Economic activity nowcasting method and system based on remote sensing observation effectiveness

Abstract

The application discloses an economic activity present-time prediction method and system based on remote sensing observation effectiveness, and particularly relates to the technical fields of remote sensing image processing and economic present-time prediction. The method comprises the following steps: extracting original remote sensing features of each economic activity spatial unit at each time point and under each remote sensing mode, and determining an original observation effectiveness index; fusing the to-be-corrected remote sensing features and a completion result based on the index to obtain corrected remote sensing features; determining a completion reliability index according to the source data of the completion result, and combining the original observation effectiveness index to determine effective quality; performing hierarchical aggregation based on the effective quality to obtain an economic activity index and hierarchical aggregation effectiveness; adjusting high-frequency remote sensing input weights according to the hierarchical aggregation effectiveness of each high-frequency time point to generate a present-time prediction result, and determining prediction period aggregation effectiveness to control the output state of the result.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing image processing and real-time economic forecasting, and particularly to a method and system for real-time economic activity forecasting based on the effectiveness of remote sensing observations. Background Technology

[0002] Current economic forecasting typically involves directly inputting multi-source remote sensing features such as nighttime light, surface temperature, and atmospheric pollutant concentrations into the prediction model, or first performing quality screening, cloud removal, and time-series synthesis on the remote sensing images. When cloud cover obscures data, effective pixels are missing, observation times are inconsistent, or sensor quality fluctuates, enterprise-level remote sensing features are prone to abnormal declines or absences, leading to misjudgments of economic activity status, biases in industry or regional aggregation results, and difficulty in determining the stability and reliability of the prediction results. Summary of the Invention

[0003] The purpose of this invention is to solve the technical problem that when the quality of multi-source remote sensing observations is unstable, it is difficult to retain the original observation quality of the correction results of low-quality remote sensing features and transmit them to the hierarchical aggregation and prediction input weights. This invention provides a method and system for real-time prediction of economic activities based on the effectiveness of remote sensing observations.

[0004] The first aspect of this invention provides a method for real-time prediction of economic activities based on the effectiveness of remote sensing observations, comprising: acquiring multi-source remote sensing data and corresponding remote sensing quality data of multiple economic activity spatial units at multiple time points; extracting original remote sensing features of each economic activity spatial unit at each time point and under each remote sensing mode from the multi-source remote sensing data; determining the original observation effectiveness index corresponding to each original remote sensing feature based on the remote sensing quality data; determining remote sensing features to be corrected based on the original observation effectiveness index; determining the completion result corresponding to the remote sensing features to be corrected; and fusing the remote sensing features to be corrected and the completion result based on the original observation effectiveness index to obtain corrected remote sensing features, wherein the original observation effectiveness index is not updated with the corrected remote sensing features; determining the completion reliability index corresponding to the completion result based on the observation effectiveness and completion performance of the completion source data used to generate the completion result; and performing real-time prediction of economic activity spatial units at each time point and under each remote sensing mode from the multi-source remote sensing data. The original observation effectiveness indices under each remote sensing modality are aggregated and combined with the corresponding completion reliability index to determine the effective quality of each economic activity spatial unit at each time point. If the completion result exists, the effective quality is constrained using a preset quality upper limit. The aggregation weight of each economic activity spatial unit is determined based on the effective quality. The corrected remote sensing features and the effective quality are then aggregated hierarchically according to the aggregation weights to obtain the economic activity index and hierarchical aggregation effectiveness. The economic activity index at each high-frequency time point is used as the high-frequency remote sensing input to the mixed-frequency real-time prediction model. The weights of the corresponding high-frequency remote sensing inputs are adjusted according to the hierarchical aggregation effectiveness at each high-frequency time point to obtain the real-time prediction result of the target economic indicator. The aggregation effectiveness for the prediction period is determined based on the hierarchical aggregation effectiveness at each high-frequency time point. The output state of the real-time prediction result is controlled based on the comparison between the aggregation effectiveness for the prediction period and a preset prediction output threshold.

[0005] In one embodiment of the present invention, the remote sensing quality data includes at least two of the following: cloud occlusion information, pixel availability information, time matching information, sensor quality information, anomaly deviation information, spatial processing error information, and historical prediction error information. The pixel availability information is determined based on the proportion of effective pixels or the proportion of missing pixels. The step of determining the original observation effectiveness index corresponding to each original remote sensing feature based on the remote sensing quality data includes: determining the quality components corresponding to each remote sensing quality data item; normalizing each quality component and weighting and summing the normalized quality components according to the quality weights corresponding to each remote sensing quality data item to obtain a weighted summation result; and constraining the weighted summation result to a preset index range to obtain the original observation effectiveness index, wherein the sum of each quality weight is 1.

[0006] In one embodiment of the present invention, determining the remote sensing feature to be corrected based on the original observation validity index includes: determining the original remote sensing feature whose original observation validity index is lower than a preset validity threshold of the corresponding remote sensing mode as the remote sensing feature to be corrected; fusing the remote sensing feature to be corrected and the completion result based on the original observation validity index includes: determining the original value retention ratio based on the original observation validity index corresponding to the remote sensing feature to be corrected; performing weighted fusion of the remote sensing feature to be corrected and the completion result according to the original value retention ratio; when the remote sensing feature to be corrected is completely missing, marked as invalid by quality control, or the corresponding original observation validity index is not greater than a preset lower limit, setting the original value retention ratio to zero, wherein the preset lower limit is not higher than the preset validity threshold.

[0007] In one embodiment of the present invention, the completion source data includes at least one of the following: historical remote sensing features of the corresponding economic activity spatial unit, remote sensing features of other economic activity spatial units belonging to the same region and industry as the corresponding economic activity spatial unit, other remote sensing modal features of the corresponding economic activity spatial unit, and historical economic activity features; determining the completion result corresponding to the remote sensing feature to be corrected and the completion reliability index corresponding to the completion result includes: generating candidate completion values ​​based on each completion source data; determining the source score corresponding to each completion source data according to the observation validity, historical completion accuracy, and inverse score of cross-validation error of each completion source data; determining the corresponding combination weight according to each source score, wherein, when the completion source data includes one source, the combination weight corresponding to that source is set to 1, and when the completion source data includes multiple sources, the source scores are normalized to obtain the combination weight corresponding to each source; weighting and fusing each candidate completion value according to the combination weight to obtain the completion result, and weighting each source score according to the combination weight to obtain the completion reliability index.

[0008] In one embodiment of the present invention, determining the effective quality of each economic activity spatial unit at each time point includes: weighted aggregation of the original observation effectiveness indices of each economic activity spatial unit at each time point and under each remote sensing mode to obtain the basic effectiveness index of the corresponding economic activity spatial unit at the corresponding time point; weighted aggregation of the completion reliability indices corresponding to each remote sensing mode with completion results to obtain a comprehensive completion reliability index, wherein the completion reliability index corresponding to the remote sensing mode without completion results is set to zero; when there is at least one completion result, calculating the sum of the comprehensive completion reliability index scaled by the completion gain coefficient and the basic effectiveness index, and determining the smaller value between the sum and the preset quality upper limit as the effective quality; when there is no completion result, determining the basic effectiveness index as the effective quality.

[0009] In one embodiment of the present invention, determining the aggregation weight of each economic activity spatial unit based on the effective quality, and performing hierarchical aggregation of the modified remote sensing features and the effective quality according to the aggregation weight, includes: determining the modal fusion weight based on the modal effective quality and historical prediction contribution corresponding to each remote sensing modality, and fusing the modified remote sensing features of each economic activity spatial unit under each remote sensing modality according to the modal fusion weight to obtain the remote sensing activity features of each economic activity spatial unit, wherein the modal effective quality is based on the original observation effectiveness index of the corresponding remote sensing modality and the corresponding completion capability when completion results exist. The reliability index is determined; based on the scale weight, sample representativeness weight, historical predictability weight, and effective quality corresponding to each economic activity spatial unit, the aggregate score corresponding to each economic activity spatial unit is determined; within the industry group or regional group to which each economic activity spatial unit belongs, the aggregate score is normalized to obtain the aggregate weight corresponding to each economic activity spatial unit; the remote sensing activity characteristics of each economic activity spatial unit are weighted and aggregated according to the aggregate weight to obtain the economic activity index, and the effective quality of each economic activity spatial unit is weighted and aggregated according to the aggregate weight to obtain the hierarchical aggregation effectiveness.

