Adaptive filtering method, device and equipment for elevation data of land leveler and medium

Abstract

The invention discloses an adaptive filtering method, device and equipment for elevation data of a land leveler and a medium. The method includes: receiving current original elevation data of the land leveler to obtain a current trend sequence; detecting and correcting the current original data by using a matching rule according to the historical data volume to obtain the current preprocessed elevation data; performing residual analysis on the current and historical preprocessed data to obtain a noise fluctuation coefficient, and further obtaining micro-scale and mesoscale filtering results; calculating a dynamic trend prediction result according to the current trend sequence and the trend tracking parameters, and solving a deviation value between the current preprocessed data and the result; a fusion weight is calculated according to the deviation value and the noise fluctuation coefficient, and a target filtering result is obtained by fusing the dynamic trend prediction, the micro-scale result and the mesoscale result for the land leveler to adjust a scraper knife; and updating the trend sequence by using the current original and preprocessed data for the next round of filtering. According to the invention, through adaptive singular point detection, multi-scale filtering and dynamic trend fusion, precise processing of elevation data of the land leveler and adaptive control of the action of the scraper knife are realized.

Description

Technical Field

[0001] This invention relates to the field of engineering machinery technology, and in particular to an adaptive filtering method, device, equipment and medium for elevation data of a grader. Background Technology

[0002] In 3D leveling operations using graders, differential positioning technology is required to obtain accurate elevation data in order to control the blade movement and achieve site leveling. However, the elevation data from differential positioning is susceptible to multiple interferences: on the one hand, satellite signals may experience instantaneous fluctuations due to obstruction, multipath effects, etc., resulting in singularities (abrupt changes in elevation values); on the other hand, vibrations from the grader's operation and the noise from the sensors themselves can introduce random noise, and the blade lifting and lowering process also exhibits linear or nonlinear dynamic trends.

[0003] In existing technologies, traditional filtering techniques have obvious drawbacks when processing this type of elevation data: moving average filtering causes trend tracking lag due to fixed windows and is prone to distortion due to singularities; Kalman filtering relies on a precise linear model, which is difficult to adapt to complex working conditions and has poor robustness to singularities; wavelet filtering requires manual parameter setting, which is not universal enough and cannot adapt to the real-time dynamic characteristics of the data, making it difficult to balance noise suppression and elevation change response speed. Summary of the Invention

[0004] Based on this, the present invention provides an adaptive filtering method, device, equipment and medium for grader elevation data to solve the problem of low adaptability and accuracy of traditional filtering algorithms for differential elevation data of graders in complex operating scenarios, thereby further solving the problem of static instability or dynamic lag in traditional filtering, achieving static stability and dynamic delay-free operation, and ensuring blade control.

[0005] In a first aspect, embodiments of the present invention provide an adaptive filtering method for elevation data of a grader, the method comprising:

[0006] Upon receiving the current raw elevation data transmitted by the grader, the most recently updated current trend sequence is obtained. The current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data.

[0007] Based on the amount of historical data stored in the current trend sequence, singularity detection is performed on the current original elevation data using matching singularity detection rules, and the current original elevation data is corrected according to the singularity detection results to obtain the current preprocessed elevation data.

[0008] Residual analysis is performed on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient. Based on the noise fluctuation coefficient, multi-scale filtering is performed on the current preprocessed elevation data to obtain micro-scale filtering results and meso-scale filtering results.

[0009] Based on the current trend sequence and predefined trend tracking parameters, calculate the dynamic trend prediction result of the current preprocessed elevation data, and calculate the deviation between the current preprocessed elevation data and the dynamic trend prediction result;

[0010] The fusion weight parameters are calculated based on the deviation value and the noise fluctuation coefficient. Then, the dynamic trend prediction results, microscale filtering results and mesoscale filtering results are fused according to the fusion weight parameters to obtain the target filtering result. The target filtering result is then provided to the grader to adjust the grader blades in real time.

[0011] The current trend sequence is updated based on the current raw elevation data and the current preprocessed elevation data, in order to provide filtering for the new raw elevation data received in the next round.

[0012] Secondly, embodiments of the present invention provide an adaptive filtering device for grader elevation data, the device comprising:

[0013] The trend sequence acquisition module is used to acquire the most recently updated current trend sequence when it receives the current raw elevation data transmitted by the grader. The current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data.

[0014] The singularity detection module is used to perform singularity detection on the current raw elevation data according to the number of historical data stored in the current trend sequence and the matching singularity detection rules, and to correct the current raw elevation data according to the singularity detection results to obtain the current preprocessed elevation data.

[0015] The multi-scale filtering result calculation module is used to perform residual analysis on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient, and to perform multi-scale filtering on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain micro-scale filtering results and meso-scale filtering results.

[0016] The deviation value calculation module is used to calculate the dynamic trend prediction result of the current preprocessed elevation data based on the current trend sequence and predefined trend tracking parameters, and to calculate the deviation value between the current preprocessed elevation data and the dynamic trend prediction result.

[0017] The filtering result output module is used to calculate the fusion weight parameters based on the deviation value and noise fluctuation coefficient, and then fuse the dynamic trend prediction result, microscale filtering result and mesoscale filtering result according to the fusion weight parameters to obtain the target filtering result. The target filtering result is then provided to the grader to adjust the grader blade in real time.

[0018] The trend sequence update module is used to update the current trend sequence based on the current raw elevation data and the current preprocessed elevation data, so as to provide filtering processing for the new raw elevation data received in the next round.

[0019] Thirdly, embodiments of the present invention provide an electronic device, the electronic device comprising:

[0020] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform an adaptive filtering method for grader elevation data according to any embodiment of the present invention.

[0021] Fourthly, a computer-readable storage medium is also provided, the computer-readable storage medium storing computer instructions, the computer instructions being used to cause a processor to execute and implement an adaptive filtering method for grader elevation data as described in any embodiment of the present invention.

[0022] The technical solution of this invention achieves adaptive full-scene operation. By detecting singularities in different states and dynamically adjusting the filtering window and fusion weights, it accurately responds to the dynamic changes in data accumulation, noise intensity, and terrain trends, solving the problems of easy misjudgment and poor adaptability of traditional methods. It balances multi-dimensional needs, taking into account terrain detail preservation, medium-to-strong noise suppression, and macro trend tracking without manual intervention, adapting to complex working conditions. Closed-loop iterative optimization continuously updates the trend sequence, allowing the data processing accuracy to gradually improve as the operation progresses. It has strong real-time performance and practicality, with a simple and efficient algorithm that can meet the real-time control requirements of grader blades and can be directly implemented in engineering applications.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of an adaptive filtering method for grader elevation data provided in Embodiment 1 of the present invention;

[0026] Figure 2 This is a flowchart of another adaptive filtering method for grader elevation data provided in Embodiment 2 of the present invention;

[0027] Figure 3 This is a schematic diagram of the structure of an adaptive filtering device for grader elevation data provided in Embodiment 3 of the present invention;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device that implements an adaptive filtering method for grader elevation data according to an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Example 1

[0032] Figure 1This is a flowchart illustrating an adaptive filtering method for elevation data of a motor grader according to an embodiment of the present invention. This embodiment is applicable to situations where elevation data containing dynamic trends, singularities, and instantaneous interference in 3D leveling operations using a motor grader is filtered to balance noise suppression and elevation change response speed. This method can be executed by an adaptive filtering device for motor grader elevation data, which can be implemented in hardware and / or software and can be configured in a motor grader used for 3D leveling. Figure 1 As shown, the method includes:

[0033] S110. Upon receiving the current raw elevation data transmitted by the grader, obtain the most recently updated current trend sequence, wherein the current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data.

[0034] The current raw elevation data refers to the original elevation information collected in real time by the grader through elevation sensors during 3D leveling operations, without any filtering or correction processing. The current trend sequence refers to the data set stored in the filtering system that reflects the historical elevation change trend; it is an ordered record of historical operation data. The time extension order is the order from earliest to latest timestamp (older data first, newer data last), consistent with the time flow of the grader's operations. Historical data consists of each record stored in the current trend sequence, containing two core data sets: historical raw elevation data and historical preprocessed elevation data. The historical preprocessed elevation data is the preprocessed data obtained after singularity detection and correction of the historical raw elevation data. After receiving the current raw elevation data, the filtering system first retrieves the most recently updated current trend sequence. From a practical perspective, the current trend sequence needs to have a preset storage capacity (e.g., retaining the most recent 100 historical data entries) to avoid data redundancy leading to processing delays; the retrieval operation is implemented through the system's internal data interface to ensure real-time performance and adapt to the dynamic control requirements of grader operations.

[0035] S120. Based on the number of historical data stored in the current trend sequence, use the matching singularity detection rules to perform singularity detection on the current original elevation data, and correct the current original elevation data according to the singularity detection results to obtain the current preprocessed elevation data.

[0036] Singularity detection rules are used to identify outliers that deviate from the normal range of elevation changes. Singularity detection refers to the process of determining whether a value is an outlier (singularity) by calculating indicators such as the degree of deviation between the current original elevation data and historical trends, and the data mutation rate, based on matching singularity detection rules.

[0037] S130. Perform residual analysis on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient. Then, perform multi-scale filtering on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain the micro-scale filtering result and the meso-scale filtering result.

[0038] Residual analysis is an analytical method that quantifies the noise intensity of current preprocessed elevation data by calculating the statistical difference between the current preprocessed elevation data and historical preprocessed elevation data. A larger residual indicates more severe noise interference in the current data. The noise fluctuation coefficient is a quantitative indicator obtained based on residual analysis. Multi-scale filtering refers to a filtering method that uses different window lengths for smoothing at different noise intensities. By dynamically adapting the window length, it balances noise suppression with the preservation of terrain details, resulting in both micro-scale and meso-scale filtering results. Micro-scale filtering results refer to those obtained using a small window for moving average filtering; meso-scale filtering results refer to those obtained using a medium window for moving average filtering.

