Microwave radar data missing value classification and imputation method and system based on reinforcement learning

Abstract

The application belongs to the technical field of microwave radar, and particularly relates to a microwave radar data missing value classification and interpolation method and system based on reinforcement learning. The method obtains microwave radar measurement data, determines the data source of the missing item based on the variation trend of the to-be-judged parameter, and then classifies the radar parameter data into an interpolatable item or an uninterpolatable item, and removes the uninterpolatable item. The effective data segments adjacent in time of the interpolatable item are divided into multiple sub-data segments, the pre-interpolation gap of each sub-data segment is calculated, and the time interval in the adjacent data area is matched to determine the fluctuation parameter, and the interpolation compensation value is calculated by combining a preset interpolation formula. The application avoids introducing false signals into the data set, ensures the physical meaning and logical rationality of the interpolation, and improves the overall quality and reliability of the repaired data set.

Description

Technical Field

[0001] This invention belongs to the field of microwave radar technology, specifically relating to a method and system for classifying and imputing missing values ​​in microwave radar data based on reinforcement learning. Background Technology

[0002] Microwave radar, as an electronic device that uses electromagnetic waves for detection, ranging, speed measurement, and imaging, provides data support for continuous data acquisition in fields such as weather forecasting, aerospace, and autonomous driving. These applications rely on the accuracy, continuity, and completeness of the acquired microwave radar data. In actual deployment environments, due to multiple factors such as complex electromagnetic interference, severe weather conditions, momentary equipment failures, or signal blockage, data loss inevitably occurs in radar data sequences. Data loss has become one of the technical bottlenecks restricting the improvement of radar technology application effectiveness.

[0003] The problem of missing data in radar data sequences is addressed by existing regression interpolation or time-series interpolation methods, which only fit the data from a single mathematical or temporal dimension, ignoring the complex physical causes and environmental background that lead to the missing data. These methods struggle to distinguish between transient missing data caused by occasional signal interference and persistent data anomalies due to equipment aging, resulting in imputed results that lack physical validity and impacting data reliability. Existing interpolation methods using conventional neural networks are typically trained on specific datasets. When microwave radar faces dynamic and varied operating environments, such as changing weather types and interference sources, the interpolation accuracy can become unstable, potentially leading to misleading data. Furthermore, conventional neural network interpolation methods lack the ability to identify and classify missing data patterns. They employ a uniform interpolation logic and struggle to dynamically select repair strategies based on the length, frequency, and surrounding data characteristics of the missing data, making it difficult to handle diverse data missing issues in complex scenarios.

[0004] In view of this, the present invention provides a method and system for classifying and imputing missing values ​​in microwave radar data based on reinforcement learning. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for classifying and imputing missing values ​​in microwave radar data based on reinforcement learning.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A reinforcement learning-based method for classifying and imputing missing values ​​in microwave radar data includes the following steps: Acquire microwave radar measurement data corresponding to the target area; Identify missing items from the acquired microwave radar measurement data; Obtain radar parameter data within the data bands preceding and following the missing item in time; Based on the changing trend of radar parameter data, the data source was determined to be radar wave obstruction or noise signal interference. When the data source is radar wave obstruction, the missing items are classified as imputable items; When the data source is noise interference, missing items are removed; Calculate the interpolation compensation value based on the valid data segment adjacent to the interpolable term; Interpolation compensation values ​​are used to interpolate the interpolable terms; Perform an optimization evaluation process to update the interpolation compensation value based on the optimal interpolation scheme.

[0007] Preferably, determining the data source as radar wave obstruction or noise signal interference based on the changing trend of radar parameter data includes: The deviation between the missing item and its temporally nearest data node in the data band is extracted as the parameter to be determined. If the trend of the parameter to be determined is continuous, the data source is determined to be radar wave blockage. If the trend of the parameter to be determined is instantaneous change, the data source is determined to be noise signal interference.

[0008] Preferably, calculating the interpolation compensation value based on the valid data segment adjacent to the interpolable term includes: The effective data segment is divided into multiple sub-data segments according to the preset number of unit nodes; for any sub-data segment, its pre-interpolation gap and fluctuation parameters are determined; based on the fluctuation parameters, the interpolation compensation value is calculated by using the interpolation formula with preset interpolation parameters.

