66 results about "Reference window" patented technology

In various examples, systems, devices and methods are disclosed relating to enhancing the efficiency and reducing the delay of error resilience in multimedia communication systems. The systems, devices and methods can include a computer system determining a sliding frame window (or a sliding reference window) for a video stream. The size of the sliding frame window represents a minimum separation between reference frames and corresponding inter frames. The computer system can encode a plurality of video frames of the video stream according to the size of the sliding frame window, such that a separation between any pair of video frames of the plurality of video frames including a reference frame and a corresponding inter-frame is greater than or equal to the size of the sliding frame window.

The invention belongs to the field of radar signal processing, and particularly relates to an improved variable index constant false alarm rate detection algorithm. According to the property of the reference window before and after constant false alarm detection, the corresponding constant false alarm processing method is adaptively selected, and the detection performance of the detector in a multi-target background environment can be effectively improved. Aiming at the situation that a traditional variable index constant false alarm detector (VI-CFAR) has targets in front and back reference windows and the detection performance is poor when a detector selection unit averagely selects a small constant false alarm detector (SO-CFAR), the detector is improved by adopting an adaptive deletion selection small constant false alarm detector (ACSO-CFAR), and when a target exists in a single-side reference window, the detection performance of the detector selection unit is improved by adopting an adaptive deletion selection small constant false alarm detector (ACSO-CFAR). An improved unit average maximum constant false alarm rate detector (CA-CFAR) is selected, and compared with a traditional CA-CFAR, the detection performance is improved under the multi-target background.

The invention discloses a millimeter wave water surface flow velocity measurement radar CFAR detection method and system, and the method comprises the steps: eliminating an outlier in advance through a local outlier factor, carrying out the estimation of the quartile distance of a sample through the screened data, obtaining a cut-off threshold value of the sample, and finally obtaining a background clutter power estimation value through the maximum likelihood estimation. According to the method, the selection of the quartile optimization truncation threshold is introduced, and the reference window sample is screened in advance by using the local outlier factor, so that the robustness of the quartile and the truncation threshold is enhanced; according to the method, the environmental adaptability of the algorithm is enhanced, the target detection performance of the algorithm in a multi-target and clutter edge environment is effectively improved, and a feasible new scheme is provided for target detection of the millimeter wave water surface flow velocity measurement radar.

The present application provides a method and device for window occlusion detection, which belongs to the field of sensor technology and can be used for assisted driving and autonomous driving. The method includes: obtaining a first window echo intensity and a second window echo intensity, the first window echo intensity represents the reference window echo intensity, and the second window echo intensity represents the window echo intensity at the current moment; obtaining the detection parameters of the target point at the first moment and the detection parameters at the second moment, the detection parameters including the reflectivity and / or position coordinates of the target point; according to the first window echo intensity, the second window echo intensity, the detection parameters at the first moment and the detection parameters at the second moment, obtaining a window occlusion parameter, the window occlusion parameter is used to indicate the window occlusion state. This method improves the accuracy of lidar window occlusion detection, so as to improve the terminal's advanced driving assistance system (ADAS) capabilities in autonomous driving or assisted driving.

The invention provides an optical integrated multiband on-line detection device, which belongs to the field of laser photoelectric system detection and is arranged at a laser output end of a laser photoelectric system. Comprising a reference unit, a beam shrinking unit, a near-field detection unit, a self-calibration unit, a far-field detection unit and an adjusting unit which are arranged in sequence, wherein the adjusting unit is connected with each unit; the reference unit comprises a reference window mirror with a cross wire; the beam shrinking unit comprises a primary mirror, a secondary mirror and a focusing mechanism for moving the secondary mirror which are sequentially arranged along a light path; the near-field detection unit comprises a near-field detector and a first spectroscope; the self-calibration unit comprises a self-calibration light source, a second spectroscope and a pyramid; the first spectroscope and the second spectroscope are sequentially arranged on a light path of a light beam output by the secondary mirror; the far-field detection unit comprises a far-field detector. According to the invention, the optical axis and pupil deviation of the laser beam relative to the reference can be rapidly and accurately tested.