[0039] Optionally, residual analysis can be performed on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend series to obtain the noise fluctuation coefficient, which may include:

[0040] Extract all historical preprocessed elevation data from the historical data stored in the current trend sequence to form a historical preprocessed elevation data sequence arranged in chronological order.

[0041] Based on the size of the preset basic sliding window, the window length is divided for each historical preprocessed elevation data in the historical preprocessed elevation data sequence as the window center.

[0042] For each historical preprocessed elevation data, the arithmetic mean of the historical preprocessed elevation data falling within the window length is calculated and used as the historical smoothed value corresponding to the current window center. Finally, a historical smoothed value sequence corresponding one-to-one with the historical preprocessed elevation data sequence is obtained.

[0043] The difference between each historical preprocessed elevation data and its corresponding historical smoothed value is calculated to obtain a single historical residual. All the single historical residuals are combined to form a complete historical residual sequence, and the overall historical noise standard deviation of the historical residual sequence is calculated based on the sample standard deviation formula.

[0044] In the historical preprocessed elevation data sequence, the last target historical preprocessed elevation data is selected as the center of the reference window, and a corresponding number of target adjacent historical preprocessed elevation data are selected based on the size of the preset basic sliding window.

[0045] Calculate the arithmetic mean of the corresponding number of target adjacent historical preprocessed elevation data, target historical preprocessed elevation data and current preprocessed data, and use it as a smoothed reference value corresponding to the center of the reference window;

[0046] The difference between the current preprocessed elevation data and the smoothed reference value is calculated to obtain the reference residual. The reference residual is then normalized by combining it with the overall historical noise standard deviation to obtain the noise fluctuation coefficient of the current preprocessed elevation data.

[0047] The historical preprocessed elevation data sequence refers to a one-dimensional data array formed by arranging all historical preprocessed elevation data extracted from the current trend sequence in chronological order. It serves as the historical sample basis for subsequent residual analysis. The preset basic sliding window refers to a predefined fixed-length window used to extract local data. Its length is set according to the sampling frequency of the grader operation to ensure the window covers a reasonable time span. The window center refers to each historical data point in the historical preprocessed elevation data sequence, serving as the core reference point around which the window expands. The window length refers to the total number of data points included in the sliding window, specifically divided into "window center + n adjacent data points before and after". The historical smoothing value refers to the arithmetic mean of all historical preprocessed elevation data within a window centered on a given historical data point, used to characterize the local average trend at that window center. The historical smoothing value sequence is a sequence formed by arranging the historical smoothing values ​​corresponding to each historical data point in their original chronological order. It corresponds one-to-one with the historical preprocessed elevation data sequence and serves as a benchmark for measuring the degree to which historical data deviates from the local trend. A single historical residual refers to the difference between a given historical preprocessed elevation data point and its corresponding historical smoothed value, reflecting the degree to which the historical data deviates from its local trend. The historical residual sequence is a sequence formed by arranging all individual historical residuals in chronological order, comprehensively recording the noise distribution characteristics of the historical data. The overall historical noise standard deviation is a statistic calculated based on the sample standard deviation formula, used to quantify the overall noise level of the historical data. The target historical preprocessed elevation data refers to the last data point in the historical preprocessed elevation data sequence, i.e., the most recent historical preprocessed elevation data in time (closest to the timestamp of the current preprocessed elevation data). The reference window center is a sliding window reference point centered on the target historical preprocessed elevation data, used to construct a local reference window closely aligned with the current data. The target adjacent historical preprocessed elevation data refers to the adjacent historical data before and after the target historical preprocessed elevation data within the reference window. The smoothed reference value is the arithmetic mean of the "target adjacent historical preprocessed elevation data, target historical preprocessed elevation data, and current preprocessed data" within the reference window, used to characterize the local trend benchmark of the current preprocessed elevation data. The reference residual refers to the difference between the current preprocessed elevation data and the smoothed reference value, reflecting the degree to which the current data deviates from its local trend. Normalization involves dividing the reference residual by the overall historical noise standard deviation to eliminate the influence of the absolute magnitude of noise under different scenarios, obtaining a dimensionless relative noise index. In this embodiment, the process uses historical data as a benchmark and current data as the core, employing a closed-loop logic of "time-series processing → local trend extraction → historical noise quantification → current benchmark construction → normalization output" to accurately quantify the noise intensity of the current preprocessed elevation data.

[0048] S140. Based on the current trend sequence and predefined trend tracking parameters, calculate the dynamic trend prediction result of the current preprocessed elevation data, and calculate the deviation between the current preprocessed elevation data and the dynamic trend prediction result.

[0049] Trend tracking parameters are a predefined set of core parameters that control the sensitivity of trend prediction, determining the degree to which the prediction result follows historical trends. Dynamic trend prediction results refer to the theoretical elevation trend value at the current moment, calculated using a trend prediction algorithm based on historical preprocessed elevation data in the current trend sequence. The deviation value is the absolute difference between the current preprocessed elevation data and the dynamic trend prediction result; it is a core indicator for measuring the degree of agreement between the current data and historical trends. A small deviation value indicates that the current data conforms to historical trends, while a large deviation value suggests the possibility of abrupt terrain changes or incompletely eliminated anomalies.

[0050] S150. Calculate the fusion weight parameters based on the deviation value and noise fluctuation coefficient, and then fuse the dynamic trend prediction results, microscale filtering results and mesoscale filtering results according to the fusion weight parameters to obtain the target filtering result. Provide the target filtering result to the grader so as to adjust the grader blade in real time.

[0051] The fusion weight parameter consists of a set of three independent weights (corresponding to the dynamic trend prediction result, microscale filtering result, and mesoscale filtering result, respectively), whose sum is always 1. Its value is dynamically adjusted based on the deviation value and noise fluctuation coefficient. Its core function is to allocate the proportional weights of the three results in the final fusion. The target filtering result refers to the final output data obtained by weighting and summing the three input results according to the fusion weight parameter. It is the optimal solution that balances trend tracking, noise suppression, and detail preservation, and can be directly used for blade control.

[0052] Furthermore, the fusion weight parameters calculated based on the deviation value and the noise fluctuation coefficient may include:

[0053] In the multi-scale filtering configuration, query the predefined basic ratio, and assign an initial first weight to the microscale filtering result, an initial second weight to the mesoscale filtering result, and an initial third weight to the dynamic trend prediction result according to the basic ratio, with a total of 1.

[0054] In the multi-scale filtering configuration, a preset trend matching threshold is queried. If the deviation value is not greater than the preset trend matching threshold, the increase of the initial third weight is determined according to the predefined deviation adjustment ratio corresponding to the dynamic trend prediction result. The increase of the initial third weight is then distributed as the deduction of the initial first weight and the deduction of the initial second weight according to the base ratio, so as to obtain the first weight, the second weight, and the third weight adjusted for the first time.

[0055] If the deviation value is greater than the preset trend matching threshold, the reduction amount of the initial third weight is determined according to the deviation adjustment ratio, and the reduction amount is distributed as the increase amount of the initial first weight and the increase amount of the initial second weight according to the base ratio, so as to obtain the first weight, the second weight and the third weight of the first adjustment.

[0056] In the multi-scale filtering configuration, a preset noise level threshold is queried, and the noise fluctuation level is determined by comparing the noise fluctuation coefficient with the noise level threshold.

[0057] If it is a high-level noise fluctuation, the increase of the second weight of the first adjustment is determined according to the predefined noise adjustment ratio corresponding to the mesoscale filtering result. At the same time, the increase of the second weight of the first adjustment is used as the deduction of the first weight of the first adjustment, so as to obtain the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment.

[0058] If it is a low-level noise fluctuation, the increase of the first weight of the first adjustment is determined according to the noise adjustment ratio corresponding to the microscale filtering result. At the same time, the increase of the first weight of the first adjustment is used as the deduction of the second weight of the first adjustment, so as to obtain the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment.

[0059] The attention weight algorithm is used to normalize the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment to obtain the fusion weight parameters corresponding to the microscale filtering result, the mesoscale filtering result, and the dynamic trend prediction result, respectively.