[0009] Preferably, determining the pre-interpolation gap and fluctuation parameters for any sub-data segment includes: Within the adjacent data region of the sub-data segment, calculate the time interval between multiple pairs of adjacent data nodes; compare the pre-interpolation gap with each time interval, calculate the difference, and determine the adjacent data node pairs whose difference is less than the preset judgment threshold as matching node pairs; if there are matching node pairs, determine the average of the differences of all matching node pairs as the fluctuation parameter.

[0010] Preferably, the step of performing an optimization evaluation process to update the interpolation compensation value based on the optimal interpolation scheme includes: Generate predicted values ​​for multiple candidate interpolation terms; use each predicted value to perform simulated interpolation on the interpolation term to obtain predicted simulated data corresponding to each predicted value; compare each predicted simulated data with the interpolated data to be verified, calculate the deviation, and output the supplementary accuracy corresponding to the predicted value that meets the preset allowable range based on the comparison result of the deviation and the preset allowable range; sort the supplementary accuracy to determine the optimal interpolation scheme based on the sorting result.

[0011] Preferably, the step of sorting the interpolation accuracy to determine the optimal interpolation scheme based on the sorting results includes: Select the predicted value corresponding to the highest supplementation accuracy, and determine the prediction parameters on which the generated predicted value is based as the optimal parameters; define the optimal interpolation scheme as replacing the preset interpolation parameters with the optimal parameters.

[0012] The microwave radar data missing value classification and imputation system based on reinforcement learning includes the following modules: The data acquisition module is used to acquire microwave radar measurement data corresponding to the target area; The missing item identification and classification module is used to identify missing items from the acquired microwave radar measurement data; acquire radar parameter data in the data band before and after the missing item in time; determine whether the data source is radar wave blockage or noise signal interference based on the changing trend of radar parameter data; when the data source is radar wave blockage, classify the missing item as an imputable item. The interpolation strategy optimization module is used to calculate the interpolation compensation value based on the valid data segments adjacent to the interpolable item; interpolate the interpolable item using the interpolation compensation value; and execute the optimization evaluation process to update the interpolation compensation value based on the optimal interpolation scheme. The data update module is used to interpolate the interpolable items using the updated interpolation compensation values ​​and to remove missing items.

[0013] Preferably, determining the data source as radar wave obstruction or noise signal interference based on the changing trend of radar parameter data includes: The deviation between the missing item and its temporally nearest data node in the data band is extracted as the parameter to be determined. If the trend of the parameter to be determined is continuous, the data source is determined to be radar wave blockage. If the trend of the parameter to be determined is instantaneous change, the data source is determined to be noise signal interference.

[0014] Preferably, calculating the interpolation compensation value based on the valid data segment adjacent to the interpolable term includes: The effective data segment is divided into multiple sub-data segments according to the preset number of unit nodes; for any sub-data segment, its pre-interpolation gap and fluctuation parameters are determined; based on the fluctuation parameters, the interpolation compensation value is calculated by using the interpolation formula with preset interpolation parameters.

[0015] Preferably, determining the pre-interpolation gap and fluctuation parameters for any sub-data segment includes: Within the adjacent data region of the sub-data segment, calculate the time interval between multiple pairs of adjacent data nodes; compare the pre-interpolation gap with each time interval, calculate the difference, and determine the adjacent data node pairs whose difference is less than the preset judgment threshold as matching node pairs; if there are matching node pairs, determine the average of the differences of all matching node pairs as the fluctuation parameter.

[0016] Beneficial effects This invention addresses any missing item in microwave radar measurement data by acquiring radar parameter data within the data band preceding and following it in time. Based on the changing trend of the parameter to be determined, the data source of the missing item is identified. Subsequently, the radar parameter data is classified into imputable or non-imputable items, and non-imputable items are removed. This avoids introducing erroneous signals into the dataset, ensures the physical meaning and logical rationality of the imputation, and improves the overall quality and reliability of the repaired dataset.

[0017] This invention divides the effective data segments that are adjacent in time to the interpolation items into multiple sub-data segments. By calculating the pre-interpolation gap of each sub-data segment and matching the time interval in its neighboring data area to determine the fluctuation parameter, the interpolation compensation value is calculated in combination with the preset interpolation formula. This allows the interpolation process to adapt to the actual fluctuation of the data, ensures the consistency between the interpolation result and the original data in terms of the change pattern, and improves the accuracy of interpolation.