The invention belongs to the technical field of anomaly detection, and provides an anomaly detection method and system based on context residual learning, and the method comprises the steps: obtaining detection data of a target domain, enabling the detection data to be a time sequence, carrying out the sliding segmentation according to a given window length, selecting a plurality of context prompt windows for each to-be-detected query window, obtaining a curvature perception basic residual error by using a pre-trained time sequence anomaly detection model, and determining abnormal data based on a comparison result of the curvature perception basic residual error and a set threshold value; during training, based on the time attention representation and the frequency domain attention representation, constructing context residuals of a time domain and a frequency domain; and training by using a multi-term loss function. By capturing the relative difference mode between the query window and the corresponding normal reference window, the method has high stability in a cross-domain scene.

The invention provides an automatic marking method for a target GCP. The automatic marking method comprises the following steps: establishing a rough mesh model based on an aerial photo set of a specified area; determining a marking point of the target GCP on a reference picture; projecting the marking point to the surface of the rough mesh model and intercepting a first projection surface element; searching and determining the optimal position and the optimal orientation of the first projection surface element based on the correlation of images in a re-projection window obtained by re-projecting the first projection surface element to the plurality of evaluation pictures; and projecting an image in a reference window on the reference picture to the first projection surface element with the optimal position and the optimal orientation to obtain a second projection surface element, and searching and determining a corresponding imaging position of the target GCP on the to-be-marked picture based on an expected image obtained by re-projecting the second projection surface element to the to-be-marked picture. According to the method provided by the invention, the credibility and accuracy of automatically marking the target GCP on the to-be-marked picture are effectively improved.

The invention relates to a VIO implementation method for multi-camera asynchronous acquisition. The method comprises the following steps: acquiring timestamps of a plurality of images corresponding to a current frame; wherein the plurality of images are respectively images acquired by a plurality of cameras in a time window of the same frame; determining a timestamp of a reference window corresponding to the current frame according to the timestamps of the plurality of images; calculating the pose compensation amount between the timestamp of each image and the reference window based on the inertial measurement data and the state vector; respectively constructing an inertial measurement residual error and a visual measurement residual error corresponding to each image based on the pose compensation amount; and carrying out iterative optimization solution on the inertia measurement residual error and the vision measurement residual error based on a nonlinear optimization method. According to the scheme provided by the invention, modeling is carried out on the difference quantity of inconsistent timestamps of pictures shot by different cameras in the VIO system, so that higher precision and robustness can be realized when different cameras in the VIO system do not shoot at the same time; according to the scheme, the VIO system can detect the rotating target more precisely and more efficiently.

The present invention discloses a SAR shallow sea water depth inversion method and system based on local wave direction, which belongs to the field of synthetic aperture radar signal processing and the field of marine remote sensing application technology. The method comprises obtaining a SAR image with clear sea wave characteristics in the study area; dividing the SAR image into sub-image blocks, estimating the local wave direction of each sub-image block based on the image spectrum; calculating the reference window angle of each sub-image block based on the local wave direction, and setting the window angle range and step size of the adaptive multi-angle window based on the reference window angle; calculating each image spectrum of the image sequence divided under different window angles for each sub-image block, estimating the sea wave wavelength corresponding to each image spectrum, and performing wavelength fusion based on Kalman filtering; applying the linear dispersion relationship to the fused wavelength to invert the underwater terrain. The present invention improves the accuracy and robustness of underwater terrain detection, and provides support for the engineering application of SAR shallow sea underwater terrain detection.

The invention relates to the technical field of video processing, and discloses a video super-resolution reconstruction method and device and wearable equipment, and the method comprises the steps: obtaining a to-be-reconstructed low-resolution video; reconstructing the to-be-reconstructed low-resolution video through a preset reconstruction model to obtain a target high-resolution video; wherein the preset reconstruction model at least comprises a disparity map generation module, and the disparity map generation module is used for determining a reference window and a candidate window of the to-be-reconstructed low-resolution video, performing depth information extraction based on the reference window and the candidate window to generate a disparity map, and reconstructing the to-be-reconstructed low-resolution video according to the disparity map. Compared with the prior art that the depth information is not considered, the method can extract the depth information for reconstruction, thereby improving the reconstruction effect.