[0010] In one embodiment of the present invention, the mixed-frequency current prediction model is a mixed-frequency data sampling model; adjusting the weights of the high-frequency remote sensing inputs according to the hierarchical aggregation effectiveness includes: obtaining the preset benchmark input weights of the mixed-frequency data sampling model for each high-frequency time point in the economic activity index; determining the product of each benchmark input weight and the hierarchical aggregation effectiveness of the corresponding high-frequency time point, and normalizing each product to obtain the effective input weights corresponding to each high-frequency time point; inputting the daily, weekly, ten-day, or monthly cumulative economic activity index into the mixed-frequency data sampling model according to each effective input weight to predict the target economic indicators for the month or quarter, and obtaining the current prediction model. The results are measured; according to the effective input weights corresponding to each high-frequency time point, the hierarchical aggregation effectiveness corresponding to each high-frequency time point is weighted and aggregated to obtain the aggregation effectiveness for the prediction period; when the aggregation effectiveness for the prediction period is not lower than the first prediction output threshold, the current prediction result is output normally; when the aggregation effectiveness for the prediction period is lower than the first prediction output threshold and the pause output condition is not met, the current prediction result is output according to the low confidence level; when the aggregation effectiveness for the prediction period is lower than the second prediction output threshold and the number of effective economic activity spatial units is lower than the preset number, the output of the current prediction result is paused, wherein the second prediction output threshold is lower than the first prediction output threshold.

[0011] A second aspect of the present invention provides a real-time prediction system for economic activities based on the effectiveness of remote sensing observations, comprising: an effectiveness determination module, configured to acquire multi-source remote sensing data and corresponding remote sensing quality data of multiple economic activity spatial units at multiple time points, extract original remote sensing features of each economic activity spatial unit at each time point and in each remote sensing mode from the multi-source remote sensing data, and determine the original observation effectiveness index corresponding to each original remote sensing feature based on the remote sensing quality data; a feature correction module, configured to determine the remote sensing features to be corrected based on the original observation effectiveness index, determine the completion result corresponding to the remote sensing features to be corrected, and fuse the remote sensing features to be corrected and the completion result based on the original observation effectiveness index to obtain the corrected remote sensing features, wherein the original observation effectiveness index is not updated with the corrected remote sensing features; a reliability determination module, configured to determine the completion reliability index corresponding to the completion result based on the observation effectiveness and completion performance of the completion source data used to generate the completion result; and a quality determination module, configured to perform quality assessment on each economic activity spatial unit. The system aggregates the original observation effectiveness indices of each economic activity spatial unit at each time point and under each remote sensing mode, and combines them with the corresponding completion reliability index to determine the effective quality of each economic activity spatial unit at each time point. It also constrains the effective quality using a preset quality upper limit when the completion result exists. A hierarchical aggregation module determines the aggregation weight of each economic activity spatial unit based on the effective quality, and performs hierarchical aggregation on the corrected remote sensing features and the effective quality according to the aggregation weights to obtain the economic activity index and hierarchical aggregation effectiveness. A prediction control module uses the economic activity index at each high-frequency time point as the high-frequency remote sensing input to the mixed-frequency current prediction model, adjusts the weights of the corresponding high-frequency remote sensing inputs according to the hierarchical aggregation effectiveness at each high-frequency time point, and obtains the current prediction result of the target economic indicator. Based on the hierarchical aggregation effectiveness at each high-frequency time point, it determines the aggregation effectiveness for the prediction period, and controls the output state of the current prediction result based on the comparison result between the aggregation effectiveness for the prediction period and a preset prediction output threshold.

[0012] A third aspect of the present invention provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the method described in the first aspect.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0014] Compared with the prior art, the technical effects achieved by the present invention are as follows:

[0015] This invention determines the original observation effectiveness index based on remote sensing quality data, and uses this index to identify and correct low-quality remote sensing features. During the completion process, the original observation effectiveness index remains unchanged with the correction results, ensuring that the corrected remote sensing features correspond to their original observation quality. The observation effectiveness and completion performance of the source data are used to form a completion reliability index, which, together with the original observation effectiveness index, determines the effective quality constrained by a quality upper limit. Effective quality participates in determining the aggregation weights of economic activity spatial units and is synchronously aggregated hierarchically with the corrected remote sensing features, ensuring that the economic activity index and corresponding hierarchical aggregation effectiveness at each high-frequency time point reflect the observation quality at that time point. The hierarchical aggregation effectiveness corresponding to each high-frequency time point is further used to adjust the weights of the corresponding high-frequency remote sensing inputs, forming the prediction period aggregation effectiveness. This prediction period aggregation effectiveness controls the output state of the current prediction results, ensuring continuous transmission of remote sensing observation quality between feature correction, hierarchical aggregation, high-frequency input weighting, and prediction output. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for real-time prediction of economic activities based on the effectiveness of remote sensing observations according to an embodiment of the present invention; Figure 2 This is a framework diagram of a real-time economic activity prediction system based on the effectiveness of remote sensing observation according to an embodiment of the present invention; Figure 3 This is a framework diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0017] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0018] The technical solution of the present invention is illustrated below through specific embodiments. It should be understood that the one or more steps mentioned in the present invention do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not for limiting the order of each method or limiting the scope of the present invention. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of the present invention.

[0019] In one embodiment of this application, a method for real-time prediction of economic activities based on the effectiveness of remote sensing observations is provided. This method can be executed by a server, a cloud computing platform, or a computing device with data processing capabilities.

[0020] In this embodiment, the economic activity spatial unit refers to a spatial region used to define the range of remote sensing pixel extraction and characterize economic activity objects. It can correspond to enterprise factory areas, production and operation areas, or other economic activity objects with spatial boundaries. Remote sensing mode refers to a data type formed by different sensors, remote sensing products, or observation bands; the time point can be a single imaging moment, or a period within a day, week, ten days, or month, or other prediction time window.

[0021] like Figure 1 As shown, the current economic activity prediction method based on the effectiveness of remote sensing observation in this embodiment may specifically include the following steps: S101, acquire multi-source remote sensing data and corresponding remote sensing quality data of multiple economic activity spatial units at multiple time points, extract the original remote sensing features of each economic activity spatial unit at each time point and in each remote sensing mode from the multi-source remote sensing data, and determine the original observation effectiveness index corresponding to each original remote sensing feature based on the remote sensing quality data.

[0022] In this embodiment, the correspondence between economic activity spatial units and spatial ranges is first established. Data from different remote sensing modalities are converted to a unified spatial reference, and intersecting pixels are extracted according to the spatial range of the economic activity spatial units. For remote sensing data with different spatial resolutions, the range of pixels participating in feature calculation can be determined based on the overlapping area between pixels and spatial ranges. For remote sensing data with different observation times, matching is performed according to the time windows corresponding to each time point, so that the original remote sensing features can correspond one-to-one with economic activity spatial units, time points, and remote sensing modalities.

[0023] Raw remote sensing features refer to the feature data formed from corresponding remote sensing pixels before low-quality observations are supplemented or corrected. These features can be pixel statistics, temporal variation values, or standardized feature values. The remote sensing quality data maintains the same spatial, temporal, and modal indices as the raw remote sensing features and is used to describe the remote sensing observation state that generated the raw remote sensing features.

[0024] Based on remote sensing quality data, multiple quality factors affecting the usability of remote sensing observations are converted into quality characterization values ​​with unified dimensions. These quality characterization values ​​are then combined to obtain the original observation effectiveness index. The original observation effectiveness index is stored in association with the corresponding original remote sensing features, characterizing the usability of those features in subsequent feature processing and model input. Different remote sensing modalities can employ different combinations and transformation relationships of quality data to adapt to the imaging characteristics of different sensors.

[0025] S102, Based on the original observation effectiveness index, determine the remote sensing features to be corrected, determine the completion results corresponding to the remote sensing features to be corrected, and fuse the remote sensing features to be corrected and the completion results based on the original observation effectiveness index to obtain the corrected remote sensing features. The original observation effectiveness index is not updated with the corrected remote sensing features.

[0026] In this embodiment, original remote sensing features whose original observation validity index does not meet the validity conditions corresponding to the current remote sensing mode are identified as remote sensing features to be corrected. These features can be low-quality features with some valid observations, or features lacking valid observations.

[0027] The completion result is generated by the completion process corresponding to the remote sensing feature to be corrected. For remote sensing features to be corrected that still contain usable information, the degree to which the original features are retained in the fusion process is determined based on their original observation validity index; the higher the original observation validity index, the more original feature information is retained in the fusion result, and the lower the original observation validity index, the larger the proportion of the completion result in the fusion result. For remote sensing features to be corrected that do not have usable observations, the completion result can be used as the corresponding corrected remote sensing feature.

[0028] When the remote sensing feature to be corrected is completely missing or there are no valid pixels that can characterize the current observation state of the remote sensing mode, its original observation validity index is set to zero. When the remote sensing feature to be corrected is marked as invalid by quality control, its original observation validity index is set to a value no higher than a preset lower limit. In the above cases, the original value retention ratio is set to zero, and the completed result is used as the corrected remote sensing feature for the corresponding remote sensing mode. This processing only indicates that the original remote sensing observation itself is unusable, not that the completed result is unusable.