[0018] After completing the initial interpolation, this invention uses the interpolation result as the data to be verified, generates predicted values ​​for multiple candidate interpolation values, calculates the deviation between each predicted value and the data to be verified through simulated interpolation, and outputs the corresponding supplementary accuracy. Based on the ranking of supplementary accuracy, the optimal interpolation scheme is determined and the interpolation compensation value is updated. Thus, guided by the supplementary accuracy, the optimal interpolation scheme is found and determined, realizing the correction and improvement of the interpolation scheme, and ensuring that the interpolation compensation value used has high confidence and accuracy. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Example 1 Please refer to Figure 1 This embodiment provides a reinforcement learning-based method for classifying and imputing missing values ​​in microwave radar data. This method is applied to address data loss issues caused by radar wave obstruction, noise interference, and other phenomena during microwave radar measurement data acquisition. The method includes the following steps: S1. Perform data preprocessing and missing structure identification; acquire microwave radar measurement data corresponding to the target area in time sequence, scan the microwave radar measurement data, identify all invalid or empty data records, and define these records as missing items; define the missing items in a structured manner and analyze the distribution pattern of the missing data in depth. The structured definition of missing items includes: merging multiple missing items that occur consecutively in time to define a consecutive missing segment; defining a sequence segment containing valid data that exists between any two missing items, or a sequence segment containing valid data that exists between two consecutive missing segments, as a missing interval segment. When defining the boundaries of consecutive missing segments, a time interval check is introduced to ensure the accuracy of the determination. The time interval check is a preset time threshold used to determine whether two time-adjacent missing items should belong to the same consecutive missing segment. The time interval check can be 50 milliseconds. Defining the boundaries of consecutive missing segments includes: determining whether the time interval between any missing item and its immediate next missing item is less than the time interval check value; if it is less than the time interval check value, then the two missing items are considered consecutive missing items and belong to the same consecutive missing segment; if it is not less than the time interval check value, then it indicates that effective data acquisition has occurred between the two missing items and they do not belong to the same consecutive missing segment. In this case, the next missing item is determined as the new comparison benchmark, and the process continues to proceed. The confirmation rules enhance the robustness of the judgment and avoid erroneously truncating complete consecutive missing segments due to a single instantaneously recovered non-missing item. The confirmation rules include: when four consecutive sets of missing items are identified, it indicates that the missing is caused by persistent factors, and at this time, the system is confirmed to enter a stable consecutive missing state; the current missing item in the fourth set is identified as the last missing item, and the data node of the first non-missing item that follows it is identified as the end point of the consecutive missing segment; thereby effectively filtering out transient signal recovery phenomena and more accurately capturing and defining the missing process caused by complete radar wave obstruction or continuous noise signal interference.

[0022] S2. Classify the causes of missing items. For any identified missing item, analyze its contextual data characteristics to infer its data source and determine whether it is due to radar wave blockage or noise signal interference. Among them, radar wave blockage is the phenomenon of signal strength smoothly attenuating to ineffectiveness caused by physical obstructions; noise signal interference is the phenomenon of signal exhibiting drastic and irregular jumps caused by external electromagnetic interference factors. Specifically, radar parameter data for the missing item is obtained in the two valid data bands before and after the missing item in time. The timestamp of the missing item is used as the time reference center. The valid data segment before the time is defined as the front band, and the valid data segment after the time is defined as the back band. Through a band dynamic calibration mechanism, it is ensured that the selected band is a stable and representative band. The band dynamic calibration mechanism includes: calculating the average deviation of all data nodes in the front and back bands respectively, and determining these two average deviations as parameters to be calibrated; setting a preset calibration value, and determining whether any parameter to be calibrated is greater than the calibration value; wherein, the preset calibration value represents the maximum reasonable fluctuation range of the data under normal acquisition conditions, and the preset calibration value can be 0.05; If any parameter to be checked is greater than the check value, it indicates that the currently selected band may contain other anomalous data and is not suitable as an analysis benchmark. In this case, the time window is extended forward or backward to include more data points to adjust the time range of the preceding or following band, and the parameter to be checked is recalculated based on the adjusted time range. The band dynamic calibration mechanism is executed cyclically until both parameters to be checked are less than or equal to the check value, thereby determining the preceding and following bands used for causal analysis and ensuring the stability and reliability of the data on which the classification decision is based.