The invention relates to a two-dimensional constant false alarm threshold calculation method based on an integral image, and belongs to the field of radar technology. The method includes constructing an integral image matrix; calculating the vertex coordinate positions corresponding to the reference window and the protection window according to a given reference window and a protection window, and the element values ​​of the corresponding points of the integral image matrix can be known from the vertex coordinate positions; calculating the cumulative sum of all elements in the reference window and the protection window; calculating the total number of elements contained in the reference window and the protection window; then calculating the mean value and sum of the clutter amplitude in the zigzag window area; and calculating the CFAR threshold value according to the mean value and sum of the clutter amplitude in the zigzag window area. Compared with the traditional method, the number of calculations is independent of the given reference window and the protection window size, and the total number of calculations is less than that of the traditional two-dimensional constant false alarm threshold calculation method, which greatly reduces the algorithm time consumption, can meet the real-time requirements of the project, and has stronger applicability.

The application discloses a kind of hysteresis reference window sliding window adaptive threshold surge detection system and method, belong to aero-engine surge detection field.The method includes: real-time acquisition aero-engine pressure sensor output analog voltage signal, and carry out partial pressure, filtering and analog-digital conversion, obtain discrete sampling sequence;With sliding window intercept analysis window and carry out linear fitting, calculate its slope and residual standard deviation;Established by historical analysis window eigenvalue, with current window maintains hysteresis distance reference window;Based on the statistics of eigenvalue sequence in reference window, dynamically calculate adaptive threshold;By comparing current window eigenvalue with threshold value, and when the eigenvalue of continuous multiple windows exceeds threshold value, determine surge.The detection scheme provided by the application significantly reduces the amount of calculation through local window processing and dynamic threshold adjustment, solves the problem that existing algorithms are difficult to be deployed in real time in embedded hardware, and improves the detection accuracy and working condition adaptability.

The invention relates to the technical field of video coding, and discloses a caching and access management method and device for inter-frame prediction reference data, initial reference window data of a reference frame is stored in an on-chip cache according to columns, the on-chip cache comprises M columns of storage spaces, and M is greater than the basic column number of reference data of a reference window required by motion estimation of a coding unit; when the whole pixel search operation of the Nth coding unit is carried out, the sub-pixel search operation of the (N-1) th coding unit is carried out, and reference windows required by the whole pixel search operation and the sub-pixel search operation have overlapped columns and different starting columns; in response to the completion of the sub-pixel search operation of the (N-1) th coding unit, determining a to-be-replaced reference data column in the on-chip cache; and obtaining a replacement reference data column corresponding to the to-be-replaced reference data column from the reference frame to replace the to-be-replaced reference data column. Therefore, according to the method, the high-definition video coding performance can be ensured, the off-chip storage bandwidth pressure is not increased, and meanwhile, the on-chip storage cost required by the search window cache in the inter-frame prediction process is reduced.

An object detection device is configured to be mounted to a movable body and configured to detect an object present around the movable body, and includes: a wave transmitter configured to transmit a transmission wave corresponding to a transmission signal; a wave receiver configured to receive, as a reception wave, the transmission wave reflected by the object; a relative speed estimation part configured to estimate a relative speed between the movable body and the object; a correlation processing part configured to change, in accordance with the relative speed, a window width of a reference window that defines the number of pulses of the transmission signal to be correlated with a reception signal corresponding to the reception wave when a correlation value corresponding to similarity between the transmission signal and the reception signal is obtained; and a detection part configured to detect information on the object, in a case where the correlation value is determined to indicate similarity at a level equal to or higher than a predetermined level.

The application discloses a ground reconnaissance radar CFAR detection method and system, the method comprises the following steps: based on the minimum description length criterion, model selection and clutter edge identification are carried out on reference window samples, and a reference sample set homogeneous with a to-be-detected unit background is dynamically selected; bilateral quartile range is used for bilateral truncation on the selected reference samples, and abnormal interference samples on both sides of the reference window are removed; based on the effective samples after truncation, accurate estimation of the background clutter power is completed through maximum likelihood estimation of the truncated Weibull distribution, and an adaptive detection threshold is calculated according to a preset false alarm probability. The application has both environmental adaptability of regional decision and anti-interference robustness of truncated statistics, and can effectively balance target detection probability and false alarm rate stable control in complex clutter scenes such as uniform background, multiple targets, clutter edge and the combination of the two, and significantly improves the target detection performance and environmental adaptability of the ground reconnaissance radar in a complex working environment.