[0029] The corrected remote sensing features and the original observation validity index are stored separately. After generating the corrected remote sensing features, the original observation validity index is not recalculated or increased based on the completeness or numerical smoothness of the completion results, nor is the original observation validity index replaced by the quality assessment results corresponding to the corrected remote sensing features. Therefore, subsequent processing can still identify the original remote sensing observation state corresponding to the corrected remote sensing features.

[0030] Optionally, the same spatial unit identifier, time point identifier, and remote sensing mode identifier can be configured for the original remote sensing features, the completed results, the corrected remote sensing features, and the original observation effectiveness index, so that any corrected remote sensing feature can be traced back to the corresponding original remote sensing feature and the original observation effectiveness index.

[0031] S103, Based on the observation validity and completion performance of the completion source data used to generate the completion result, determine the completion reliability index corresponding to the completion result.

[0032] In this embodiment, the source data for completion refers to the data used to derive, estimate, or generate the completion result during the completion process. For each completion result, the source data for completion and the completion process method are recorded, and the observation validity of the source data itself is obtained.

[0033] Completion performance characterizes the completion capability of a corresponding completion processing method on historical or validation data. It can be determined based on the deviation between the completed value and the known true observation, the consistency of the completion direction, or the stability of the completion result in subsequent tasks. Combining the observational validity of the source data with the completion performance forms a completion reliability index. When the observational validity of the source data is low or the historical completion performance of the completion processing method is low, the completion reliability index of the corresponding completion result decreases; conversely, the completion reliability index increases.

[0034] The completion reliability index characterizes the credibility of the completion result itself, and it corresponds to different quality objects than the original observation validity index. The original observation validity index describes the usability of the remote sensing observations before correction, while the completion reliability index describes the credibility of the completion result generated based on the completion source data. The two are not interchangeable.

[0035] S104 aggregates the original observation validity indices of each economic activity spatial unit at each time point and under each remote sensing mode, and combines them with the corresponding completion reliability index to determine the effective quality of each economic activity spatial unit at each time point. If there are completion results, the effective quality is constrained by a preset quality upper limit.

[0036] In this embodiment, for the same economic activity spatial unit and the same time point, the original observation effectiveness indices corresponding to each remote sensing mode are collected, and the basic observation quality of the economic activity spatial unit at that time point is formed according to the preset mode aggregation relationship.

[0037] When completion results exist, the corresponding completion reliability index is used as a supplementary measure of the completion information to the sample quality, and together with the basic observation quality, it forms the effective quality. Effective quality retains the influence of the original remote sensing observation quality while also reflecting the usability of the completion results in the current sample. The completion reliability index does not directly replace the basic observation quality, but rather participates in the determination of effective quality according to a pre-defined combination relationship.

[0038] Optionally, for remote sensing modes with completion results, the effective quality of the mode is also determined based on the original observation validity index and the corresponding completion reliability index of the remote sensing mode. The effective quality of the mode is used to calculate the subsequent mode fusion weights, and its target differs from that of the effective quality at the economic activity spatial unit level: the effective quality of the mode is used to determine whether and to what extent the corresponding remote sensing mode after completion participates in the formation of remote sensing activity features, while the effective quality at the economic activity spatial unit level is used to control the contribution of the economic activity spatial unit in hierarchical aggregation.

[0039] In the absence of completion results, the basic observation quality is directly determined as the effective quality. In the presence of completion results, the basic observation quality and the completion reliability index are combined according to a preset relationship to form a combined quality, and the smaller of the combined quality and a preset quality upper limit is determined as the effective quality. The preset quality upper limit is not lower than the corresponding basic observation quality, and is used to limit the quality increment brought about by the completion reliability.

[0040] S105. Based on the effective quality, determine the aggregation weight of each economic activity spatial unit. Then, perform hierarchical aggregation of the corrected remote sensing features and effective quality according to the aggregation weight to obtain the economic activity index and hierarchical aggregation effectiveness.

[0041] In this embodiment, multiple economic activity spatial units are divided into corresponding data groups according to a preset hierarchical relationship. For each economic activity spatial unit in the same group, an aggregation weight is determined based on its effective quality at the current time point, and each aggregation weight satisfies a normalization condition within that group.

[0042] Based on the aggregation weights corresponding to each economic activity spatial unit, the corrected remote sensing features are weighted to obtain the economic activity index of the current group at the corresponding time point. The corrected remote sensing features are used to provide the characteristic values ​​of the economic activity index, and the effective quality is used to control the contribution of each economic activity spatial unit to the economic activity index, so that spatial units with low original observation quality and limited completion reliability will not be aggregated as high-quality samples with equal weights simply because they have obtained complete corrected remote sensing features.

[0043] By employing aggregation relationships within the same spatial unit range, the effective quality of each economic activity spatial unit is weighted to obtain the hierarchical aggregation effectiveness corresponding to the economic activity index. The economic activity index and the hierarchical aggregation effectiveness share the same group identifier and time point identifier. The former characterizes the remote sensing economic activity characteristics of the corresponding group, while the latter characterizes the overall observation quality of the spatial unit set that forms the economic activity index at the current time point.

[0044] As an feasible approach, when there are multi-level grouping relationships, economic activity indices and hierarchical aggregation effectiveness can be generated first at the lower level, and then the aggregation results at the lower level can be used as inputs for the next higher level aggregation to form economic activity indices and hierarchical aggregation effectiveness at different spatial or object levels.

[0045] S106, the economic activity index at each high-frequency time point is used as the high-frequency remote sensing input of the mixed-frequency current prediction model. The weight of the corresponding high-frequency remote sensing input is adjusted according to the hierarchical aggregation effectiveness corresponding to each high-frequency time point to obtain the current prediction result of the target economic indicator. The aggregation effectiveness of the prediction period is determined based on the hierarchical aggregation effectiveness corresponding to each high-frequency time point. Based on the comparison result of the aggregation effectiveness of the prediction period and the preset prediction output threshold, the output state of the current prediction result is controlled.

[0046] In this embodiment, the economic activity index and its corresponding hierarchical aggregation validity are configured with the same group identifier and time point identifier. For a target prediction period, the economic activity index and its corresponding hierarchical aggregation validity for each high-frequency time point within the target prediction period are obtained according to the group identifier and time point identifier, and the economic activity index for each high-frequency time point is determined as the high-frequency remote sensing input of the mixed-frequency real-time prediction model.

[0047] The mixed-frequency current forecasting model is used to handle the time-frequency difference between high-frequency remote sensing input and target economic indicators. The economic activity index at each high-frequency time point is input into the mixed-frequency current forecasting model according to its time position within the target forecast period, so that the mixed-frequency current forecasting model can generate the current forecast results of the corresponding target economic indicators based on the economic activity indexes already formed within the target forecast period.

[0048] Based on the hierarchical aggregation effectiveness corresponding to each high-frequency time point, the weights of the corresponding high-frequency remote sensing inputs are adjusted. Hierarchical aggregation effectiveness is used to characterize the overall quality of the set of economic activity spatial units that form the corresponding economic activity index at that high-frequency time point. When the hierarchical aggregation effectiveness is high, the corresponding economic activity index retains a higher input weight in the mixed-frequency current prediction model; when the hierarchical aggregation effectiveness is low, the input weight of the corresponding economic activity index is reduced accordingly. The adjusted high-frequency remote sensing inputs and their weights jointly participate in the generation of the current prediction results of the target economic indicator.

[0049] Based on the hierarchical aggregation effectiveness corresponding to each high-frequency time point within the target prediction period, the aggregation effectiveness for the prediction period is determined according to a preset aggregation relationship. The aggregation effectiveness for the prediction period is associated with the target prediction period identifier and is used to characterize the overall quality of each high-frequency remote sensing input participating in this current prediction throughout the entire target prediction period. When the economic activity index and hierarchical aggregation effectiveness of a new high-frequency time point are added within the target prediction period, the weights of the corresponding high-frequency remote sensing inputs are updated, and the current prediction results and the aggregation effectiveness for the prediction period are redefined.

[0050] The aggregation validity during the prediction period is compared with a preset prediction output threshold, and the output state of the current prediction result is determined based on the comparison result. When the aggregation validity during the prediction period meets the corresponding output condition, the current prediction result is output according to the corresponding output state; when the aggregation validity during the prediction period does not meet the corresponding output condition, the output state of the current prediction result is adjusted.