[0023] S3. Extract the deviation between the missing item and its temporally nearest valid data node in the front and back bands as the parameter to be determined, and determine the data source of the missing item based on the changing trend of the parameter to be determined. Among them, the parameters to be determined directly reflect the transition characteristics of data when it transitions from a normal state to a missing state and from a missing state back to a normal state. Determining the source of missing data based on the changing trend of the parameter to be determined includes: if the changing trend of the parameter to be determined shows a smooth decrease in data value to the missing state or a smooth recovery from the missing state to the valid value, and this smooth and continuous changing trend is consistent with the signal attenuation characteristics when a physical entity gradually enters or leaves the radar beam range, then the source of data is determined to be radar wave obstruction; if the changing trend of the parameter to be determined shows a sudden and irregular jump in data value before and after the missing point, then the source of data is determined to be noise signal interference.

[0024] S4. Based on the classification results of the causes of missing items, implement differentiated data processing strategies; If the source of the missing data is determined to be radar wave obstruction, then the missing data is considered to be due to a physical process, and its trend is continuous and predictable, making it suitable for data imputation. The missing data is then classified as an imputable item. If the source of the data is determined to be noise interference, then the data at the missing point is considered to be an unreliable random value. Imputing it would not have physical meaning and might even introduce incorrect information, polluting the entire dataset. The missing item is then classified as an unimpeded item. All missing items classified as unimpeded items are removed from the dataset to prevent the spread of noise interference.

[0025] S5. Impute all missing items that are classified as imputable items; All interpolable items are uniformly identified as interpolation items. For each interpolation item, the effective data segments that are temporally adjacent are identified as interpolation data segments. Each interpolation data segment is divided into multiple sub-data segments according to a preset number of unit nodes. Each sub-data segment is defined as a sub-sample to be interpolated for feature matching and parameter calculation. The interpolation data segment contains the local temporal features required for interpolation calculation, and the preset number of unit nodes can be 10 data nodes.

[0026] S6. For each subsample to be supplemented, determine the pre-insertion gap and quantify its internal temporal non-uniformity. Determining the pre-interpolation gap includes: acquiring the start and end point data of each sub-data segment; calculating the time interval between the start point data and its preceding data node in time to obtain the previous interval value; and calculating the time interval between the end point data and its following data node in time to obtain the subsequent interval value. Calculate the absolute value of the difference between the previous interval value and the subsequent interval value, and determine this absolute value as the pre-interpolation gap of the sub-data segment. This pre-interpolation gap value reflects the change in the data sampling rate at the boundary of the sub-data segment and is an important basis for interpolation calculation.

[0027] S7. Based on the subsamples to be supplemented and their corresponding pre-interpolation gaps, determine the fluctuation parameters used for interpolation calculation, and match the parameters that best fit the local dynamic characteristics of the current data for the interpolation calculation process. Specifically, a preset judgment threshold is used to measure the similarity of time intervals, where the preset judgment threshold can be 5ms; for each subsample to be supplemented, the time interval between multiple pairs of adjacent data nodes is calculated in its neighboring data region; the pre-insertion gap of the sample is compared with each calculated time interval, and the difference between the two is calculated. If the difference is less than the preset judgment threshold, it indicates that the time fluctuation pattern of the pair of adjacent data nodes is similar to that of the subsample to be supplemented, and they are identified as a matching node pair; the number of all matching node pairs is counted to obtain the number of matches, and the average difference of the difference of all matching node pairs is calculated to obtain the average difference, where the average difference represents the typical time fluctuation in the region that is most similar to the feature of the subsample to be supplemented. If the average difference is less than the preset judgment threshold, it indicates that a reliable local fluctuation pattern has been found, and the average difference is determined as the fluctuation parameter of the subsample to be supplemented; if the average difference is higher than or equal to the preset judgment threshold, it indicates that a sufficiently similar pattern has not been found in the neighboring area, and the matching result is determined to be invalid in order to avoid interpolation based on unreliable information.