This invention discloses a Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on a large-scale selection logic, belonging to the field of image detection technology. The steps of this invention include: first, setting a detection window, which includes a detection unit, a reference window composed of a leading edge window and a trailing edge window, and a protection region; then, using the detection window to traverse the SAR ship remote sensing image pixel by pixel, with each movement distance equal to the width of the detection unit; subsequently, calculating the mean ratio (MR) of the leading edge and trailing edge windows, and using a threshold to determine whether the clutter is uniform or non-uniform; then, selecting a reference window based on the determination result; for uniform clutter, selecting the entire reference window, and for non-uniform clutter, selecting the leading edge / trailing edge window with the larger mean and widening it by 1; finally, calculating the detection statistic. R Set detection threshold T ,according to R and T The relationship between the detection unit and the target is determined. This invention maintains an approximately constant false alarm rate in non-uniform backgrounds at clutter edges, significantly outperforming the excessive increase in false alarms observed with traditional Wilcoxon technology, thus greatly improving false alarm control capabilities.

The invention relates to a target detection method based on background noise estimation self-updating. The method comprises the following steps: judging whether a to-be-detected unit and a reference window around the to-be-detected unit belong to a homogeneous region or not; when the to-be-detected unit belongs to the mask, taking the background noise of the to-be-detected unit as updated background noise; when the to-be-detected unit does not belong to the mask, background noise of the to-be-detected unit is updated; calculating local noise estimation of the reference window, fusing the local noise estimation and the updated background noise, and calculating a fused noise estimation threshold; judging whether a target is detected in the to-be-detected unit according to the amplitude of the to-be-detected unit and the fusion noise estimation threshold; according to a detection result, calculating an actual false alarm rate, and according to a difference between the actual false alarm rate and a set false alarm rate, dynamically adjusting a fusion noise estimation threshold multiplying power to obtain an adjusted threshold multiplying power; and repeating the steps to adjust the actual false alarm rate until a preset condition is met, and obtaining a detection result. According to the method, target detection under a complex clutter background is realized.

The invention provides a three-dimensional CFAR detection method based on semantic segmentation, and belongs to the field of radar target detection. Aiming at performance reduction caused by target amplitude flicker in single-frame RD image detection, a CFAR detector reference window is extended to range, Doppler and frame three dimensions so as to contain more effective samples, a reference unit is selected in combination with semantic information, and the method comprises the following steps: integrating multiple frames of RD images to form a three-dimensional data matrix, then carrying out semantic segmentation on three-dimensional radar data, and obtaining a three-dimensional data matrix; and constructing a three-dimensional semantic sign matrix. In CFAR detection, a reference window clutter unit is eliminated by means of the matrix so as to estimate background power, and finally whether a target exists or not is judged. According to the method, the problem that the detection performance is reduced due to target amplitude flicker in single-frame detection is effectively avoided, the adaptability of a detection algorithm to target fluctuation change is remarkably improved, the calculated amount is small, and engineering implementation is easy.

The present disclosure relates to an infrared moisture meter with a rapid calibration device and a calibration and automatic calibration method thereof, and belongs to the technical field of moisture detection. The infrared moisture meter with the rapid calibration device comprises a main control unit, a motor, a light source, a chopper, a plano-convex lens, a measurement probe, a punching mirror, a measurement window, a reference probe, a rapid calibration device assembly, a mirror, a reference window, a first condenser, a second condenser and a first quick connector. The rapid calibration device assembly comprises a structural shell, a ventilation hole, a metal container, a heating wire, a weight sensor and a second quick connector. The technical solution of the present disclosure improves the accuracy of equipment detection and the applicability of different materials when the equipment is operated for a long time or the material to be detected is changed. The present disclosure solves the complex problem of on-site calibration of existing infrared online moisture meters, improves the calibration efficiency of the equipment and greatly reduces the maintenance cost of the equipment.

This invention discloses a PON gateway fault detection method, system, device, and medium, relating to the field of data processing technology. The method includes: acquiring a PON gateway status dataset; acquiring a first data type and a second data type; acquiring first state data and second state data; forming a first data sequence and a second data sequence; acquiring a first sub-fluctuation index; acquiring a second sub-fluctuation index; acquiring a reference time node; acquiring a fuzzy quantity n; acquiring n consecutive acquisition time nodes that include the reference time node as a reference window; determining whether the largest second sub-fluctuation index is located within the reference window; if so, obtaining the fault probability based on the positions of the first state data, second state data, and second state data of the reference time node within the reference window; if not, obtaining a fault detection strategy based on the first state data of the reference time node and a safety threshold. This invention has the advantages of high fault detection efficiency, low false alarm rate, and low false alarm rate.

