclination angle change amplitudes of each axis in the same time period, analyzing the distribution state of the inclination angle differences over time, and screening the effective sections of the inclination angle changes are specifically as follows: S101: Obtain the inclination angle data of the X, Y, and Z axes. Identify the slope directions of multiple time points in each axis inclination angle sequence, extract the inflection point positions in each inclination angle curve, calculate the time intervals between adjacent inflection points in each curve and record the inflection point indexes, and generate an inclination angle change trend node group; S102: Based on the inclination angle change trend node group, call the inclination angle values of the X, Y, and Z axes in the same time period. By comparing the inclination angle change amplitudes of each axis in the same time period, analyze the distribution state of the differences in the inclination angle change amplitudes over time, and generate an inclination angle multi-axis difference change interval; S103: According to the inclination angle multi-axis difference change interval, identify the continuous change trends according to the distribution state of the differences in the inclination angle change amplitudes over time, and screen the effective sections of the inclination angle changes.

4. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 3, characterized in that: The steps of obtaining the three-dimensional displacement data of the structure based on the effective sections of the inclination angle changes, calculating the included angle and modulus length changes of the displacement vectors between adjacent time points, judging the consistency of the displacement directions, analyzing the stability of the main axis direction, identifying the time periods when the displacement directions are consistent with the inclination angle change trends, and generating the inclination angle linkage mode intervals are specifically as follows: S201: Based on the effective section of the inclination angle change, obtain the X, Y, and Z three-dimensional displacement data within the corresponding time period. According to the time series order, construct a spatial displacement vector group for each adjacent time point, calculate the included angle and modulus change value of each group of vectors, and record the direction change trend and length change amplitude corresponding to each time point to generate a spatial displacement continuous change feature group; S202: Call the spatial displacement continuous change feature group to perform consistency analysis on the change direction of the vector included angle within a continuous time period, analyze the stability of the displacement main axis within the corresponding time period, extract the direction stable section, and generate a set of main axis direction stable intervals; S203: According to the set of main axis direction stable intervals, obtain the inclination angle change trend sequence within the corresponding time period, identify the time period when the displacement direction and the inclination angle change trend are consistent, and generate an inclination angle linkage mode interval.

5. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 4, wherein: The steps of calling the inclination angle linkage mode interval, extracting the corresponding inclination angle and settlement data, calculating the change amplitude relative to the mean value of the sample library, judging the deviation degree of the data in each time period, and constructing a sample weight distribution table are specifically as follows: S301: Call the inclination angle linkage mode interval, extract the inclination angle sequence and settlement sequence within the corresponding time period, calculate the difference between the inclination angle value at each time point and the corresponding inclination angle mean value in the sample library, and calculate the difference between the settlement value and the settlement mean value in the sample library to generate an inclination angle settlement deviation amplitude group; S302: Based on the inclination angle settlement deviation amplitude group, sort the relative positions of the fluctuation amplitudes of each time period in the sample set, and generate a sample deviation level distribution label for the deviation degree of the data in each time period; S303: According to the sample deviation level distribution label, adjust the training weights of the corresponding samples according to the deviation degree, and establish a sample weight distribution table.

6. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 5, characterized in that: The steps of analyzing the inclination angle change sequence, detecting micro jump points according to the inclination angle linkage mode interval and the sample weight distribution table, identifying the peak section of the change frequency according to the distribution density of the micro jump points, setting the starting position of the trend prediction path, and predicting the settlement behavior of the elevator shaft to generate settlement prediction information are specifically as follows: S401: According to the inclination angle linkage mode interval and the sample weight distribution table, extract the inclination angle change sequence within the corresponding time period, detect micro jump points by comparing with the micro jump identification threshold, and generate micro jump point information; S402: Based on the micro jump point information, analyze the distribution density of the micro jump points, identify the peak section of the change frequency according to the position of the jump points in the time series, and set it as the starting position of the trend prediction path to generate a trend path starting position parameter; S403: According to the trend path starting position parameter, combine the inclination angle change direction of continuous time points within the corresponding section to predict the settlement behavior of the elevator shaft and generate settlement prediction information.

7. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 6, wherein: The specific formula for identifying the peak section of the change frequency is: ； Calculate the jump distribution density index; wherein, is the jump distribution density index within the th time window, is the index of the jump point in the current time window, is the index of the time window, is the original time index of the th jump point, is the original time index of the th jump point, is the total time length of the current time window, is the training sample weight value corresponding to the th jump point, is the normalized value of the inclination change amplitude of the th jump point, is the residual stability correction term constant for preventing division by zero, is the total number of micro jump points detected within the th time window.

8. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 1, wherein It also includes: S5: Invoke the settlement prediction information, obtain the residual sequence between the predicted value and the actual value, analyze the prediction error at each time step, adjust the translation step size within the time window, calculate the error mean value in each translation state, identify the optimal translation state and the corresponding time offset distance by comparing the residual means of each translation step, and generate the structural prediction offset trend data; The structural prediction offset trend data includes the model time response offset, the minimum residual value of translation matching, and the trend error classification label.

9. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 8, wherein: The steps of invoking the settlement prediction information, obtaining the residual sequence between the predicted value and the actual value, analyzing the prediction error at each time step, adjusting the translation step size within the time window, calculating the error mean value in each translation state, identifying the optimal translation state and the corresponding time offset distance by comparing the residual means of each translation step, and generating the structural prediction offset trend data are specifically as follows: S501: Invoke the settlement prediction information, obtain the predicted value sequence and the actual value sequence, and construct the difference sequence between the data sequences to generate the prediction error residual sequence; S502: Invoke the prediction error residual sequence, set multiple consecutive time windows, adjust the translation step size of the predicted data sequence relative to the actual monitoring data sequence, and calculate the residual average value of the corresponding time window in each translation state to generate the translation state error mean value list; S503: Invoke the translation state error mean value list, identify the optimal translation state and the corresponding time offset distance by analyzing the stability of the error and residual fluctuations of each translation step size, and mark it as the response offset of the model prediction to generate the structural prediction offset trend data.

10. The method for predicting the settlement of a steel structure elevator shaft based on the LSTM model according to claim 9, characterized in that: The specific formula for analyzing the stability of the error and residual fluctuations of each translation step size is: ； Calculate the residual fluctuation score value; Among them, is the normalized residual fluctuation score value with the translation step number . is the arithmetic mean of all prediction residual values within the time window with the translation step number . is the single prediction residual value with the translation step number and the time point index . is the number of time points within the current time window. is the maximum value among the absolute values of the residual means with all translation step numbers . is the constant correction term value. represents the current translation step number for analysis. represents the discrete time point index within the current time window. represents the traversal index number of all translation steps to be compared.

Images