[0028] S8. Based on the effective fluctuation parameters, calculate the interpolation compensation value of the interpolable item according to the preset interpolation calculation rules, and use the calculated interpolation compensation value to interpolate the corresponding item to be interpolated, thus completing the preliminary interpolation process. The preset interpolation calculation rule can be a Lagrange interpolation or spline interpolation calculation rule that dynamically adjusts the weights of multiple adjacent data nodes based on the fluctuation parameter.

[0029] S9. The interpolation results are evaluated and iteratively optimized through an evaluation and feedback mechanism; the evaluation and feedback mechanism forms a closed-loop feedback, enabling the interpolation method to improve itself. The evaluation and feedback mechanism includes: determining the preliminary interpolation results of the interpolable item as the data to be verified; fine-tuning the prediction parameters used in the current interpolation calculation rule based on the preset parameter perturbation rule to generate a set of distinct candidate parameters; using each candidate parameter set, generating corresponding candidate interpolation values ​​as predicted values ​​for the interpolable item according to the interpolation calculation rule; and using each predicted value to perform simulated interpolation on the interpolable item to obtain the predicted simulation data corresponding to each predicted value. The preset parameter perturbation rules include: slightly increasing or decreasing the basic weights and fluctuation parameters used in the interpolation calculation rules according to a preset ratio of 0.1%; Each predicted simulation data is compared with the data to be verified, and the deviation between the two is calculated. Based on the comparison result of the deviation with the preset allowable range, the supplementary accuracy corresponding to each predicted value is output. The supplementary accuracy is a quantitative reflection of the deviation. The smaller the deviation, the higher the accuracy. Specifically, for each predicted simulation data, the absolute value of the total deviation between the predicted simulation data and the data to be verified is calculated; a preset allowable deviation value is used to screen out qualified predictions; wherein, the preset allowable deviation value is 0.05; If the absolute value of the total deviation is less than the allowable deviation, the supplementary accuracy is calculated and output based on the total deviation, and the corresponding prediction value is determined to be a reliable prediction value; if the absolute value of the total deviation is not less than the allowable deviation, the corresponding prediction value is determined to be an unreliable prediction value and will not participate in the subsequent optimization process. The multiple supplementary accuracies corresponding to the interpolable item are sorted, and all supplementary accuracies corresponding to the reliable prediction value are arranged in descending order of value. The optimal interpolation scheme is determined based on this sorting result. The item with the highest supplementary accuracy in the sorting result is selected, and the prediction parameter on which the prediction value corresponding to this item is generated is determined as the optimal parameter. The interpolation parameter used in generating the interpolation compensation value is replaced with the optimal parameter. Based on the updated interpolation parameter, the interpolation compensation value is recalculated and updated. This process is iterated continuously, so that the interpolation calculation process can dynamically adapt to changes in the data flow and continuously improve the interpolation accuracy.

[0030] Example 2 Please refer to Figure 2 This embodiment provides a microwave radar data missing value classification and imputation system based on reinforcement learning, including the following modules: The data acquisition module is used to acquire raw microwave radar measurement data of the target area within a specific time range from one or more microwave radar devices or data repositories.

[0031] The missing item identification and classification module is used to preprocess the acquired raw data, identify the missing items, and classify them according to their causes to determine the processing strategy; it scans the time series of microwave radar measurement data to identify missing items that are discontinuous or marked as invalid. After identifying the missing item, obtain the radar parameter data of the missing item in the two valid data bands before and after it in time. Determine the data source of the missing item based on the changing trend of the radar parameter data. Extract the deviation between the missing item and the valid data node that is closest to it in time in the data band, and use this deviation as the parameter to be determined. If the parameter to be determined changes continuously over time, the data source is determined to be radar wave obstruction, indicating that the target signal is temporarily blocked but its trajectory is continuous and predictable. Therefore, the missing item is classified as an interpolable item and the interpolable item is interpolated. If the parameter to be determined shows a sudden change and has no logical connection with the surrounding data, the data source is determined to be noise signal interference, indicating that the missing data is usually invalid and unpredictable. Therefore, the missing data is classified as an imputable item and the imputable item is removed.