[0060] Multi-scale filtering configuration refers to the set of parameter configurations stored in the filtering system, including core parameters such as base ratio, trend matching threshold, deviation adjustment ratio, noise level threshold, and noise adjustment ratio. All parameters are predefined after being calibrated through experiments in typical operating scenarios and support subsequent dynamic calls. The base ratio is a predefined initial weight allocation ratio (e.g., 30% for microscale filtering results, 40% for mesoscale filtering results, and 30% for dynamic trend prediction results), which is the benchmark for weight adjustment, and the sum is always 1. The initial first / second / third weights refer to the initial ratio values ​​allocated to the microscale filtering results, mesoscale filtering results, and dynamic trend prediction results according to the base ratio. The preset trend matching threshold is a predefined critical value for judging the degree of matching between the current data and the historical trend. If the deviation value does not exceed this threshold, it means that the current data conforms to the historical trend and trend following needs to be strengthened; if it exceeds, it means that there may be a trend change and trend dependence needs to be weakened. The deviation adjustment ratio refers to the predefined adjustment magnitude for the initial third weight (dynamic trend prediction result), that is, the increase in the initial third weight = initial third weight × deviation adjustment ratio. The first / second / third weights in the initial adjustment are weight values ​​obtained after adjusting for deviation values. The core change is to increase the weight of the dynamic trend prediction result, while distributing the deduction according to the basic ratio to ensure that the total after adjustment is still 1. The preset noise level threshold refers to the predefined critical value for classifying noise intensity, set based on the statistical distribution of the noise fluctuation coefficient. The noise fluctuation level is a noise intensity category classified according to the comparison result of the noise fluctuation coefficient and the noise level threshold to ensure real-time performance and adapt to the dynamic control requirements of grader operation. The noise adjustment ratio refers to the predefined adjustment range of the weights after the initial adjustment. For high-level noise scenarios, the adjustment target is the second weight (mesoscale filtering result) of the initial adjustment. The first / second weights in the second adjustment are the adjusted weight values ​​triggered by high-level noise fluctuations. The core change is to increase the weight of the mesoscale filtering result and decrease the weight of the microscale filtering result. The third weight (dynamic trend prediction result) of the initial adjustment remains unchanged because trend adaptation has been completed through the initial adjustment. In this embodiment, the attention weight algorithm specifically refers to a simple statistical algorithm used for weight normalization. Its core function is to recalibrate the three weights after secondary adjustment to ensure that the sum is always 1, avoiding weight overflow due to the cumulative error of the two adjustments. The fusion weight parameter refers to the final set of weights output after normalization, containing three proportional values ​​that correspond one-to-one with the microscale filtering result, the mesoscale filtering result, and the dynamic trend prediction result. It serves as the direct basis for subsequent weighted fusion. This embodiment achieves dynamic generation of the fusion weight parameter through a four-stage logic of "initial benchmark setting → trend adaptation adjustment → noise adaptation adjustment → normalization calibration".

[0061] Optionally, according to the fusion weight parameters, the dynamic trend prediction results, microscale filtering results, and mesoscale filtering results are fused to obtain the target filtering result. This target filtering result is then provided to the grader to adjust the grader blades' movements in real time. This may include:

[0062] The fusion weight corresponding to the microscale filtering result is defined as the first fusion weight, the fusion weight corresponding to the mesoscale filtering result is defined as the second fusion weight, and the fusion weight corresponding to the dynamic trend prediction result is defined as the third fusion weight.

[0063] The product of the microscale filtering result and the first fusion weight, the product of the mesoscale filtering result and the second fusion weight, and the product of the dynamic trend prediction result and the third fusion weight are calculated separately. The target filtering result is obtained by adding the results of the three products.

[0064] The target filtering result is encapsulated into a height adjustment command that conforms to the data format of the grader control system through a preset communication interface. This command is used to drive the grader blade to adjust its movements. The system also receives the movement adjustment completion signal synchronously fed back by the grader control system, thus forming a closed-loop control.

[0065] The first fusion weight refers to the proportion of the fusion weight parameters specifically corresponding to the micro-scale filtering results. Its core function is to allocate the weight ratio for terrain detail preservation, and it is completely consistent with the first weight in the second adjustment. The second fusion weight refers to the proportion of the fusion weight parameters specifically corresponding to the mesoscale filtering results. Its core function is to allocate the weight ratio for noise suppression, and it is completely consistent with the second weight in the second adjustment. The third fusion weight refers to the proportion of the fusion weight parameters specifically corresponding to the dynamic trend prediction results. Its core function is to allocate the weight ratio for trend following, and it is completely consistent with the third weight in the first adjustment. The preset communication interface refers to the pre-agreed signal transmission interface between the filtering system and the grader control system, which needs to be adapted to industrial control scenarios; the elevation adjustment command refers to the executable command formed by encapsulating the target filtering result; closed-loop control refers to the complete link of "filtering system sending command → grader executing → feedback of execution result → filtering system confirmation". This embodiment completes the final implementation of "data processing → action execution" through a three-order logic of "explicit mapping → weighted fusion → reliable transmission + closed-loop control". The entire process is the final implementation of the technical steps mentioned above, such as residual analysis and weight adjustment. It transforms the abstract filtering algorithm into the scraping action that the grader can execute, directly serving the goal of improving the accuracy of 3D leveling operations.

[0066] S160. Update the current trend sequence based on the current raw elevation data and the current preprocessed elevation data, so as to provide filtering processing for the new raw elevation data received in the next round.

[0067] The current raw elevation data and preprocessed elevation data are added to the current trend sequence in chronological order (added to the end of the sequence). Simultaneously, the oldest historical data exceeding the preset storage capacity is deleted to prevent the sequence from becoming too long and reducing processing efficiency. The updated current trend sequence will serve as a historical reference for the next round of processing new raw elevation data. That is, the next round of singularity detection rule matching, residual analysis, and trend prediction will all be based on this updated sequence. Through closed-loop updates, the adaptive capability of the entire filtering method is continuously strengthened. As the operation progresses and historical data becomes richer, the accuracy of singularity detection, the accuracy of trend prediction, and the rationality of weight adjustments will gradually improve, ultimately achieving dynamic optimization of the grader blade control accuracy.

[0068] This invention achieves adaptive full-scene operation by detecting singularities in different states, dynamically adjusting the filtering window and fusion weights, and accurately responding to dynamic changes in data accumulation stages, noise intensity, and terrain trends. This solves the problems of misjudgment and poor adaptability of traditional methods. It balances multi-dimensional needs, achieving terrain detail preservation, medium-to-strong noise suppression, and macro-trend tracking without manual intervention, adapting to complex working conditions. Closed-loop iterative optimization continuously updates the trend sequence, allowing data processing accuracy to gradually improve as the operation progresses. It has strong real-time performance and practicality, with a simple and efficient algorithm that meets the real-time control requirements of grader blades and can be directly implemented in engineering applications.

[0069] Example 2

[0070] Figure 2 This is a flowchart of another adaptive filtering method for grader elevation data provided in Embodiment 2 of the present invention. This embodiment is a refinement based on the above embodiment, and correspondingly, as shown below. Figure 2 As shown, the method specifically includes:

[0071] S210. Upon receiving the current raw elevation data transmitted by the grader, obtain the most recently updated current trend sequence, wherein the current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data.

[0072] S220. If the number of historical data stored in the current trend sequence is less than the preset full window number threshold, calculate the absolute value of the first difference between the current original elevation data and the arithmetic mean of each historical preprocessed elevation data; if the absolute value of the first difference exceeds the preset fixed deviation threshold, detect the current original elevation data as a singularity in a non-full window state.

[0073] The preset full-window quantity threshold is a predefined critical value for judging whether historical data is "sufficient." It is set based on the sampling frequency and trend stability requirements of the grader elevation data. When the data volume reaches this value, a reliable trend can be fitted; otherwise, it is considered "insufficient data," requiring a simplified detection logic. The absolute value of the first difference is the absolute difference between the current raw elevation data and the arithmetic mean of all historical preprocessed elevation data, used to measure the deviation of the current data from the historical average level. The preset fixed deviation threshold is a fixed critical value set for the "non-full-window state" (insufficient data), which does not fluctuate with historical data, ensuring a consistent detection standard and avoiding threshold distortion caused by insufficient data. A singularity in the non-full-window state is the current raw elevation data in the non-full-window state where the absolute value of the first difference exceeds the fixed deviation threshold; essentially, it is an outlier deviating from the historical average level. The core logic of this embodiment is: when data is insufficient, singularities are detected based on the historical average level. When the amount of historical data is less than the full-window quantity threshold, a stable trend cannot be fitted; in this case, the arithmetic mean of all historical preprocessed elevation data is used as the reference benchmark.

[0074] S230. If the number of historical data stored in the current trend sequence is equal to the preset full window number threshold, then the historical preprocessed elevation data are fitted in the order of time extension to obtain a linear trend equation, and the trend calculation value of the current original elevation data is predicted based on the linear trend equation; if the absolute value of the second difference between the trend calculation value and the current original elevation data exceeds the preset floating deviation threshold, then the current original elevation data is detected as a singularity in the full window state.

[0075] The linear trend equation is a straight line obtained by fitting all historical preprocessed elevation data using a linear regression algorithm, accurately reflecting the stable trend of historical data. The calculated trend value is the theoretical elevation value obtained by substituting the timestamp of the current raw elevation data into the linear trend equation, representing the reasonable elevation range based on historical trends at the current moment. The absolute value of the second difference is the absolute difference between the current raw elevation data and the calculated trend value, used to measure the degree of deviation of the current data from the historical stable trend. The preset floating deviation threshold refers to the dynamic critical value set for the "full window state". The singularity in the full window state refers to the current raw elevation data in the full window state where the absolute value of the second difference exceeds the floating deviation threshold, which is essentially an outlier value that deviates from the historical stable trend. The core logic of this implementation is: when there is sufficient data, the historical stable trend is used as a benchmark to detect singularities. When the amount of historical data reaches the full window number threshold, a stable trend has been formed, and the linear trend equation can accurately capture this trend, and the calculated trend value is the theoretically reasonable value at the current moment. The core difference between it and the S230 is that the full-window state uses trend fitting instead of simple averaging, and the reference benchmark is upgraded from static averaging to dynamic trend, resulting in higher detection accuracy.

[0076] S240. If the number of historical data stored in the current trend sequence is zero, then the current raw elevation data is detected as a non-singularity in a window state.

[0077] In the "empty window" state, "non-singularity" refers to a situation where there is no historical data for reference. The current raw elevation data is assumed to be normal (non-singularity). The core reason is the lack of historical benchmarks, making it impossible to determine anomalies and preventing the unfounded removal of initial data. The empty window state only occurs during the system's first run or after the trend sequence has been cleared. Once the first historical data is generated, subsequent runs will enter either the "non-full window" or "full window" state. Therefore, this logic only applies to the initial scenario and does not affect subsequent detection accuracy.

[0078] S250. Correct the current original elevation data according to the singularity detection results to obtain the current preprocessed elevation data.