[0051] This embodiment determines the original observation effectiveness index based on remote sensing quality data, and uses this index to identify and correct low-quality remote sensing features. During the completion process, the original observation effectiveness index remains unchanged with the correction results, ensuring that the corrected remote sensing features correspond to their original observation quality. The observation effectiveness and completion performance of the source data are used to form the completion reliability index, which, together with the original observation effectiveness index, determines the effective quality constrained by the quality upper limit. The effective quality participates in determining the aggregation weights of economic activity spatial units and is synchronously aggregated hierarchically with the corrected remote sensing features, ensuring that the economic activity index and corresponding hierarchical aggregation effectiveness at each high-frequency time point reflect the observation quality at that time point. The hierarchical aggregation effectiveness corresponding to each high-frequency time point is further used to adjust the weights of the corresponding high-frequency remote sensing inputs, forming the prediction period aggregation effectiveness. This prediction period aggregation effectiveness controls the output state of the current prediction results, ensuring continuous transmission of remote sensing observation quality between feature correction, hierarchical aggregation, high-frequency input weighting, and prediction output.

[0052] Optionally, in this embodiment, the remote sensing quality data includes at least two of the following: cloud occlusion information, pixel availability information, time matching information, sensor quality information, anomaly deviation information, spatial processing error information, and historical prediction error information. Each piece of remote sensing quality data is associated with the corresponding economic activity spatial unit, time point, and remote sensing mode, and is used to describe the remote sensing observation state that forms the original remote sensing features.

[0053] Cloud cover information can be determined based on the cloud coverage ratio within the economic activity spatial unit, cloud mask classification results, or clear-sky pixel ratio. Pixel availability information is used to characterize the effective pixels participating in the calculation of raw remote sensing features, and can be determined based on the effective pixel ratio or the missing pixel ratio. When the effective pixel ratio and the missing pixel ratio are complementary, one of them is selected to participate in the calculation of the raw observation effectiveness index to avoid the same pixel availability factor being counted repeatedly.

[0054] Time matching information can be determined based on the time distance between the imaging time of the remote sensing data and the corresponding time point or target time window. The closer the imaging time is to the target time window, the higher the corresponding time matching quality. Sensor quality information can be determined by the quality label carried by the remote sensing product, the observation angle, the product grade, or the sensor's operating status.

[0055] Anomaly deviation information is used to characterize the degree of deviation of the current raw remote sensing features relative to the corresponding historical sequence or similar samples. Spatial processing error information can include errors generated in the remote sensing data during projection transformation, spatial registration, resampling, and pixel cropping. Historical prediction error information is used to characterize the error level when the historical features of the corresponding remote sensing mode are used in the prediction of target economic indicators.

[0056] Each remote sensing quality data point is assigned a corresponding quality component. For remote sensing quality data where an increase in value indicates improved observation quality, its normalized value is used as the corresponding quality component. For remote sensing quality data where an increase in value indicates decreased observation quality, inverse normalization is performed to ensure that each quality component satisfies the condition that a larger value indicates a more effective remote sensing observation. All normalized quality components fall within the same preset value range.

[0057] Each quality component is normalized, and then a weighted sum is performed on the normalized quality components according to the quality weights corresponding to each remote sensing quality data point, yielding the weighted sum result. For example, the original observation effectiveness index of enterprise i at time point t and remote sensing mode m can be determined according to the following formula:

[0058] in, Indicates the first The normalized quality components corresponding to the remote sensing quality data. Indicates the first The quality weights corresponding to the remote sensing quality data. This indicates the number of mass components involved in the calculation, and ; This indicates that the weighted summation result will be constrained to... and A limited preset index range.

[0059] In one example, the preset index range is [0, 1]. When the weighted summation result is less than 0, the original observation validity index is determined to be 0; when the weighted summation result is greater than 1, the original observation validity index is determined to be 1; when the weighted summation result is in [0, 1], the weighted summation result is determined to be the original observation validity index.

[0060] As another feasible approach, different remote sensing modes employ different combinations of mass components and mass weights. For example, the mass weights for optical remote sensing modes emphasize cloud obstruction information and pixel availability information, while the mass weights for synthetic aperture radar (SAR) remote sensing modes emphasize sensor quality information, anomaly deviation information, and spatial processing error information. The mass weights for each remote sensing mode are determined based on historical valid observation samples and remain unchanged when used to calculate the validity index of the same batch of raw observations.

[0061] Optionally, in this embodiment, the original remote sensing features X of economic activity spatial unit i at time point t and remote sensing mode m are considered. i,t,m The corresponding original observation validity index BDVI i,t,m The preset validity threshold τ corresponding to this remote sensing mode m Comparison. When BDVI i,t,m Below τ m At that time, X i,t,m Identified as a remote sensing feature to be corrected; when BDVI i,t,m Not less than τ m When, retain X i,t,m As the remote sensing features of this remote sensing mode at the corresponding time point, they are not included in the completion and fusion process.

[0062] Different remote sensing modes can be configured with preset validity thresholds τ. m The preset validity threshold can be determined based on the false positive rate, false negative rate, or subsequent prediction error of the corresponding remote sensing mode in the validation sample. For example, the preset validity threshold for nighttime light, optical reflectance, and surface temperature can be set to 0.70, or it can be adjusted according to the historical quality distribution of each remote sensing mode.

[0063] After determining the completion result corresponding to the remote sensing features to be corrected, based on BDVI i,t,m Determine the proportion α of original values ​​to be retained for the remote sensing features to be corrected. i,t,m Original value retention ratio α i,t,m The value of is in [0, 1], and varies with BDVI. i,t,m The value increases with the size of the original observation, so that the original observations with high validity retain more original information, while the original observations with low validity adopt more completion results.

[0064] In one implementation, the original observation validity index is directly used as the proportion of the original value retained, i.e.:

[0065] The remote sensing features to be corrected and the completion results are weighted and fused according to the original value retention ratio to obtain the corrected remote sensing features:

[0066] in, This represents the completion result corresponding to the remote sensing features to be corrected. This indicates the corrected remote sensing features.

[0067] As another feasible approach, a preset lower limit ε is set, and the proportion of original values ​​retained is determined according to the piecewise mapping relationship:

[0068] in, This indicates that the calculation results will be constrained to... Preset lower limit Not higher than the preset validity threshold For example, It can be set to 0.05. It can be set to 0.70; the actual value can also be determined based on the validation samples of the corresponding remote sensing mode.

[0069] When the remote sensing feature to be corrected is completely missing, marked as invalid by quality control, or BDVIi,t,m is not greater than ε, αi,t,m is set to zero, and the completion result is determined as the corrected remote sensing feature, that is:

[0070] Specifically, when the remote sensing feature to be corrected is completely missing or there are no valid pixels that can characterize the current observation state of the remote sensing mode, BDVI will be used. i,t,m Set to zero; when the remote sensing feature to be corrected is marked as invalid by quality control, BDVI will be set to zero. i,t,m Set to a value no higher than ε. In the above case, missing or invalid original remote sensing features are not used in the fusion.

[0071] In one example, the original observation validity index of nighttime lights for a certain economic activity spatial unit in the target month is 0.35. When the original observation validity index is directly used as the retention ratio of the original value, the fusion weight of the original nighttime light features is 0.35, and the corresponding fusion weight of the completed result is 0.65. Based on this, the corrected nighttime light features of the economic activity spatial unit in the target month are formed.

[0072] After weighted fusion, the original remote sensing features, the completed results, the proportion of original values ​​retained, and the corrected remote sensing features are saved separately, while keeping the original observation validity index unchanged. The corrected remote sensing features are not used to update or improve the original observation validity index, so that subsequent quality aggregation can still trace back to the original observation state corresponding to the corrected remote sensing features.

[0073] Optionally, in this embodiment, the supplementary source data includes at least one of the following: historical remote sensing features of the corresponding economic activity spatial unit, remote sensing features of other economic activity spatial units belonging to the same region and industry as the economic activity spatial unit, other remote sensing modal features of the economic activity spatial unit, and historical economic activity features. Historical economic activity features may include historical operating revenue, output, employment, or other operational data related to the state of economic activity.

[0074] When using historical remote sensing features of the corresponding economic activity spatial unit for completion, the time series of the target remote sensing mode is extracted from the highly effective remote sensing observations of the economic activity spatial unit before the target time point, and candidate time series completion values ​​are generated based on the time series. The completion method used can be time series interpolation, state-space model, time series prediction model, or completion method based on historical contemporaneous change relationships.

[0075] When using remote sensing features from other economic activity spatial units within the same region and industry for completion, other economic activity spatial units with valid remote sensing observations at the target time are selected, and reference samples are determined based on their similarity to the economic activity spatial unit to be completed in terms of industry category, spatial location, or historical changes. The remote sensing features of the reference samples at the target time are then weighted and averaged or processed to obtain candidate completion values ​​for similar samples.