The invention discloses a fault diagnosis method and system based on a differential attention double-branch network and large language model interpretation, and relates to the technical field of industrial system intelligent operation and maintenance and multi-sensor fault diagnosis. The method comprises the following steps: constructing paired input of a to-be-diagnosed window and a reference window and carrying out differential modeling; in a differential branch, a channel cross attention mechanism modulated by channel selection probability prior is adopted, and adaptive focusing and contribution redistribution are carried out on a key sensor channel; meanwhile, a sharing representation branch is introduced to describe a working condition generality and a cross-sample stability mode, and fusion with a differential branch is carried out at a decision-making layer; diagnosis bases such as key sensors, statistical characteristics of the key sensors, channel attention weights and the like are normalized into structured evidences, and the structured evidences are input into a large language model to generate a natural language diagnosis report. According to the method, high diagnosis accuracy can be kept under complex working conditions, and stable and reviewable fault diagnosis results can be output.

The invention relates to the technical field of electrical digital data processing, and discloses an intelligent database operation and maintenance method based on a large language model, which comprises the following steps: establishing a reference window buffer queue and a detection window buffer queue in a memory, and respectively storing historical and real-time data to construct steady-state and transient probability distribution; the probability distribution divergence of the key index between the two windows is calculated, when the divergence exceeds a threshold value, local subgraph condition independence test is triggered, and the causal topology connection weight is updated; extracting an association index subset based on the updated topology, encoding the association index subset into a structured prompt vector, and inputting the structured prompt vector into a large language model to generate a candidate instruction; and executing the instruction in the isolation shadow database based on the snapshot, and generating a formal instruction if the state vector distance is smaller than a convergence threshold, the causal topology is repaired in real time by detecting data distribution drift, the problem of inference illusion caused by causal mismatch under the non-stationary load is solved, and the fault attribution accuracy and the operation safety are improved.

The present application relates to the technical field of biomedical image processing, and discloses a multi-scene video intelligent clipping method and system, comprising the following steps: obtaining time-synchronized original microscopic video stream and structured experimental operation log; for each operation event, constructing an event neighborhood reference window in the original microscopic video stream and performing spatial registration to generate a locally video segment after spatial correction; generating a spatial attention weight mask according to the operation space coordinate vector, calculating a structure disturbance response value sequence of each frame of the local video segment in combination with a structural similarity index, and obtaining a phase significance score sequence through smoothing filtering; generating an original candidate segment based on the discrete first-order derivative of the phase significance score sequence and baseline period statistical characteristics; according to the preset priority weight of the operation action type label corresponding to the candidate segment, performing cutting and transition processing on the time overlapping segment, and outputting an abstract video file; and generating a metadata index file associated with the abstract video file.

The application relates to a multi-camera asynchronous acquisition VIO implementation method, comprising the following steps: acquiring the timestamps of multiple images corresponding to a current frame; wherein the multiple images are images captured by multiple cameras within a same frame time window; determining the timestamp of a reference window corresponding to the current frame according to the timestamps of the multiple images; calculating the pose compensation amount between the timestamp of each image and the reference window based on inertial measurement data and a state vector; respectively constructing inertial measurement residuals and visual measurement residuals corresponding to each image based on the pose compensation amount; and iteratively optimizing and solving the inertial measurement residuals and the visual measurement residuals based on a nonlinear optimization method. The scheme of the application models the inconsistent difference amount of the timestamps of the pictures taken by different cameras in the VIO system, so that when the different cameras in the VIO system do not take pictures at the same time, higher precision and robustness can be achieved; the scheme can make the VIO system have higher detection precision and higher efficiency for a rotating target.

The invention provides a target constant false alarm rate detection method based on semantic segmentation, and belongs to the field of radar target detection. In a complex environment, a sky wave over-the-horizon radar RD map contains sea clutters and abnormal clutters, effective target detection cannot be performed in the clutters, and clutter edge target detection becomes difficult. In order to solve the problem, the target constant false alarm rate detection method based on semantic segmentation is provided and comprises the following steps of performing semantic segmentation on an RD image, constructing a semantic flag matrix, and dividing the RD image into a clutter region and a noise region. And then, according to the semantic flag matrix, clutter units in the reference window of the unit to be detected are eliminated, and effective reference units are determined. Calculating a background power estimation value and calculating a threshold factor based on the effective reference unit; and finally judging whether the ratio of the power value of the to-be-detected unit to the background power is greater than a threshold value, and judging whether a target exists. According to the method, the clutter edge target detection probability can be effectively improved, the calculation amount is small, and engineering implementation is easy.