[0079] Optionally, the original elevation data can be corrected based on the singularity detection results to obtain the current preprocessed elevation data, which may include:

[0080] If the detection result is a singularity in a non-full-window state, then it is based on the preset benchmark elevation value. and raw elevation data Calculation formula based on non-full window trend forecast value The non-full window trend prediction value was calculated. ; wherein, the The number of historical data stored in the current trend sequence;

[0081] The normalized elevation data is obtained by normalizing the original elevation data, denoted as . ;

[0082] Extract the non-full window trend prediction value Normalized elevation data values Substitute into the formula for calculating non-full-window singularity correction data: In the process, preprocessed elevation data is obtained; among them, , These are the preset fixed weight values ​​under the non-full-window singularity detection rule.

[0083] Benchmark elevation value This refers to a predefined reference elevation value, typically set based on the target elevation for grader operations or the average value of historical stable elevation ranges, serving as a benchmark anchor point for trend prediction in non-full-window scenarios. (Original elevation data) This refers to the original elevation data currently detected as a "singularity in a non-full-window state," which is the input data to be corrected. When the amount of historical data does not reach the full-window threshold, a stable trend cannot be fitted; therefore, a baseline elevation value is required. Original elevation data Combined with historical data quantity To construct trend forecast values. In the formula... It is a weighting factor based on the amount of historical data. The smaller the factor, the better ( The more controllable the impact of normalized elevation data, the better. It is the original elevation data The dimensionless values ​​obtained through normalization calculations are used to eliminate the influence of absolute elevation differences on the correction results, making subsequent weights... , Its effect is more stable. Fixed weights , Weighting parameters are used to calibrate classic non-full-window scenario experiments, balancing the need for trend rationality and the preservation of original data features. Through weighted summation, the trend prediction value is fused with the normalized original data to generate preprocessed data that corrects singularity anomalies while retaining reasonable information. This embodiment addresses singularities in non-full-window states, achieving accurate correction of anomalous data through the logic of "trend prediction → data normalization → weighted fusion." Its core value lies in: preventing singularity data from directly entering subsequent stages and causing distortion when historical data is insufficient; and generating scenario-appropriate preprocessed data through trend correlation and scale unification, providing reliable input for subsequent residual analysis and filtering, while ensuring the continuity of system operations in the early stages of data accumulation.

[0084] Furthermore, the current original elevation data is corrected based on the singularity detection results to obtain the current preprocessed elevation data, which may also include:

[0085] If the detection result is a singularity in a full-window state, then the trend intercept is obtained by calling the fitted linear trend equation. and trend slope The formula for calculating the full-window trend forecast value is as follows: The full-window trend prediction value was calculated. ;

[0086] The slope of the trend Substitute into the trend weight calculation formula: In the process, the trend weights are obtained. ;

[0087] The normalized elevation data is obtained by normalizing the original elevation data, denoted as . ;

[0088] Extract the full-window trend prediction value Trend weight and normalized elevation data values Substitute into the full-window singularity correction data calculation formula: In this process, preprocessed elevation data is obtained.

[0089] With sufficient historical data in a full-window state, a linear trend equation is used to generate a reasonable elevation value for the current moment by substituting the current timestamp into the trend equation. This value is a "benchmark reasonable value" based on historical trends and is used to correct for deviations and singularities, ensuring that the corrected data conforms to the continuity of historical trends. The dynamic trend weight is obtained by multiplying the absolute value of the trend slope by a coefficient of 0.1 and then taking the minimum value not exceeding 1.0. The larger the value, the more significant the historical trend change, and the more the correction relies on the trend forecast; conversely, a smaller value focuses more on the local characteristics of the original data. Trend Weight The trend weighting is a key parameter used to balance the proportion of trend forecasts and normalized raw data in the correction process, and is crucial for adapting to different trend strengths in full-window scenarios. The predicted trend values ​​and the normalized raw data are weighted and summed to obtain the corrected current preprocessed elevation data. In this embodiment, the singularity correction under full-window conditions achieves accurate correction of trend intensity adaptively through the logic of "trend prediction → dynamic weight adaptation → normalization fusion". The final output preprocessed elevation data not only conforms to the historical stable trend but also adapts to the true characteristics of the current terrain, providing a high-precision input foundation for subsequent residual analysis, multi-scale filtering, and other processes.

[0090] S260. Perform residual analysis on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient. Then, perform multi-scale filtering on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain the micro-scale filtering result and the meso-scale filtering result.

[0091] Optionally, multi-scale filtering can be performed on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain micro-scale filtering results and meso-scale filtering results, which may include:

[0092] Based on the noise fluctuation coefficient, the first window length required for microscale filtering is determined according to the predefined noise fluctuation coefficient and filter intensity matching rules in the multi-scale filtering configuration.

[0093] If the current trend sequence is in a non-full window state or an empty window state, then based on the fluctuation range of historical preprocessed elevation data, the preset proportion value defined in the multi-scale filtering configuration is used as the trend influence coefficient.

[0094] If the current trend sequence is in a full window state, read the predefined mapping relationship between the trend slope and the trend influence coefficient in the multi-scale filtering configuration, and calculate the trend influence coefficient based on the mapping relationship;

[0095] Extract the first historical preprocessed elevation data within the first window length according to the reverse time extension order, assign the first type of weight to each first historical preprocessed elevation data according to the predefined weight allocation rules in the multi-scale filtering configuration, and calculate the weighted sum of the first historical preprocessed elevation data.

[0096] Obtain the current working status of the grader. If the grader is stationary, determine the second type of weight of the current preprocessed elevation data as the ratio of the square of the first window length to the reference elevation value.

[0097] If the grader is in motion, the second type of weight of the current preprocessed elevation data is determined as the product of the trend influence coefficient and the length of the first window;

[0098] The microscale filtering result is calculated based on the weighted sum of the first historical preprocessed elevation data and the second type of weight of the current preprocessed elevation data.

[0099] The first window length refers to the number of data points contained in the window used to extract historical data during microscale filtering. It is a core parameter determining the degree of detail retention in microscale filtering. The smaller the window, the more details are retained, but the weaker the noise resistance; the larger the window, the slightly stronger the smoothness, but the potential loss of subtle terrain features. The noise fluctuation coefficient and filter intensity matching rule is a predefined mapping relationship in the multiscale filtering configuration, the core of which is to make the window size fit the current noise intensity. The trend influence coefficient is a parameter that quantifies the degree of influence of historical trends on the current microscale filtering result. The larger the coefficient, the closer the filtering result is to the historical trend; conversely, it relies more on the local features of the current data and neighboring historical data. The preset ratio value refers to the predefined fixed trend influence coefficient in the non-full window / empty window state. Because there is insufficient historical data to fit a reliable trend, a fixed value is used to balance the trend and local features to avoid unfounded trend skew. The mapping relationship between the trend slope and the trend influence coefficient is a predefined "absolute value of trend slope → coefficient" correspondence rule in the multiscale filtering configuration in the full window state. The larger the slope (the more drastic the trend change), the larger the coefficient, strengthening the trend's guidance on filtering. The reverse time extension order is the opposite of the time extension order, meaning historical data is extracted in "new → old" order, ensuring that the most recent data within the window has a greater impact on the filtering results. The first historical preprocessed elevation data refers to the historical preprocessed elevation data extracted from the current trend series in reverse time order that falls within the length of the first window. Reverse time order extraction ensures the window focuses on the most recent history; the first type of weight decays over time to avoid excessive interference from earlier historical data. The second type of weight refers to the weight assigned to the current preprocessed elevation data, determining the proportion of the current data in the microscale filtering results.

[0100] The current working status of the grader refers to whether it is stationary (operation paused, blade position stable) or moving (normal operation, blade dynamically adjusted according to terrain), which is fed back in real time by the grader control system. When stationary, the grader is not moving, the blade position is stable, and the current pre-processed elevation data is minimally affected by motion; therefore, the second type of weight is set as the ratio of the square of the first window length to the baseline elevation value. When moving, the grader is traveling, the terrain dynamically changes, and the current data needs to be combined with historical trends; therefore, the second type of weight is set as the product of the trend influence coefficient and the first window length, ensuring that the weight of the current data is dynamically adjusted according to the trend intensity. The calculation logic of the microscale filtering result is: dynamic fusion of the historical weighted sum and the current data, usually calculated using a weighted average formula. The calculation of microscale filtering results is achieved through a four-order logic of "window adaptation to noise → coefficient correlation with trend → weight highlighting of recent trends → state adjustment of current proportion". This realizes adaptive filtering that prioritizes detail preservation while taking into account noise and trends. The final output microscale filtering results can accurately capture subtle changes in terrain while avoiding interference from noise and trend deviations, providing core inputs of detailed dimensions for subsequent multi-result fusion.

[0101] Furthermore, multi-scale filtering of the current preprocessed elevation data is performed based on the noise fluctuation coefficient to obtain micro-scale filtering results and meso-scale filtering results, which may also include:

[0102] The second window length required for mesoscale filtering is determined based on the first window length and the multi-scale filtering configuration; wherein, the multi-scale filtering configuration defines a multiple relationship between the second window length and the first window length;

[0103] If the current trend sequence is not in a full window state, the second historical preprocessed elevation data within the second window length is extracted in reverse time extension order, and the linear fitting value of the second historical preprocessed elevation data is used as the mesoscale trend prediction value.

[0104] If the current trend sequence is in a full window state, the acquisition time of the current original elevation data is obtained, the acquisition time is substituted into the fitted linear trend equation, and the mesoscale trend prediction value is obtained by combining the second window length.

[0105] If the current trend sequence is in a blank state, the benchmark elevation value is directly used as the mesoscale trend prediction value.