[0076] When using features from other remote sensing modalities for completion, the completion model is trained on historical samples with highly effective observations of both the target and other remote sensing modalities. At the target time point, the features from other remote sensing modalities are input into the trained completion model to obtain cross-modal candidate completion values. The completion model can employ a multivariate regression model, a K-nearest neighbor model, a matrix completion model, a random forest model, or a gradient boosting tree model. For example, cross-modal candidate completion values ​​can be determined according to the following formula:

[0077] in, Represents spatial units of economic activities At time Next target remote sensing mode Cross-modal candidate completion values, Indicates remote sensing modes other than the target Other than One remote sensing modal feature, This indicates the historical characteristics of the corresponding spatial unit of economic activity. This refers to spatial units of other economic activities within the same region and industry at a given point in time. Statistical values ​​of remote sensing features , , and This represents the model parameters determined based on historically highly effective samples.

[0078] When using historical economic activity features for completion, a completion model is established based on the correspondence between historically highly effective remote sensing features and historical economic activity features. The historical economic activity features of the target economic activity spatial unit are then input into the completion model to obtain candidate completion values ​​for economic activities.

[0079] Regarding the first The method for identifying a source of completion involves obtaining the observation validity corresponding to the input data that constitutes that source. The accuracy rate of historical completion of the source of the completion. and the normalized value of cross-validation error The historical completion accuracy can be determined based on the numerical error or the consistency rate of change direction between the candidate completed value and the occluded true observation; the normalized cross-validation error is obtained by mapping the cross-validation error to a preset interval, and its inverse score is... .

[0080] The inverse score of the cross-validation error is determined based on observation validity, historical completion accuracy, and cross-validation error. Source rating for seed completion source For example, the source score can be determined according to the following formula:

[0081] The above coefficients can also be calibrated based on the relationship between each evaluation item and the completion error in the verification sample. When each evaluation item is normalized to [0, 1], the source score is correspondingly in [0, 1].

[0082] When only one candidate completion value is obtained, the combined weight of the completion source is set to 1, and the candidate completion value is determined as the completion result.

[0083] Optionally, when the source data for generating candidate completion values ​​does not exist, is marked as invalid by quality control, or does not meet the input conditions of the corresponding completion model, the source data is not used to generate candidate completion values, nor is the source score corresponding to the source data calculated.

[0084] When no source data exists to generate candidate completion values, no completion result is generated for the remote sensing feature to be corrected. The corresponding completion reliability index is set to zero, and the missing or invalid state of the remote sensing feature to be corrected is retained. The completion process is then re-executed once new historical remote sensing data, other remote sensing modal data, or remote sensing data for other economic activity spatial units within the same region and industry are subsequently obtained.

[0085] If at least one candidate completion value exists, calculate the sum of the source scores corresponding to each candidate completion value:

[0086] in, Represents spatial units of economic activities At time and remote sensing modes The sum of the source ratings below, Indicates the first The source score corresponding to each completed source data point This indicates the number of sources that participated in the completion.

[0087] when When it is greater than zero, the first The combined weights corresponding to the data sources for completion are determined according to the following formula:

[0088] when When the value is zero, the scores from each source are not normalized, the corresponding candidate completion values ​​are not used to generate completion results, and the completion reliability index is set to zero.

[0089] The candidate completion values ​​are weighted and fused according to the combined weights of each completion source to obtain the completion result:

[0090] in, Indicates the first Candidate completion values ​​generated from various completion sources Indicates the first The combined weights corresponding to the sources of the completion. This indicates the completion result obtained after fusion.

[0091] The source scores are weighted according to the weights of each combination to obtain the completion reliability index:

[0092] in, Represents spatial units of economic activities At time and remote sensing modes The completion reliability index corresponds to the completion result. The completion reliability index is saved in association with the completion result and used to subsequently determine the effective quality of the corresponding economic activity spatial unit.

[0093] In one example, the source scores for time series completion, same-region and same-industry completion, and cross-modal completion are 0.72, 0.81, and 0.63, respectively. The normalized combined weights are 0.333, 0.375, and 0.292, respectively, and the corresponding completion reliability index is 0.727.

[0094] For example, when the nighttime light characteristics of a certain economic activity spatial unit in the target month are identified as remote sensing characteristics to be corrected due to cloud cover, and the surface temperature characteristics, nitrogen dioxide characteristics, and nighttime light characteristics of economic activity spatial units in the same region and industry do not decrease synchronously, candidate completion values ​​are generated based on historical nighttime light sequences, other remote sensing modalities, and similar samples, respectively. Then, the combined weights are determined according to the source scores of each completion source to form the completion result.

[0095] In step S104 of this embodiment, the raw observation effectiveness index of each economic activity spatial unit at each time point and under each remote sensing mode is aggregated, and combined with the corresponding completion reliability index, the effective quality of each economic activity spatial unit at each time point is determined. If completion results exist, a preset quality upper limit is used to constrain the effective quality. For economic activity spatial unit i and time point t, the raw observation effectiveness index BDVI corresponding to each remote sensing mode m is obtained. i,t,m The original observation validity indices used in the aggregation are those determined and retained before correction, and are not recalculated based on the corrected remote sensing features.

[0096] Optionally, the original observation effectiveness index is weighted and aggregated according to the first mode weight corresponding to each remote sensing mode to obtain the basic effectiveness index:

[0097] in, Represents spatial units of economic activities At time The basic effectiveness index, Representing remote sensing modes The corresponding first mode weights, This indicates the number of remote sensing modes participating in the aggregation, and The first mode weights can be pre-configured or normalized based on the available remote sensing modes at the current time. If no corresponding observation data exists for a certain remote sensing mode at the current time, the first mode weights corresponding to that remote sensing mode are set to zero, and the remaining first mode weights are renormalized.

[0098] Optionally, before aggregating the raw observation validity indices corresponding to each remote sensing mode, it is determined whether each remote sensing mode has a valid raw observation validity index. When at least one remote sensing mode has a valid raw observation validity index, the weight of the first mode corresponding to the remote sensing mode that does not have a valid raw observation validity index is set to zero, and the weights of the remaining first modes are renormalized.

[0099] The renormalized first mode weights are determined according to the following formula: in, Representing remote sensing modes m The corresponding renormalized first mode weights, Representing remote sensing modes m The corresponding original first mode weights, Indicates validity marker; when remote sensing modality m When there is an effective raw observation validity index, Select 1 if the value is 1, otherwise select 0. M This indicates the number of remote sensing modes.

[0100] When no valid original observation validity index exists for any remote sensing mode, the basic validity index of the corresponding economic activity spatial unit at the current time point is set to zero, and the economic activity spatial unit is marked as having no valid original observation.

[0101] For the completed remote sensing modes, obtain the completion reliability index corresponding to the completion result. For remote sensing modes that have not generated completion results, the corresponding completion reliability index is set to zero. Thus, each remote sensing mode has a completion reliability index value that can participate in subsequent aggregation, where a zero value indicates that the corresponding remote sensing mode does not have a quality increment formed by the completion result at the current time point.

[0102] Based on the second mode weights corresponding to each remote sensing mode, the completion reliability indices are weighted and aggregated to obtain the comprehensive completion reliability index:

[0103] in, Represents spatial units of economic activities At time The comprehensive completion reliability index, Representing remote sensing modes The corresponding second modal weights, and When none of the remote sensing modalities generate completion results, each All are zero, resulting in a comprehensive reliability index. The corresponding value is zero.

[0104] As an feasible approach, the first modality weight and the second modality weight are configured separately to adapt to the original observation quality aggregation and the completion reliability aggregation, respectively; when using a uniform modality importance, the first modality weight and the second modality weight can also take the same value.

[0105] When determining the comprehensive completion reliability index, the completion reliability index corresponding to remote sensing modes without completion results is set to zero. When at least one remote sensing mode has completion results, the weights of the second mode corresponding to the remote sensing modes with completion results are normalized, and the completion reliability indices are aggregated based on the normalized second mode weights.

[0106] The normalized second mode weights are determined according to the following formula:

[0107] in, Representing remote sensing modes m The corresponding normalized second modal weights, Indicates the original second mode weights. Indicates the completion result marker; when remote sensing mode m When completion results exist, Select 1 if the value is 1, otherwise select 0.

[0108] When no completion results are available for any of the remote sensing modes, the overall completion reliability index is set to zero. When both the basic effectiveness index and the overall completion reliability index are zero, the effective quality of the corresponding economic activity spatial unit at the current time point is set to zero.