[0032] The interpolation strategy optimization module is used to calculate the interpolation compensation value and iteratively improve the accuracy of interpolation by optimizing the evaluation process when a missing item classified as an interpolable item is received. Based on the valid data segment adjacent to the interpolable item, the module calculates the initial interpolation compensation value. The valid data segment is divided into multiple sub-data segments according to the preset number of unit nodes. For any sub-data segment, the duration or range of its missing data is determined as the pre-interpolation gap and fluctuation parameters. The determination of the fluctuation parameter includes: calculating the time interval between multiple pairs of adjacent data nodes in the adjacent data area of ​​the sub-data segment, and comparing these time intervals with the pre-interpolation gap; if the difference between the two is less than the preset judgment threshold, then the pair of adjacent data nodes is determined as a matching node pair, and the average difference is obtained by calculating the average difference of all matching node pairs. If the average difference is less than the preset judgment threshold, the average difference is determined as the fluctuation parameter that can reflect the dynamic characteristics of local data; based on the fluctuation parameter and by using the interpolation calculation rules of the interpolation parameter, the initial interpolation compensation value is calculated. An optimization evaluation process is performed to find the optimal interpolation scheme and update the interpolation compensation value. The optimization evaluation process includes: generating multiple candidate interpolation values ​​as predicted values ​​for interpolable items, where each predicted value is generated based on different prediction parameters; using each predicted value to simulate interpolate the interpolable item to obtain a series of corresponding predicted simulation data; comparing each predicted simulation data with the interpolated data to be verified and calculating the deviation between them. Based on the comparison between the calculated deviation and the preset allowable range, supplementary accuracy is output for each predicted value that meets the allowable range. The higher the supplementary accuracy, the better the interpolation effect. All supplementary accuracies are sorted. Based on the sorting results, the predicted value corresponding to the highest supplementary accuracy is selected. The prediction parameters on which the optimal predicted value is generated are determined as the optimal parameters. The optimal interpolation scheme is defined as replacing the original interpolation parameters with the optimal parameters, updating and outputting optimized and more accurate interpolation compensation values.

[0033] The data update module is used to make final modifications to the original data based on the processing results of other modules. It receives the optimized and updated interpolation compensation value output by the interpolation strategy optimization module, and uses the value to interpolate the corresponding interpolable items to fill the data gaps. It receives the unimpeded items marked by the missing item identification and classification module, performs a removal operation on these invalid data caused by noise signal interference, and outputs high-quality microwave radar measurement data that has been classified, imputed and cleaned.

[0034] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for classifying and imputing missing values ​​in microwave radar data based on reinforcement learning, characterized in that, Includes the following steps: Acquire microwave radar measurement data corresponding to the target area; Identify missing items from the acquired microwave radar measurement data; Obtain radar parameter data within the data bands preceding and following the missing item in time; Based on the changing trend of radar parameter data, the data source was determined to be radar wave obstruction or noise signal interference. When the data source is radar wave obstruction, the missing items are classified as imputable items; When the data source is noise interference, missing items are removed; Calculate the interpolation compensation value based on the valid data segment adjacent to the interpolable term; Interpolation compensation values ​​are used to interpolate the interpolable terms; Perform an optimization evaluation process to update the interpolation compensation value based on the optimal interpolation scheme; The calculation of the interpolation compensation value based on the valid data segment adjacent to the interpolable item includes: dividing the valid data segment into multiple sub-data segments according to a preset number of unit nodes; determining the pre-interpolation gap and fluctuation parameters for any sub-data segment; and calculating the interpolation compensation value based on the fluctuation parameters using an interpolation formula with preset interpolation parameters. For any sub-data segment, determining its pre-imputation gap and volatility parameter includes: calculating the time interval between multiple pairs of adjacent data nodes in the adjacent data region of the sub-data segment; comparing the pre-imputation gap with each time interval, calculating the difference, and determining the adjacent data node pairs with differences less than a preset judgment threshold as matching node pairs; if there are matching node pairs, the average of the differences of all matching node pairs is determined as the volatility parameter. Determining the pre-interpolation gap includes: acquiring the start point data and end point data of each sub-data segment; calculating the time interval between the start point data and its preceding data node in time to obtain the previous interval value; calculating the time interval between the end point data and its following data node in time to obtain the subsequent interval value; calculating the absolute value of the difference between the previous interval value and the subsequent interval value, and determining the absolute value as the pre-interpolation gap of the sub-data segment. The optimization evaluation process for updating interpolation compensation values ​​based on the optimal interpolation scheme includes: generating predicted values ​​for multiple candidate interpolation values ​​to be used as interpolation items; performing simulated interpolation on the interpolation items using each predicted value to obtain predicted simulated data corresponding to each predicted value; comparing each predicted simulated data with the interpolated data to be verified, calculating the deviation, and outputting the supplementary accuracy corresponding to the predicted value that meets the preset allowable range based on the comparison result of the deviation and the preset allowable range; and sorting the supplementary accuracy to determine the optimal interpolation scheme based on the sorting result.