[0106] Obtain the current working status of the grader. If the grader is in motion, the microscale filtering result and the mesoscale trend prediction value are weighted and fused according to a preset first fusion ratio to obtain the mesoscale filtering result.

[0107] If the grader is stationary, compare the absolute value of the third difference with the preset mesoscale jump threshold.

[0108] If the absolute value of the third difference exceeds the mesoscale jump threshold, the microscale filtering result and the mesoscale trend prediction value are weighted and fused according to a preset second fusion ratio to obtain the mesoscale filtering result.

[0109] If the absolute value of the third difference does not exceed the preset mesoscale jump threshold, the mesoscale trend prediction value is directly used as the mesoscale filtering result.

[0110] The second window length refers to the number of data points contained in the window used to extract historical data during mesoscale filtering. Its core function is to enhance noise suppression capabilities through a larger window range while preserving mesoscale variations in terrain. The multiplier relationship is a predefined "second window length = first window length × N" (N is a fixed multiplier) in the multiscale filtering configuration, ensuring that the mesoscale window is always larger than the microscale window. The mesoscale trend prediction value is the predicted mesoscale trend elevation value at the current moment based on historical data within the second window or a preset benchmark, and is the core trend benchmark for mesoscale filtering. The second historical preprocessed elevation data refers to the historical preprocessed elevation data extracted from the current trend sequence in reverse time extension order that falls within the second window length. The linear fit value is the trend value obtained by linear regression of the second historical preprocessed elevation data, used to estimate the mesoscale trend under non-full window conditions. The third absolute difference value refers to the absolute difference between the current preprocessed elevation data and the mesoscale trend prediction value, used to measure the degree of deviation between the current data and the mesoscale trend. The first fusion ratio is the ratio of the predefined microscale filtering result to the mesoscale trend prediction value under motion conditions. The mesoscale jump threshold is a predefined critical value under static conditions, used to determine whether the deviation of the current data from the mesoscale trend constitutes an anomalous jump. The second fusion ratio is the fusion ratio when the absolute value of the third difference exceeds the threshold under static conditions, balancing the current data details with the mesoscale trend to prevent anomalous jumps from directly affecting the results.

[0111] In motion, mesoscale filtering prioritizes noise suppression while following the overall trend. A first fusion ratio is used, allowing the mesoscale trend prediction to dominate, with microscale results supplementing details. In a stationary state, with the grader still and the terrain stable, it's necessary to determine if the absolute value of the third difference represents an anomalous jump: if the difference is less than or equal to the mesoscale jump threshold, the deviation is normal detail, and the mesoscale trend prediction is used directly as the result to avoid over-smoothing and losing stable details; if the difference is greater than the threshold, there may be an anomalous jump, and a second fusion ratio is used to balance the two. The calculation of the mesoscale filtering result utilizes a three-order logic of "window scale binding → trend benchmark adaptation → stateful fusion," achieving adaptive filtering that prioritizes noise suppression while considering both mesoscale trends and details. This result complements the microscale filtering result in a "details-mesoscale trend" relationship, providing the core input for subsequent multi-result fusion in terms of noise suppression.

[0112] S270. Based on the current trend sequence and predefined trend tracking parameters, calculate the dynamic trend prediction result of the current preprocessed elevation data, and calculate the deviation between the current preprocessed elevation data and the dynamic trend prediction result.

[0113] Optionally, based on the current trend sequence and predefined trend tracking parameters, the dynamic trend prediction result of the current preprocessed elevation data is calculated, and the deviation between the current preprocessed elevation data and the dynamic trend prediction result is calculated. This may include:

[0114] Obtain the window state of the current trend sequence. If the window state is empty, then use the benchmark elevation value as the dynamic trend prediction result.

[0115] If the window state is not full, the weighted average of all historical preprocessed elevation data in the current trend sequence is calculated based on the predefined historical weighting factor, and then multiplied by the predefined trend smoothing coefficient to obtain the dynamic trend prediction result.

[0116] If the window is full, the trend slope is smoothed by combining the fitted linear trend equation with a predefined trend smoothing coefficient to obtain a corrected linear trend equation. The acquisition time of the current original elevation data is substituted into the corrected linear trend equation to obtain the dynamic trend prediction result.

[0117] Calculate the absolute difference between the current preprocessed elevation data and the dynamic trend prediction result, and use the absolute difference as the deviation value.

[0118] In the absence of historical data and trend reference, the baseline elevation value is directly used as the dynamic trend prediction result. The dynamic trend prediction result is the predicted elevation change trend value at the current moment based on historical data or a preset baseline, used to reflect the overall change pattern of the terrain. Historical weighting factors are predefined weights assigned to each historical preprocessed elevation data point in the incomplete window state. The weighted average is the sum of the products of each historical preprocessed elevation data point and its corresponding historical weighting factor, reflecting the weighted trend of recent historical data in the incomplete window state. The trend smoothing coefficient is a predefined coefficient between 0 and 1, used to reduce the fluctuation range of the weighted average (weighted average × trend smoothing coefficient), avoiding instability in trend prediction due to insufficient and highly volatile data in the incomplete window state. Since historical data is insufficient in the incomplete window state to fit a stable linear trend, the trend prediction result is generated through "recent historical weighting + smoothing processing." The core logic is to use recent data to guide the trend while mitigating fluctuations. The revised linear trend equation refers to the new equation obtained by adjusting the trend slope of the fitted linear trend equation under full-window conditions using a trend smoothing coefficient. The aim is to make trend changes smoother and avoid prediction distortion caused by abrupt slope changes. Under full-window conditions, sufficient historical data has been fitted to create a linear equation that reflects the global trend. However, the original slope may experience short-term jumps due to local fluctuations. Therefore, the slope needs to be adjusted using a trend smoothing coefficient. The core logic is to enhance the continuity and stability of the trend. The deviation value is calculated as the absolute difference between the current preprocessed elevation data and the dynamic trend prediction result. Its core function is to quantify the degree of deviation between the current data and the overall trend. If the deviation value is small, it indicates that the current data closely matches the overall trend, and the dynamic trend prediction result is highly reliable, allowing for increased weighting during subsequent fusion. If the deviation value is large, it indicates that the current data may deviate from the overall trend, reducing the reference value of the dynamic trend prediction result, requiring a reduction in its weight during subsequent fusion. In this embodiment, the calculation of the dynamic trend prediction result achieves full-scene coverage through window state adaptation, while the deviation value quantifies the deviation between the current data and the trend. The dynamic trend prediction results generated throughout the process complement the micro-scale and meso-scale results in a three-dimensional structure of "trend-detail-noise". Finally, through dynamic adjustment of the fusion weights, the target filtering results that take into account the needs of multiple dimensions are output.

[0119] S280. Calculate the fusion weight parameters based on the deviation value and noise fluctuation coefficient, and then fuse the dynamic trend prediction results, microscale filtering results and mesoscale filtering results according to the fusion weight parameters to obtain the target filtering result. Provide the target filtering result to the grader so as to adjust the grader blade in real time.

[0120] S290. Update the current trend sequence based on the current raw elevation data and the current preprocessed elevation data for filtering the new raw elevation data received in the next round.

[0121] The technical solution of this invention, through the refinement of the overall scheme and the specific implementation logic of singularity correction, multi-scale filtering, and dynamic trend prediction, ensures the accuracy and adaptability of each link. Specifically: it clarifies the differentiated correction methods for non-full-window and full-window states, adapting to scenarios with insufficient and sufficient historical data through trend prediction, data normalization, and weighted fusion, effectively eliminating the interference of singularities on the data and ensuring the accuracy of preprocessed elevation data at different data accumulation stages; it clarifies the window adaptation rules, trend influence coefficient calculation methods, and fusion strategies for micro-scale and meso-scale filtering, dynamically adjusting the filtering logic based on noise fluctuation coefficient and grader working status, achieving a balance between micro-scale terrain detail preservation and meso-scale noise suppression, and solving the adaptation defects of traditional fixed filtering under complex working conditions; it clarifies the trend prediction method for windowed states, generating dynamic trend prediction results through weighted averaging, linear trend smoothing, etc., and quantifying the deviation value between the current data and the trend, providing a direct basis for the adaptive adjustment of subsequent fusion weights, ensuring the continuity of trend tracking and the rationality of fusion decisions.

[0122] Example 3

[0123] Figure 3 This is a schematic diagram of an adaptive filtering device for grader elevation data provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes:

[0124] The trend sequence acquisition module 310 is used to acquire the most recently updated current trend sequence when it receives the current raw elevation data transmitted by the grader. The current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data.

[0125] The singularity detection module 320 is used to perform singularity detection on the current original elevation data according to the number of historical data stored in the current trend sequence and the matching singularity detection rules, and to correct the current original elevation data according to the singularity detection results to obtain the current preprocessed elevation data.

[0126] The multi-scale filtering result calculation module 330 is used to perform residual analysis on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient, and to perform multi-scale filtering on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain micro-scale filtering results and meso-scale filtering results.

[0127] The deviation value calculation module 340 is used to calculate the dynamic trend prediction result of the current preprocessed elevation data based on the current trend sequence and predefined trend tracking parameters, and to calculate the deviation value between the current preprocessed elevation data and the dynamic trend prediction result.

[0128] The filtering result output module 350 is used to calculate the fusion weight parameters based on the deviation value and the noise fluctuation coefficient, and then fuse the dynamic trend prediction result, the microscale filtering result and the mesoscale filtering result according to the fusion weight parameters to obtain the target filtering result. The target filtering result is then provided to the grader to adjust the action of the grader blade in real time.

[0129] The trend sequence update module 360 ​​is used to update the current trend sequence based on the current raw elevation data and the current preprocessed elevation data, so as to provide filtering processing for the new raw elevation data received in the next round.