[0109] Determine the effective quality based on whether there is a completion result:

[0110]

[0111] Where Qeff represents the effective quality of the economic activity spatial unit at the corresponding time point, κ represents the completion gain coefficient, and Qcap represents the preset quality upper limit. The completion gain coefficient is used to limit the incremental contribution of the comprehensive completion reliability index to the effective quality, and it can be taken from 0.2 to 0.4. The preset quality upper limit is used to limit the quality increment brought about by completion reliability, and can be pre-configured into multiple levels according to the value range of the basic effectiveness index. The preset quality upper limit selected for the current economic activity spatial unit is not lower than its basic effectiveness index. For example, when the basic effectiveness index is lower than 0.70, the preset quality upper limit can be 0.70, or it can be set according to historical verification samples.

[0112] When completion results exist, if the sum of the basic validity index and the scaled comprehensive completion reliability index is not greater than a preset quality upper limit, the sum is determined as the valid quality; if the sum is greater than the preset quality upper limit, the preset quality upper limit is determined as the valid quality. If no completion results are generated, the comprehensive completion reliability index is zero, and the basic validity index is directly determined as the valid quality, without adjusting the basic validity index using the preset quality upper limit.

[0113] In step S105 of this embodiment, the aggregation weight of each economic activity spatial unit is determined based on the effective quality. The corrected remote sensing features and effective quality are then aggregated hierarchically according to the aggregation weights to obtain the economic activity index and hierarchical aggregation effectiveness. This step first fuses multiple remote sensing modalities of the same economic activity spatial unit at the same time point to form spatial unit-level remote sensing activity features, and then performs weighted aggregation of multiple economic activity spatial units according to industry grouping or regional grouping.

[0114] The corrected remote sensing features X of economic activity spatial unit i at time point t and remote sensing mode m i,t,m, obtain the raw observation effectiveness index (BDVI) corresponding to this remote sensing mode. i,t,m Complete the reliability index Rcomp i,t,m and historical prediction contribution score P i,m The historical prediction contribution score is used to characterize the degree of contribution of the corresponding remote sensing mode to the prediction of the target economic indicator in historical samples. It can be determined based on the prediction accuracy, coefficient of determination, direction judgment accuracy or the inverse score of prediction error of the validation set, and normalized to [0, 1].

[0115] Optionally, the modal score corresponding to each remote sensing mode is determined based on the product of the modal effective quality and the historical prediction contribution score:

[0116] Among them, A i,t,m Qmod represents the modal score of remote sensing mode m in economic activity spatial unit i and time point t. i,t,m P represents the modal effective mass. i,m This represents the historical prediction contribution score corresponding to remote sensing mode m. Modal effective quality is used to enable reliably completed remote sensing modes to participate in mode fusion in a constrained manner without updating the original observation effectiveness index.

[0117] In one implementation, the modal effective mass is determined according to the following relationship:

[0118]

[0119] Among them, κ m This represents the completion gain coefficient corresponding to remote sensing mode m. This indicates the upper limit of the quality that the remote sensing mode can achieve after completion. The upper limit of the quality selected for the current remote sensing mode is not lower than the original observation validity index corresponding to the remote sensing mode, so that the completion process will not reduce the observation quality that the remote sensing mode already has before completion.

[0120] For completely missing remote sensing modes, the original observation validity index is set to zero; when the completion reliability index is lower than the preset reliability threshold, the modal effective quality of the remote sensing mode is set to zero or it is not allowed to participate in modal fusion; when the completion reliability index is not lower than the preset reliability threshold, the remote sensing mode participates in modal fusion based on the modal effective quality.

[0121] In one example, the nighttime light mode of a certain economic activity spatial unit is completely missing at time t. Its original observation validity index is 0, the completion reliability index is 0.80, the completion gain coefficient is 0.50, and the modal quality upper limit is 0.60. Then the modal effective quality of this nighttime light mode is 0.40. When its historical prediction contribution score is 0.90, the corresponding modal score is 0.36.

[0122] The modal scores of various remote sensing modes for the same economic activity spatial unit at the same time point are normalized to obtain the modal fusion weights:

[0123] Wherein, represents the mode fusion weight corresponding to the remote sensing mode, represents any remote sensing mode participating in the fusion, represents the number of remote sensing modes participating in the fusion, and the sum of the mode fusion weights is 1. Remote sensing modes that have neither available original observations nor completion results that meet the preset reliability threshold at the current time point are not included in the normalization process, and their mode fusion weight is set to zero; for remote sensing modes that are completely missing but whose completion reliability index is not lower than the preset reliability threshold, the mode score is calculated based on the effective quality of the mode and participates in the normalization process.

[0124] Optionally, when determining the modality fusion weights, the sum of the modality scores corresponding to each remote sensing modality for the same economic activity spatial unit at the same time point is calculated:

[0125] in, Represents spatial units of economic activities i At time t The sum of the corresponding modal scores, Representing remote sensing modes m The corresponding modal score.

[0126] when When greater than zero, remote sensing mode m The corresponding modal fusion weights are determined according to the following formula:

[0127] when When the value is zero, the corresponding economic activity spatial unit is marked as having no effective remote sensing mode, and no remote sensing activity features of the economic activity spatial unit at the current time point are generated, and the economic activity spatial unit does not participate in the subsequent hierarchical aggregation at the current time point.

[0128] For modified remote sensing features with different units of measurement or ranges of variation, standardization can be performed first based on the historical sample mean and standard deviation of the corresponding remote sensing modality. Then, the standardized modified remote sensing features are weighted and fused according to the fusion weights of each modality to obtain remote sensing activity features at the spatial unit level of economic activity.

[0129] Here, represents the remote sensing activity characteristics of economic activity spatial units at a given point in time. When the original observation effectiveness index of a certain remote sensing mode decreases at the current point in time and there is no reliable complement, the mode fusion weight corresponding to that remote sensing mode decreases accordingly; when the remote sensing mode is completely missing but the complement reliability index reaches a preset reliability threshold, the remote sensing mode participates in fusion with a limited weight corresponding to its effective mode quality; when both its effective mode quality and historical prediction contribution score are high, the corresponding mode fusion weight increases.

[0130] After determining the characteristics of remote sensing activities, multiple economic activity spatial units are divided into corresponding groups g according to the industry or region to which they belong. For economic activity spatial unit i in group g, the size weight, sample representativeness weight, historical predictability weight, and effective quality corresponding to time point t are obtained respectively.

[0131] Size weights can be determined based on the operating revenue, asset size, production capacity, number of employees, or land area of ​​the enterprises corresponding to the economic activity spatial unit, and are normalized within the corresponding group. Sample representativeness weights characterize the representativeness of the economic activity spatial unit to its industry or region, and can be determined based on industry output share, regional economic scale share, sampling weight, or sample coverage. Historical predictability weights can be determined based on the historical prediction accuracy or the inverse fraction of historical prediction errors of the economic activity spatial unit. The corresponding aggregate weights are formed by enterprise size, sample representativeness, historical predictability, and observational effectiveness, and are normalized within the same group.

[0132] The aggregate score corresponding to the spatial unit of economic activity is determined by multiplying the size weight, sample representativeness weight, historical predictability weight, and effective quality:

[0133] in, Represents spatial units of economic activities In groups and time point The aggregated rating below Indicates size weight, Indicates the representativeness weight of the sample. Indicates the weight of historical predictability. Indicates effective quality.

[0134] Within group g, the aggregate scores corresponding to each economic activity spatial unit are normalized to obtain the aggregate weights:

[0135] in, Represents spatial units of economic activities In groups and time point Aggregate weights below Indicates grouping For any economic activity spatial unit in the data, the sum of the aggregation weights within the same group is 1.

[0136] When determining the aggregation weight of each economic activity spatial unit within a group, the sum of the aggregation scores of each economic activity spatial unit within the corresponding group is calculated:

[0137] in, Indicates group g at time point t The sum of the corresponding aggregated scores, This represents the aggregate score corresponding to spatial unit i of economic activity.

[0138] when When it is greater than zero, the spatial unit of economic activity i The corresponding aggregation weights are determined according to the following formula:

[0139] when When the value is zero, the corresponding group is marked as having no effective economic activity spatial unit, the economic activity index of the group at the current time is not generated, and the hierarchical aggregation validity of the group at the current time is set to zero.

[0140] Based on the aggregation weights corresponding to each economic activity spatial unit, the remote sensing activity characteristics are weighted and aggregated to obtain the economic activity index of group g at time t:

[0141] in, Indicates industry grouping or regional grouping At time The economic activity index is determined by the quality of the spatial units representing economic activity. The lower the effective quality of these units, the lower their aggregation score and weight, thus limiting the contribution of low-quality remote sensing samples to the economic activity index.

[0142] By applying the same aggregation weight to the effective quality of each economic activity spatial unit, the hierarchical aggregation effectiveness of group g at time point t is obtained:

[0143] in, Indices related to economic activity index The corresponding stratified aggregation validity. The economic activity index and the stratified aggregation validity are configured with the same group identifier and time identifier, which are used to represent the remote sensing economic activity status of the corresponding group and the overall quality of the sample that forms the economic activity index, respectively.