2. The microwave radar data missing value classification and imputation method based on reinforcement learning according to claim 1, characterized in that, The determination of the data source as radar wave obstruction or noise signal interference based on the changing trend of radar parameter data includes: The deviation between the missing item and its temporally nearest data node in the data band is extracted as the parameter to be determined. If the trend of the parameter to be determined is continuous, the data source is determined to be radar wave blockage. If the trend of the parameter to be determined is instantaneous change, the data source is determined to be noise signal interference.

3. The microwave radar data missing value classification and imputation method based on reinforcement learning according to claim 1, characterized in that, The process of ranking the interpolation accuracy to determine the optimal interpolation scheme based on the ranking results includes: Select the predicted value corresponding to the highest supplementation accuracy, and determine the prediction parameters on which the generated predicted value is based as the optimal parameters; define the optimal interpolation scheme as replacing the preset interpolation parameters with the optimal parameters.

4. A microwave radar data missing value classification and imputation system based on reinforcement learning, used to implement the microwave radar data missing value classification and imputation method based on reinforcement learning as described in any one of claims 1-3, characterized in that, Includes the following modules: The data acquisition module is used to acquire microwave radar measurement data corresponding to the target area; The missing item identification and classification module is used to identify missing items from the acquired microwave radar measurement data; and to acquire radar parameter data in the data bands before and after the missing items in time. Based on the changing trend of radar parameter data, the data source was determined to be radar wave obstruction or noise signal interference. When the data source is radar wave obstruction, the missing items are classified as imputable items; The interpolation strategy optimization module is used to calculate the interpolation compensation value based on the valid data segments adjacent to the interpolable item; and to interpolate the interpolable item using the interpolation compensation value. Perform an optimization evaluation process to update the interpolation compensation value based on the optimal interpolation scheme; The data update module is used to interpolate the interpolable items using the updated interpolation compensation values ​​and to remove missing items.

5. The microwave radar data missing value classification and imputation system based on reinforcement learning according to claim 4, characterized in that, The determination of the data source as radar wave obstruction or noise signal interference based on the changing trend of radar parameter data includes: The deviation between the missing item and its temporally nearest data node in the data band is extracted as the parameter to be determined. If the trend of the parameter to be determined is continuous, the data source is determined to be radar wave blockage. If the trend of the parameter to be determined is instantaneous change, the data source is determined to be noise signal interference.

6. The microwave radar data missing value classification and imputation system based on reinforcement learning according to claim 4, characterized in that, The calculation of the interpolation compensation value based on the valid data segment adjacent to the interpolable term includes: The effective data segment is divided into multiple sub-data segments according to the preset number of unit nodes; for any sub-data segment, its pre-interpolation gap and fluctuation parameters are determined; based on the fluctuation parameters, the interpolation compensation value is calculated by using the interpolation formula with preset interpolation parameters.

7. The microwave radar data missing value classification and imputation system based on reinforcement learning according to claim 6, characterized in that, The determination of the pre-interpolation gap and fluctuation parameters for any given sub-data segment includes: Within the adjacent data region of the sub-data segment, calculate the time interval between multiple pairs of adjacent data nodes; compare the pre-interpolation gap with each time interval, calculate the difference, and determine the adjacent data node pairs whose difference is less than the preset judgment threshold as matching node pairs; if there are matching node pairs, determine the average of the differences of all matching node pairs as the fluctuation parameter.

Images