[0130] In this embodiment of the invention, adaptive full-scene operation is achieved. By detecting singularities in different states and dynamically adjusting the filtering window and fusion weights, it accurately responds to the dynamic changes in data accumulation stages, noise intensity, and terrain trends, solving the problems of easy misjudgment and poor adaptability of traditional methods. It balances multi-dimensional needs, taking into account terrain detail preservation, medium-to-strong noise suppression, and macro trend tracking without manual intervention, adapting to complex working conditions. Closed-loop iterative optimization continuously updates the trend sequence, allowing data processing accuracy to gradually improve as the operation progresses. It has strong real-time performance and practicality, with a simple and efficient algorithm that can meet the real-time control requirements of grader blades and can be directly implemented in engineering applications.

[0131] Optionally, based on the above embodiments, the singularity detection module 320 may include:

[0132] The non-full window state detection unit is used to calculate the absolute value of the first difference between the current original elevation data and the arithmetic mean of each historical preprocessed elevation data if the number of historical data stored in the current trend sequence is less than the preset full window number threshold; if the absolute value of the first difference exceeds the preset fixed deviation threshold, the current original elevation data is detected as a singularity in a non-full window state.

[0133] The full-window state detection unit is used to fit a linear trend equation to each historical preprocessed elevation data in chronological order if the number of historical data stored in the current trend sequence is equal to a preset full-window number threshold, and predict the trend calculation value of the current original elevation data based on the linear trend equation; if the absolute value of the second difference between the trend calculation value and the current original elevation data exceeds a preset floating deviation threshold, the current original elevation data is detected as a singularity in the full-window state.

[0134] The empty window state detection unit is used to detect the current raw elevation data as a non-singularity in an empty window state if the number of historical data stored in the current trend sequence is zero.

[0135] Optionally, based on the above embodiments, the singularity detection module 320 may further include:

[0136] The non-full-window trend prediction calculation unit is used to calculate the predicted value based on a preset benchmark elevation value if the detection result is a singularity in a non-full-window state. and raw elevation data Calculation formula based on non-full window trend forecast value The non-full window trend prediction value was calculated. ;

[0137] Normalization unit, used to normalize the original elevation data to obtain normalized elevation data values, denoted as ;

[0138] The non-full-window singularity correction unit is used to extract the non-full-window trend prediction value. Normalized elevation data values Substitute into the formula for calculating non-full-window singularity correction data: In the process, preprocessed elevation data is obtained; among them, , These are the preset fixed weight values ​​under the non-full-window singularity detection rule.

[0139] Optionally, based on the above embodiments, the singularity detection module 320 may further include:

[0140] The full-window trend prediction calculation unit is used to obtain the trend intercept by calling the fitted linear trend equation if the detection result is a singularity in the full-window state. and trend slope The formula for calculating the full-window trend forecast value is as follows: The full-window trend prediction value was calculated. ; wherein, the The number of historical data stored in the current trend sequence;

[0141] The trend weight calculation unit is used to calculate the trend slope. Substitute into the trend weight calculation formula: In the process, the trend weights are obtained. ;

[0142] Normalization unit, used to normalize the original elevation data to obtain normalized elevation data values, denoted as ;

[0143] The full-window singularity correction unit is used to extract the full-window trend prediction value. Trend weight and normalized elevation data values Substitute into the full-window singularity correction data calculation formula: In this process, preprocessed elevation data is obtained.

[0144] Optionally, based on the above embodiments, the multi-scale filtering result calculation module 330 may include:

[0145] The historical data extraction unit is used to extract all historical preprocessed elevation data from the historical data stored in the current trend sequence, forming a historical preprocessed elevation data sequence arranged in chronological order.

[0146] The window length division unit is used to divide the window length for each window center according to the size of the preset basic sliding window, taking each historical preprocessed elevation data in the historical preprocessed elevation data sequence as the window center in turn.

[0147] The historical smoothing value calculation unit is used to calculate the arithmetic mean of the historical preprocessed elevation data falling within the window length for each historical preprocessed elevation data, and use it as the historical smoothing value corresponding to the current window center, so as to finally obtain a historical smoothing value sequence that corresponds one-to-one with the historical preprocessed elevation data sequence.

[0148] The historical noise standard deviation calculation unit is used to calculate the difference between each historical preprocessed elevation data and its corresponding historical smoothing value to obtain a single historical residual. All the single historical residuals are combined to form a complete historical residual sequence, and the overall historical noise standard deviation of the historical residual sequence is calculated based on the sample standard deviation formula.

[0149] The adjacent data determination unit is used to select the last target historical preprocessed elevation data in the historical preprocessed elevation data sequence as the center of the reference window, and select a corresponding number of target adjacent historical preprocessed elevation data based on the size of the preset basic sliding window.

[0150] The smoothing reference value calculation unit is used to calculate the arithmetic mean of the corresponding number of target adjacent historical preprocessed elevation data, target historical preprocessed elevation data and current preprocessed data, as a smoothing reference value corresponding to the center of the reference window;

[0151] The residual normalization unit is used to calculate the difference between the current preprocessed elevation data and the smoothed reference value to obtain the reference residual. The reference residual is then normalized by combining the overall historical noise standard deviation to obtain the noise fluctuation coefficient of the current preprocessed elevation data.

[0152] Optionally, based on the above embodiments, the multi-scale filtering result calculation module 330 may include:

[0153] The microscale filtering window length determination unit is used to determine the first window length required for microscale filtering based on the noise fluctuation coefficient and according to the predefined noise fluctuation coefficient and filtering intensity matching rules in the multiscale filtering configuration.

[0154] The first type of trend influence coefficient calculation unit is used to calculate the trend influence coefficient based on the fluctuation range of historical preprocessed elevation data and the preset ratio value defined in the multi-scale filtering configuration if the current trend sequence is not in a full window state or in an empty window state.

[0155] The second type of trend influence coefficient calculation unit is used to read the predefined mapping relationship between trend slope and trend influence coefficient in the multi-scale filtering configuration if the current trend sequence is in a full window state, and calculate the trend influence coefficient based on the mapping relationship.

[0156] The historical data weighted sum calculation unit is used to extract the first historical preprocessed elevation data within the first window length in reverse time extension order, assign the first type of weight to each first historical preprocessed elevation data according to the predefined weight allocation rules in the multi-scale filtering configuration, and calculate the weighted sum of the first historical preprocessed elevation data.

[0157] The stationary state second type weight determination unit is used to obtain the current working state of the grader. If the grader is stationary, the second type weight of the current preprocessed elevation data is determined as the ratio of the square of the first window length to the reference elevation value.

[0158] The second type of weight determination unit for motion state is used to determine the second type of weight of the current preprocessed elevation data as the product of the trend influence coefficient and the length of the first window if the grader is in motion state.

[0159] The second type of weight calculation unit is used to calculate the microscale filtering result based on the weighted sum of the first historical preprocessed elevation data and the second type of weight of the current preprocessed elevation data.

[0160] Optionally, based on the above embodiments, the multi-scale filtering result calculation module 330 may further include:

[0161] The mesoscale filtering window length determination unit is used to determine the second window length required for mesoscale filtering based on the first window length and the multi-scale filtering configuration; wherein, the multi-scale filtering configuration defines a multiple relationship between the second window length and the first window length;

[0162] The non-full-window state mesoscale trend prediction calculation unit is used to extract the second historical preprocessed elevation data within the second window length in reverse time extension order if the current trend sequence is in a non-full-window state, and to calculate the linear fitting value of the second historical preprocessed elevation data as the mesoscale trend prediction value.

[0163] The full-window state mesoscale trend prediction calculation unit is used to obtain the acquisition time of the current original elevation data if the current trend sequence is in a full-window state, substitute the acquisition time into the fitted linear trend equation, and combine it with the second window length to obtain the mesoscale trend prediction value.

[0164] The mesoscale trend prediction calculation unit in the empty window state is used to directly use the benchmark elevation value as the mesoscale trend prediction value if the current trend sequence is in the empty window state.

[0165] The motion state mesoscale filtering calculation unit is used to obtain the current working state of the grader. If the grader is in motion, the microscale filtering result and the mesoscale trend prediction value are weighted and fused according to a preset first fusion ratio to obtain the mesoscale filtering result.

[0166] A stationary state threshold extraction unit is used to compare the absolute value of the third difference with the preset mesoscale jump threshold if the grader is stationary.

[0167] The first type of mesoscale filtering result calculation unit is used to calculate the mesoscale filtering result by weighting and fusing the microscale filtering result and the mesoscale trend prediction value according to a preset second fusion ratio if the absolute value of the third difference exceeds the mesoscale jump threshold.

[0168] The second type of mesoscale filtering result calculation unit is used to directly use the mesoscale trend prediction value as the mesoscale filtering result if the absolute value of the third difference does not exceed the preset mesoscale jump threshold.

[0169] Optionally, based on the above embodiments, the deviation value calculation module 340 may include:

[0170] The window state dynamic trend prediction result determination unit is used to obtain the window state of the current trend sequence. If the window state is an empty window state, the benchmark elevation value is used as the dynamic trend prediction result.

[0171] The non-full window state dynamic trend prediction result determination unit is used to calculate the weighted average of all historical preprocessed elevation data in the current trend sequence based on the predefined historical weighting factor if the window state is non-full window state, and then multiply it by the predefined trend smoothing coefficient to obtain the dynamic trend prediction result.

[0172] The full-window state dynamic trend prediction result determination unit is used to, if the window state is full, adjust the trend slope based on the fitted linear trend equation and a predefined trend smoothing coefficient to obtain a corrected linear trend equation, and substitute the current original elevation data acquisition time into the corrected linear trend equation to obtain the dynamic trend prediction result.