[0144] In step S106 of this embodiment, the economic activity index is used as the high-frequency remote sensing input of the mixed-frequency current prediction model. The weight of the high-frequency remote sensing input is adjusted according to the hierarchical aggregation effectiveness to obtain the current prediction result of the target economic indicator. Based on the comparison result of the aggregation effectiveness during the prediction period and the preset prediction output threshold, the output state of the current prediction result is controlled.

[0145] In this embodiment, the mixed-frequency current forecasting model employs a mixed-frequency data sampling model to map daily, weekly, ten-day, or monthly cumulative economic activity indices to monthly or quarterly target economic indicators. For grouping... Target forecast period The number of high-frequency time points participating in this prediction is denoted as , will the The economic activity index corresponding to each high-frequency time point is denoted as The effectiveness of hierarchical aggregation corresponding to this high-frequency time point is denoted as ,in, The high-frequency time point index within the target prediction period, with a value ranging from 1 to... .

[0146] Obtain the preset baseline input weights for each high-frequency time point in the mixing data sampling model. The baseline input weight corresponding to the h-th high-frequency time point is denoted as... This is used to characterize the time-series contribution of the economic activity index at this high-frequency point in time to the target economic indicator. The baseline input weights can be generated by a preset time-delay weight function, or they can be determined as model parameters based on historical training samples.

[0147] Calculate the product of each baseline input weight and the corresponding high-frequency time point's hierarchical aggregation effectiveness, and then calculate the sum of these products:

[0148] in, Let g represent the weighted normalized denominator of group g in the target prediction period T, and k represent any high-frequency time point participating in this prediction.

[0149] When D g,T When the value is greater than zero, the effective input weight corresponding to the h-th high-frequency time point is determined according to the following formula:

[0150] After expanding the denominator to normalize the weights, the effective input weights corresponding to the h-th high-frequency time point are expressed as:

[0151] in, This represents the effective input weight corresponding to the h-th high-frequency time point, and the sum of the effective input weights corresponding to all high-frequency time points is 1. When the effectiveness of hierarchical aggregation decreases, the effective input weight of the corresponding high-frequency time point decreases accordingly; when the effectiveness of hierarchical aggregation is zero, the effective input weight of the corresponding high-frequency time point is zero.

[0152] when When the result is equal to zero, the products are not normalized, no current prediction results for the current target prediction period are generated, and the current target prediction period is set to a paused output state.

[0153] Based on the effective input weights corresponding to each high-frequency time point, the corresponding economic activity index is input into the mixed-frequency data sampling model to obtain the current forecast result of the target economic indicator. For example, the quality-corrected high-frequency input can be formed first according to the following formula:

[0154] in, This represents the high-frequency input for quality correction of group g at the target prediction period T.

[0155] The quality-corrected high-frequency input is input into the mixing data sampling model to obtain the current prediction result:

[0156] in, This represents the current forecast result of the target economic indicator for group g within the target forecast period T. Represents the intercept parameter. This represents the model parameters corresponding to the high-frequency input for quality correction.

[0157] Optionally, if the mixed-frequency data sampling model is further configured with control variables, the current prediction result is determined according to the following formula:

[0158] in, δ represents the control variable, and δ represents the model parameter corresponding to the control variable.

[0159] Based on the effective input weights corresponding to each high-frequency time point, the hierarchical aggregation effectiveness corresponding to each high-frequency time point is weighted and aggregated to obtain the aggregation effectiveness for the prediction period:

[0160] in, This represents the aggregation validity of group g within the prediction period corresponding to the target prediction period T. This aggregation validity serves as the comparison object controlling the output state of the current prediction result. When the current prediction contains only one high-frequency time point, the hierarchical aggregation validity corresponding to that high-frequency time point is directly determined as the aggregation validity of the prediction period.

[0161] The aggregation validity during the prediction period is compared with a first prediction output threshold and a second prediction output threshold, wherein the second prediction output threshold is lower than the first prediction output threshold. The first and second prediction output thresholds can be determined based on the correspondence between the aggregation validity during the prediction period and the prediction error in historical prediction samples, and remain unchanged in the same prediction task.

[0162] When the aggregation validity of the forecast period is not lower than the first forecast output threshold, the current forecast results of the target economic indicators are output normally, and the current forecast results are associated with the corresponding group identifier and the target forecast period identifier.

[0163] When the aggregation validity during the prediction period is lower than the first prediction output threshold and the condition for pausing output is not met, the current prediction result is output according to the low confidence level, and the low confidence level is associated with the current prediction result.

[0164] The number of valid economic activity spatial units participating in the aggregation of the economic activity index within the target forecast period refers to the number of duplicate economic activity spatial units that form valid quality and actually participate in at least one of the aggregations of the economic activity index within the target forecast period.

[0165] When the aggregate validity of the forecast period falls below the second forecast output threshold, and the number of effective economic activity spatial units is less than a preset number, the output of the current forecast result is paused. While output is paused, the already formed economic activity index, the aggregate validity of the forecast period, and the current forecast result are retained, but this current forecast result is not used as the output result for the current target forecast period.

[0166] As new high-frequency remote sensing data arrives within the target prediction period, the economic activity index and hierarchical aggregation validity corresponding to the newly added high-frequency time points are redefined, and the effective input weights, current prediction results, and prediction period aggregation validity are updated. The output status of the current prediction results is then determined again based on the prediction period aggregation validity and the number of effective economic activity spatial units.

[0167] Another embodiment of the present invention relates to a real-time economic activity prediction system based on the effectiveness of remote sensing observations, such as... Figure 2 As shown, it includes: The validity determination module 20 is used to acquire multi-source remote sensing data and corresponding remote sensing quality data of multiple economic activity spatial units at multiple time points, extract the original remote sensing features of each economic activity spatial unit at each time point and in each remote sensing mode from the multi-source remote sensing data, and determine the original observation validity index corresponding to each original remote sensing feature based on the remote sensing quality data. The feature correction module 21 is used to determine the remote sensing feature to be corrected based on the original observation effectiveness index, determine the completion result corresponding to the remote sensing feature to be corrected, and fuse the remote sensing feature to be corrected and the completion result based on the original observation effectiveness index to obtain the corrected remote sensing feature, wherein the original observation effectiveness index is not updated with the corrected remote sensing feature. The reliability determination module 22 is used to determine the completion reliability index corresponding to the completion result based on the observation validity and completion performance of the completion source data used to generate the completion result; The quality determination module 23 is used to aggregate the original observation effectiveness index of each economic activity spatial unit at each time point and in each remote sensing mode, and combine it with the corresponding completion reliability index to determine the effective quality of each economic activity spatial unit at each time point, and to constrain the effective quality by using a preset quality upper limit when the completion result exists. The hierarchical aggregation module 24 is used to determine the aggregation weight of each economic activity spatial unit based on the effective quality, and to perform hierarchical aggregation on the modified remote sensing features and the effective quality according to the aggregation weight to obtain the economic activity index and the hierarchical aggregation effectiveness. The prediction control module 25 is used to take the economic activity index at each high-frequency time point as the high-frequency remote sensing input of the mixed-frequency real-time prediction model, adjust the weight of the corresponding high-frequency remote sensing input according to the hierarchical aggregation effectiveness corresponding to each high-frequency time point, and obtain the real-time prediction result of the target economic indicator; determine the aggregation effectiveness of the prediction period based on the hierarchical aggregation effectiveness corresponding to each high-frequency time point, and control the output state of the real-time prediction result based on the comparison result of the aggregation effectiveness of the prediction period and the preset prediction output threshold.

[0168] Optionally, in this embodiment, the remote sensing quality data includes at least two of the following: cloud occlusion information, pixel availability information, time matching information, sensor quality information, anomaly deviation information, spatial processing error information, and historical prediction error information. The pixel availability information is determined based on the proportion of valid pixels or the proportion of missing pixels. The validity determination module 20 includes: The first determination submodule is used to determine the quality components corresponding to each remote sensing quality data item; The first processing submodule is used to normalize each of the mass components and perform a weighted summation of the normalized mass components according to the mass weights corresponding to each remote sensing quality data to obtain a weighted summation result. The second processing submodule is used to constrain the weighted summation result to a preset index range to obtain the original observation validity index, wherein the sum of each quality weight is 1.