The embodiment of the invention relates to the technical field of information security, in particular to an API abnormal access detection method, computer equipment and a storage medium. The method comprises the steps that an access request received by a target API in a target time window is acquired, the access request carries key information, and the key information comprises a timestamp; updating the key information to a target bitmap corresponding to the target time window based on the timestamp; obtaining a historical bitmap based on the target time sequence database; calculating a target confidence score based on the target bitmap and the historical bitmap; and if the target confidence score is greater than a preset confidence threshold, determining that the target API has abnormal access. According to the embodiment of the invention, the key information of the API access request is compressed and stored in the time sequence database through the bitmap, the required storage space is reduced, the bitmap operation efficiency is high, the target confidence score is calculated by comparing the bitmap of the target time window with the bitmap of the historical reference window, the false alarm and the missing report are reduced, and the user experience is improved. And rapid real-time detection and response to the abnormal access of the API are realized.

Provided is a method of detecting a surface defect. The method includes acquiring a target image, detecting at least one defect area from the target image, generating a first defect matrix having a size corresponding to the target image and an element that is a first defect score calculated based on the number of defect areas, calculating a plurality of second defect scores by summing first defect scores in a range corresponding to a predetermined reference window in the first defect matrix, and extracting an image corresponding to the reference window from the target image based on the second defect score and generating a defect image.

The invention relates to the technical field of radar target detection, in particular to an HFSWR target intelligent detection method based on background subspace projection filtering and a deep learning network, and the method comprises the following steps: S1, carrying out the unbiased covariance estimation and feature decomposition on a multi-channel distance-Doppler spectrum through employing a protection-reference window, and obtaining an HFSWR target; constructing a local clutter main subspace and projecting the whitening snapshot to obtain an energy tensor changing along with a subspace rank k; according to the method, array phase information is effectively utilized by adopting background matching subspace projection, the separation degree of target and non-target components is remarkably improved, projection energy of different ranks is stacked into a three-dimensional tensor according to channels, a frame of distance-Doppler spectrum is expanded in a distance-Doppler-rank three-dimensional space, and the distance-Doppler-rank three-dimensional space is expanded. And the deep network can learn the distance-Doppler spectrum structure features and the cross-channel complementary features under the multi-rank hypothesis at the same time.

This invention discloses a method and related products for detecting GNSS step-type deception interference, belonging to the field of satellite navigation signal processing technology. The GNSS step-type deception interference detection method provided by this invention constructs a signal power sample set through sliding window segmentation processing and builds a feature sequence to characterize the statistical properties of the signal power samples. It utilizes a long short-term memory network model for time-series modeling, effectively capturing the temporal dynamic characteristics of the signal power samples. By calculating and fusing the distribution difference measure between the current detection window and the historical reference window, a comprehensive distribution drift index is constructed, which can sensitively and accurately reflect the drift of the signal power sample distribution. Based on the comprehensive distribution drift index, a dynamic threshold is constructed for decision-making, enabling the detection standard to adapt to environmental changes, thereby improving the reliability and accuracy of GNSS step-type deception interference detection under dynamic environments and varying environmental noise characteristics.

The application provides a risk prediction window selection method and device based on multi-target cooperative evaluation and electronic equipment, relates to the technical field of machine learning and data modeling, and comprises the following steps: obtaining time sequence historical data of a target object, and dividing the time sequence historical data into a training subset and a test subset used for simulating future prediction according to time sequence; a candidate prediction window set containing at least two different time lengths is set, for each window length in the set, a corresponding binary classification prediction task label is defined based on the length, and a corresponding prediction model is trained by using the training subset; taking the model performance corresponding to the longest reference window in the candidate window as a reference, the optimal window length is selected from the candidate window set according to a preset selection criterion. The application ensures that the selected window length directly serves the accuracy target of the final prediction model, so that the shortest effective window is automatically found under the premise of ensuring that the prediction performance of the long window reference model is equivalent.