[0173] The absolute difference calculation unit is used to calculate the absolute difference between the current preprocessed elevation data and the dynamic trend prediction result, and uses the absolute difference as the deviation value.

[0174] Optionally, based on the above embodiments, the filtering result output module 350 may include:

[0175] The initial weight allocation unit is used to query a predefined base ratio in the multi-scale filtering configuration, and allocate an initial first weight to the microscale filtering result, an initial second weight to the mesoscale filtering result, and an initial third weight to the dynamic trend prediction result according to the base ratio, with a total of 1.

[0176] The first type of initial weight adjustment unit is used to query a preset trend matching threshold in the multi-scale filtering configuration. If the deviation value is not greater than the preset trend matching threshold, the increase of the initial third weight is determined according to the predefined deviation adjustment ratio corresponding to the dynamic trend prediction result. The increase of the initial third weight is then distributed as the deduction of the initial first weight and the deduction of the initial second weight according to the basic ratio, so as to obtain the first weight, the second weight, and the third weight of the first adjustment.

[0177] The second type of initial weight adjustment unit is used to determine the reduction amount of the initial third weight according to the deviation adjustment ratio if the deviation value is greater than the preset trend matching threshold, and to distribute the reduction amount as the increase amount of the initial first weight and the increase amount of the initial second weight according to the base ratio, so as to obtain the first weight, the second weight and the third weight of the initial adjustment.

[0178] The noise fluctuation level determination unit is used to query a preset noise level threshold in the multi-scale filtering configuration and determine the noise fluctuation level by comparing the noise fluctuation coefficient with the noise level threshold.

[0179] A high-level secondary weight adjustment unit is used to determine the increase of the second weight of the first adjustment according to a predefined noise adjustment ratio corresponding to the mesoscale filtering result if the noise fluctuation is high-level. At the same time, the increase of the second weight of the first adjustment is used as the deduction of the first weight of the first adjustment, so as to obtain the first weight of the secondary adjustment, the second weight of the secondary adjustment, and the third weight of the first adjustment.

[0180] The low-level secondary weight adjustment unit is used to determine the increase of the first weight of the first adjustment according to the noise adjustment ratio corresponding to the microscale filtering result if it is a low-level noise fluctuation, and at the same time use the increase of the first weight of the first adjustment as the deduction of the second weight of the first adjustment, so as to obtain the first weight of the secondary adjustment, the second weight of the secondary adjustment, and the third weight of the first adjustment.

[0181] The weight fusion unit is used to normalize the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment through the attention weight algorithm to obtain the fusion weight parameters corresponding to the microscale filtering result, the mesoscale filtering result, and the dynamic trend prediction result, respectively.

[0182] Optionally, based on the above embodiments, the filtering result output module 350 may further include:

[0183] The fusion weight definition unit is used to define the fusion weight corresponding to the microscale filtering result as the first fusion weight, the fusion weight corresponding to the mesoscale filtering result as the second fusion weight, and the fusion weight corresponding to the dynamic trend prediction result as the third fusion weight.

[0184] The target filtering result calculation unit is used to calculate the product of the microscale filtering result and the first fusion weight, the product of the mesoscale filtering result and the second fusion weight, and the product of the dynamic trend prediction result and the third fusion weight, respectively. The target filtering result is obtained by adding the results of the three products.

[0185] The elevation adjustment command sending unit is used to encapsulate the target filtering result into an elevation adjustment command that conforms to the data format of the grader control system through a preset communication interface, and to drive the grader blade to adjust its movement. It also receives the movement adjustment completion signal synchronously fed back by the grader control system, thus forming a closed-loop control.

[0186] The adaptive filtering device for grader elevation data provided in this embodiment of the invention can execute the adaptive filtering method for grader elevation data provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0187] Example 4

[0188] Figure 4 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0189] like Figure 4As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0190] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0191] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as an adaptive filtering method for grader elevation data.

[0192] That is, when the current raw elevation data transmitted by the grader is received, the most recently updated current trend sequence is obtained. The current trend sequence includes multiple historical data stored in chronological order. Each historical data includes historical raw elevation data and historical preprocessed elevation data.

[0193] Based on the amount of historical data stored in the current trend sequence, singularity detection is performed on the current original elevation data using matching singularity detection rules, and the current original elevation data is corrected according to the singularity detection results to obtain the current preprocessed elevation data.

[0194] Residual analysis is performed on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient. Based on the noise fluctuation coefficient, multi-scale filtering is performed on the current preprocessed elevation data to obtain micro-scale filtering results and meso-scale filtering results.

[0195] Based on the current trend sequence and predefined trend tracking parameters, calculate the dynamic trend prediction result of the current preprocessed elevation data, and calculate the deviation between the current preprocessed elevation data and the dynamic trend prediction result;

[0196] The fusion weight parameters are calculated based on the deviation value and the noise fluctuation coefficient. Then, the dynamic trend prediction results, microscale filtering results and mesoscale filtering results are fused according to the fusion weight parameters to obtain the target filtering result. The target filtering result is then provided to the grader to adjust the grader blades in real time.

[0197] The current trend sequence is updated based on the current raw elevation data and the current preprocessed elevation data, in order to provide filtering for the new raw elevation data received in the next round.

[0198] In some embodiments, an adaptive filtering method for grader elevation data can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the adaptive filtering method for grader elevation data described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform an adaptive filtering method for grader elevation data by any other suitable means (e.g., by means of firmware).

[0199] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0200] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0201] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0202] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0203] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0204] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0205] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0206] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An adaptive filtering method for grader elevation data, characterized in that, include: Upon receiving the current raw elevation data transmitted by the grader, the most recently updated current trend sequence is obtained. The current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data. Based on the amount of historical data stored in the current trend sequence, singularity detection is performed on the current original elevation data using matching singularity detection rules, and the current original elevation data is corrected according to the singularity detection results to obtain the current preprocessed elevation data. Residual analysis is performed on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient. Based on the noise fluctuation coefficient, multi-scale filtering is performed on the current preprocessed elevation data to obtain micro-scale filtering results and meso-scale filtering results. Based on the current trend sequence and predefined trend tracking parameters, calculate the dynamic trend prediction result of the current preprocessed elevation data, and calculate the deviation between the current preprocessed elevation data and the dynamic trend prediction result; The fusion weight parameters are calculated based on the deviation value and the noise fluctuation coefficient. Then, the dynamic trend prediction results, microscale filtering results and mesoscale filtering results are fused according to the fusion weight parameters to obtain the target filtering result. The target filtering result is then provided to the grader to adjust the grader blades in real time. The current trend sequence is updated based on the current raw elevation data and the current preprocessed elevation data, in order to provide filtering for the new raw elevation data received in the next round.

2. The method according to claim 1, characterized in that, Based on the amount of historical data stored in the current trend sequence, singularity detection rules are used to detect singularities in the current raw elevation data. The raw elevation data is then corrected according to the singularity detection results to obtain the current preprocessed elevation data, including: If the number of historical data stored in the current trend sequence is less than the preset full window number threshold, then the absolute value of the first difference between the current raw elevation data and the arithmetic mean of each historical preprocessed elevation data is calculated; if the absolute value of the first difference exceeds the preset fixed deviation threshold, then the current raw elevation data is detected as a singularity in a non-full window state. If the number of historical data stored in the current trend sequence is equal to the preset full window number threshold, then the historical preprocessed elevation data are fitted in the order of time extension to obtain a linear trend equation, and the trend calculation value of the current original elevation data is predicted based on the linear trend equation; if the absolute value of the second difference between the trend calculation value and the current original elevation data exceeds the preset floating deviation threshold, then the current original elevation data is detected as a singularity in the full window state. If the number of historical data stored in the current trend sequence is zero, then the current raw elevation data is detected as a non-singularity in a window state.

3. The method according to claim 2, characterized in that, Based on the singularity detection results, the current original elevation data is corrected to obtain the current preprocessed elevation data, including: If the detection result is a singularity in a non-full-window state, then it is based on the preset benchmark elevation value. and raw elevation data Calculation formula based on non-full window trend forecast value The non-full window trend prediction value was calculated. ; wherein, the The number of historical data stored in the current trend sequence; The normalized elevation data is obtained by normalizing the original elevation data, denoted as . ; Extract the non-full window trend prediction value Normalized elevation data values Substitute into the formula for calculating non-full-window singularity correction data: In the process, preprocessed elevation data is obtained; among them, , These are the preset fixed weight values ​​under the non-full-window singularity detection rule.

4. The method according to claim 2, characterized in that, Based on the singularity detection results, the current original elevation data is corrected to obtain the current preprocessed elevation data, which also includes: If the detection result is a singularity in a full-window state, then the trend intercept is obtained by calling the fitted linear trend equation. and trend slope The formula for calculating the full-window trend forecast value is as follows: The full-window trend prediction value was calculated. ; The slope of the trend Substitute into the trend weight calculation formula: In the middle, the trend weight is obtained. ; The normalized elevation data is obtained by normalizing the original elevation data, denoted as . ; Extract the full-window trend prediction value Trend weight and normalized elevation data values Substitute into the full-window singularity correction data calculation formula: In this process, preprocessed elevation data is obtained.

5. The method according to claim 1, characterized in that, Residual analysis was performed on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend series to obtain the noise fluctuation coefficient, including: Extract all historical preprocessed elevation data from the historical data stored in the current trend sequence to form a historical preprocessed elevation data sequence arranged in chronological order. Based on the size of the preset basic sliding window, the window length is divided for each historical preprocessed elevation data in the historical preprocessed elevation data sequence as the window center. For each historical preprocessed elevation data, the arithmetic mean of the historical preprocessed elevation data falling within the window length is calculated and used as the historical smoothed value corresponding to the current window center. Finally, a historical smoothed value sequence corresponding one-to-one with the historical preprocessed elevation data sequence is obtained. The difference between each historical preprocessed elevation data and its corresponding historical smoothed value is calculated to obtain a single historical residual. All the single historical residuals are combined to form a complete historical residual sequence, and the overall historical noise standard deviation of the historical residual sequence is calculated based on the sample standard deviation formula. In the historical preprocessed elevation data sequence, the last target historical preprocessed elevation data is selected as the center of the reference window, and a corresponding number of target adjacent historical preprocessed elevation data are selected based on the size of the preset basic sliding window. Calculate the arithmetic mean of the corresponding number of target adjacent historical preprocessed elevation data, target historical preprocessed elevation data and current preprocessed data, and use it as a smoothing reference value corresponding to the center of the reference window; The difference between the current preprocessed elevation data and the smoothed reference value is calculated to obtain the reference residual. The reference residual is then normalized by combining it with the overall historical noise standard deviation to obtain the noise fluctuation coefficient of the current preprocessed elevation data.

6. The method according to claim 2 or 3, characterized in that, Multi-scale filtering is performed on the current preprocessed elevation data based on the noise fluctuation coefficient, yielding micro-scale filtering results and meso-scale filtering results, including: Based on the noise fluctuation coefficient, the first window length required for microscale filtering is determined according to the predefined noise fluctuation coefficient and filter intensity matching rules in the multi-scale filtering configuration. If the current trend sequence is in a non-full window state or an empty window state, then based on the fluctuation range of historical preprocessed elevation data, the preset proportion value defined in the multi-scale filtering configuration is used as the trend influence coefficient. If the current trend sequence is in a full window state, read the predefined mapping relationship between the trend slope and the trend influence coefficient in the multi-scale filtering configuration, and calculate the trend influence coefficient based on the mapping relationship; Extract the first historical preprocessed elevation data within the first window length according to the reverse time extension order, assign the first type of weight to each first historical preprocessed elevation data according to the predefined weight allocation rules in the multi-scale filtering configuration, and calculate the weighted sum of the first historical preprocessed elevation data. Obtain the current working status of the grader. If the grader is stationary, determine the second type of weight of the current preprocessed elevation data as the ratio of the square of the first window length to the reference elevation value. If the grader is in motion, the second type of weight of the current preprocessed elevation data is determined as the product of the trend influence coefficient and the length of the first window; The microscale filtering result is calculated based on the weighted sum of the first historical preprocessed elevation data and the second type of weight of the current preprocessed elevation data.

7. The method according to claim 6, characterized in that, Multi-scale filtering is performed on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain micro-scale filtering results and meso-scale filtering results, which also include: The second window length required for mesoscale filtering is determined based on the first window length and the multi-scale filtering configuration; wherein, the multi-scale filtering configuration defines a multiple relationship between the second window length and the first window length; If the current trend sequence is not in a full window state, the second historical preprocessed elevation data within the second window length is extracted in reverse time extension order, and the linear fitting value of the second historical preprocessed elevation data is used as the mesoscale trend prediction value. If the current trend sequence is in a full window state, the acquisition time of the current original elevation data is obtained, the acquisition time is substituted into the fitted linear trend equation, and the mesoscale trend prediction value is obtained by combining the second window length. If the current trend sequence is in a blank state, the benchmark elevation value is directly used as the mesoscale trend prediction value. Obtain the current working status of the grader. If the grader is in motion, the microscale filtering result and the mesoscale trend prediction value are weighted and fused according to a preset first fusion ratio to obtain the mesoscale filtering result. If the grader is stationary, compare the absolute value of the third difference with the preset mesoscale jump threshold. If the absolute value of the third difference exceeds the mesoscale jump threshold, the microscale filtering result and the mesoscale trend prediction value are weighted and fused according to a preset second fusion ratio to obtain the mesoscale filtering result. If the absolute value of the third difference does not exceed the preset mesoscale jump threshold, the mesoscale trend prediction value is directly used as the mesoscale filtering result.

8. The method according to claim 3, characterized in that, Based on the current trend sequence and predefined trend tracking parameters, calculate the dynamic trend prediction result of the current preprocessed elevation data, and calculate the deviation between the current preprocessed elevation data and the dynamic trend prediction result, including: Obtain the window state of the current trend sequence. If the window state is empty, then use the benchmark elevation value as the dynamic trend prediction result. If the window state is not full, the weighted average of all historical preprocessed elevation data in the current trend sequence is calculated based on the predefined historical weighting factor, and then multiplied by the predefined trend smoothing coefficient to obtain the dynamic trend prediction result. If the window is full, the trend slope is smoothed by combining the fitted linear trend equation with a predefined trend smoothing coefficient to obtain a corrected linear trend equation. The acquisition time of the current original elevation data is substituted into the corrected linear trend equation to obtain the dynamic trend prediction result. Calculate the absolute difference between the current preprocessed elevation data and the dynamic trend prediction result, and use the absolute difference as the deviation value.

9. The method according to claim 1, characterized in that, The fusion weight parameters are calculated based on the deviation value and the noise fluctuation coefficient, including: In the multi-scale filtering configuration, query the predefined basic ratio, and assign an initial first weight to the microscale filtering result, an initial second weight to the mesoscale filtering result, and an initial third weight to the dynamic trend prediction result according to the basic ratio, with a total of 1. In the multi-scale filtering configuration, a preset trend matching threshold is queried. If the deviation value is not greater than the preset trend matching threshold, the increase of the initial third weight is determined according to the predefined deviation adjustment ratio corresponding to the dynamic trend prediction result. The increase of the initial third weight is then distributed as the deduction of the initial first weight and the deduction of the initial second weight according to the base ratio, so as to obtain the first weight, the second weight, and the third weight adjusted for the first time. If the deviation value is greater than the preset trend matching threshold, the reduction amount of the initial third weight is determined according to the deviation adjustment ratio, and the reduction amount is distributed as the increase amount of the initial first weight and the increase amount of the initial second weight according to the base ratio, so as to obtain the first weight, the second weight and the third weight of the first adjustment. In the multi-scale filtering configuration, a preset noise level threshold is queried, and the noise fluctuation level is determined by comparing the noise fluctuation coefficient with the noise level threshold. If it is a high-level noise fluctuation, the increase of the second weight of the first adjustment is determined according to the predefined noise adjustment ratio corresponding to the mesoscale filtering result. At the same time, the increase of the second weight of the first adjustment is used as the deduction of the first weight of the first adjustment, so as to obtain the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment. If it is a low-level noise fluctuation, the increase of the first weight of the first adjustment is determined according to the noise adjustment ratio corresponding to the microscale filtering result. At the same time, the increase of the first weight of the first adjustment is used as the deduction of the second weight of the first adjustment, so as to obtain the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment. The attention weight algorithm is used to normalize the first weight of the second adjustment, the second weight of the second adjustment, and the third weight of the first adjustment to obtain the fusion weight parameters corresponding to the microscale filtering result, the mesoscale filtering result, and the dynamic trend prediction result, respectively.

10. The method according to claim 9, characterized in that, According to the fusion weight parameters, the dynamic trend prediction results, micro-scale filtering results, and meso-scale filtering results are fused to obtain the target filtering result. This target filtering result is then provided to the grader to adjust the grader blade's movement in real time, including: The fusion weight corresponding to the microscale filtering result is defined as the first fusion weight, the fusion weight corresponding to the mesoscale filtering result is defined as the second fusion weight, and the fusion weight corresponding to the dynamic trend prediction result is defined as the third fusion weight. The product of the microscale filtering result and the first fusion weight, the product of the mesoscale filtering result and the second fusion weight, and the product of the dynamic trend prediction result and the third fusion weight are calculated separately. The target filtering result is obtained by adding the results of the three products. The target filtering result is encapsulated into a height adjustment command that conforms to the data format of the grader control system through a preset communication interface. This command is used to drive the grader blade to adjust its movements. The system also receives the movement adjustment completion signal synchronously fed back by the grader control system, thus forming a closed-loop control.

11. An adaptive filtering device for grader elevation data, characterized in that, include: The trend sequence acquisition module is used to acquire the most recently updated current trend sequence when it receives the current raw elevation data transmitted by the grader. The current trend sequence includes multiple historical data stored in chronological order, and each historical data includes historical raw elevation data and historical preprocessed elevation data. The singularity detection module is used to perform singularity detection on the current raw elevation data according to the number of historical data stored in the current trend sequence and the matching singularity detection rules, and to correct the current raw elevation data according to the singularity detection results to obtain the current preprocessed elevation data. The multi-scale filtering result calculation module is used to perform residual analysis on the current preprocessed elevation data and each historical preprocessed elevation data in the current trend sequence to obtain the noise fluctuation coefficient, and to perform multi-scale filtering on the current preprocessed elevation data based on the noise fluctuation coefficient to obtain micro-scale filtering results and meso-scale filtering results. The deviation value calculation module is used to calculate the dynamic trend prediction result of the current preprocessed elevation data based on the current trend sequence and predefined trend tracking parameters, and to calculate the deviation value between the current preprocessed elevation data and the dynamic trend prediction result. The filtering result output module is used to calculate the fusion weight parameters based on the deviation value and noise fluctuation coefficient, and then fuse the dynamic trend prediction result, microscale filtering result and mesoscale filtering result according to the fusion weight parameters to obtain the target filtering result. The target filtering result is then provided to the grader to adjust the grader blade in real time. The trend sequence update module is used to update the current trend sequence based on the current raw elevation data and the current preprocessed elevation data, so as to provide filtering processing for the new raw elevation data received in the next round.

12. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform an adaptive filtering method for grader elevation data according to any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute an adaptive filtering method for grader elevation data according to any one of claims 1-10.

Images