[0169] Optionally, in this embodiment, the feature correction module 21 includes: The second determining submodule is used to determine the original remote sensing features whose original observation validity index is lower than the preset validity threshold of the corresponding remote sensing mode as the remote sensing features to be corrected. The third determining submodule is used to determine the original value retention ratio based on the original observation validity index corresponding to the remote sensing feature to be corrected; The third processing submodule is used to perform weighted fusion of the remote sensing features to be corrected and the completion result according to the original value retention ratio; The fourth processing submodule is used to set the original value retention ratio to zero when the remote sensing feature to be corrected is completely missing, marked as invalid by quality control, or the corresponding original observation validity index is not greater than a preset lower limit, wherein the preset lower limit is not higher than the preset validity threshold.

[0170] Optionally, in this embodiment, the supplementary source data includes at least one of the following: historical remote sensing features of the corresponding economic activity spatial unit, remote sensing features of other economic activity spatial units belonging to the same region and industry as the corresponding economic activity spatial unit, other remote sensing modal features of the corresponding economic activity spatial unit, and historical economic activity features; the reliability determination module 22 includes: The fifth processing submodule is used to generate candidate completion values ​​based on the completion data from each source. The fourth determination submodule is used to determine the source score corresponding to each source data for completion based on the observation validity, historical completion accuracy, and inverse score of cross-validation error of each source data for completion. The fifth determination submodule is used to determine the corresponding combined weight based on the scores of each source. In the case that the supplementary source data includes one source, the combined weight corresponding to that source is set to 1. In the case that the supplementary source data includes multiple sources, the scores of each source are normalized to obtain the combined weight corresponding to each source. The sixth processing submodule is used to perform weighted fusion of each candidate completion value according to the combined weight to obtain the completion result, and to perform weighted fusion of each source score according to the combined weight to obtain the completion reliability index.

[0171] Optionally, in this embodiment, the quality determination module 23 includes: The seventh processing submodule is used to perform weighted aggregation of the original observation effectiveness indices of each economic activity spatial unit at each time point and under each remote sensing mode, so as to obtain the basic effectiveness index of the corresponding economic activity spatial unit at the corresponding time point. The eighth processing submodule is used to perform weighted aggregation of the completion reliability index corresponding to each remote sensing modality with completion results to obtain a comprehensive completion reliability index. The completion reliability index corresponding to the remote sensing modality without completion results is set to zero. The sixth determining submodule is used to calculate the sum of the comprehensive completion reliability index scaled by the completion gain coefficient and the basic effectiveness index when at least one completion result exists, and determine the smaller value between the sum and the preset quality upper limit as the effective quality; when no completion result exists, the basic effectiveness index is determined as the effective quality.

[0172] Optionally, in this embodiment, the hierarchical aggregation module 24 includes: The seventh determination submodule is used to determine the mode fusion weight based on the mode effective quality and historical prediction contribution corresponding to each remote sensing mode, and to fuse the corrected remote sensing features of each economic activity spatial unit under each remote sensing mode according to the mode fusion weight to obtain the remote sensing activity features of each economic activity spatial unit. The mode effective quality is determined based on the original observation effectiveness index of the corresponding remote sensing mode and the completion reliability index when completion results exist. The eighth determination submodule is used to determine the aggregate score corresponding to each economic activity spatial unit based on the scale weight, sample representativeness weight, historical predictability weight and the effective quality corresponding to each economic activity spatial unit. The ninth processing submodule is used to normalize the aggregated score within the industry group or regional group to which each economic activity spatial unit belongs, so as to obtain the aggregated weight corresponding to each economic activity spatial unit. The tenth processing submodule is used to perform weighted aggregation of the remote sensing activity features of each economic activity spatial unit according to the aggregation weight to obtain the economic activity index, and to perform weighted aggregation of the effective quality of each economic activity spatial unit according to the aggregation weight to obtain the hierarchical aggregation effectiveness.

[0173] Optionally, in this embodiment, the mixing current prediction model is a mixing data sampling model; the prediction control module 25 includes: The acquisition submodule is used to acquire the preset benchmark input weights of the mixed-frequency data sampling model for each high-frequency time point in the economic activity index; The ninth determination submodule is used to determine the product of each benchmark input weight and the corresponding high-frequency time point hierarchical aggregation validity, and to normalize each product to obtain the valid input weight corresponding to each high-frequency time point; The prediction submodule is used to input the daily, weekly, ten-day or monthly cumulative economic activity index into the mixed-frequency data sampling model according to each effective input weight, so as to predict the target economic indicators for the month or quarter. The aggregation validity determination submodule is used to perform weighted aggregation of the hierarchical aggregation validity corresponding to each high-frequency time point according to the effective input weights corresponding to each high-frequency time point, so as to obtain the aggregation validity for the prediction period. The first output submodule is used to output the current prediction result normally when the aggregation validity during the prediction period is not lower than the first prediction output threshold. The second output submodule is used to output the current prediction result at a low confidence level when the aggregation validity during the prediction period is lower than the first prediction output threshold and the pause output condition is not met. The third output submodule is used to pause the output of the current prediction result when the aggregation validity during the prediction period is lower than the second prediction output threshold and the number of effective economic activity spatial units is lower than a preset number, wherein the second prediction output threshold is lower than the first prediction output threshold.

[0174] Another embodiment of the present invention relates to an electronic device, such as Figure 3 As shown, it includes: at least one processor 301; and a memory 302 communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions executable by the at least one processor 301, the instructions being executed by the at least one processor 301 to enable the at least one processor 301 to perform the real-time economic activity prediction method based on the effectiveness of remote sensing observations in the above embodiments.

[0175] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0176] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0177] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0178] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0179] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0180] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for real-time forecasting of economic activities based on the effectiveness of remote sensing observations, characterized in that, include: Acquire multi-source remote sensing data and corresponding remote sensing quality data of multiple economic activity spatial units at multiple time points. Extract the original remote sensing features of each economic activity spatial unit at each time point and in each remote sensing mode from the multi-source remote sensing data. Determine the original observation effectiveness index corresponding to each original remote sensing feature based on the remote sensing quality data. Based on the original observation effectiveness index, the remote sensing features to be corrected are determined, the completion results corresponding to the remote sensing features to be corrected are determined, and the remote sensing features to be corrected and the completion results are fused based on the original observation effectiveness index to obtain the corrected remote sensing features. The original observation effectiveness index is not updated with the corrected remote sensing features. Based on the observation validity and completion performance of the completion source data used to generate the completion result, the completion reliability index corresponding to the completion result is determined; The original observation effectiveness index of each economic activity spatial unit at each time point and under each remote sensing mode is aggregated, and combined with the corresponding completion reliability index, the effective quality of each economic activity spatial unit at each time point is determined. In the case of the completion result, the effective quality is constrained by a preset quality upper limit. The aggregation weight of each economic activity spatial unit is determined based on the effective quality. The modified remote sensing features and the effective quality are then aggregated in layers according to the aggregation weight to obtain the economic activity index and the layered aggregation effectiveness. The economic activity index at each high-frequency time point is used as the high-frequency remote sensing input to the mixed-frequency current prediction model. The weights of the corresponding high-frequency remote sensing inputs are adjusted according to the hierarchical aggregation effectiveness at each high-frequency time point to obtain the current prediction result of the target economic indicator. The aggregation effectiveness of the prediction period is determined based on the hierarchical aggregation effectiveness at each high-frequency time point, and the output state of the current prediction result is controlled based on the comparison result between the aggregation effectiveness of the prediction period and the preset prediction output threshold.

2. The method according to claim 1, characterized in that, The remote sensing quality data includes at least two of the following: cloud occlusion information, pixel availability information, temporal matching information, sensor quality information, anomaly deviation information, spatial processing error information, and historical prediction error information. The pixel availability information is determined based on the proportion of valid pixels or the proportion of missing pixels. The determination of the original observation effectiveness index corresponding to each original remote sensing feature based on the remote sensing quality data includes: Determine the mass components corresponding to each remote sensing quality data point; Each of the aforementioned quality components is normalized, and the normalized quality components are weighted and summed according to the quality weights corresponding to the remote sensing quality data to obtain the weighted summation result. The weighted summation result is constrained to a preset index range to obtain the original observation validity index, wherein the sum of each quality weight is 1.

3. The method according to claim 1, characterized in that, The process of determining the remote sensing features to be corrected based on the original observation effectiveness index includes: The original remote sensing features whose original observation effectiveness index is lower than the preset effectiveness threshold of the corresponding remote sensing mode are identified as the remote sensing features to be corrected. The process of fusing the remote sensing features to be corrected and the completion result based on the original observation effectiveness index includes: The proportion of original values ​​to be retained is determined based on the original observation validity index corresponding to the remote sensing features to be corrected. The remote sensing features to be corrected and the completion results are weighted and fused according to the original value retention ratio; When the remote sensing feature to be corrected is completely missing, is marked as invalid by quality control, or the corresponding original observation validity index is not greater than a preset lower limit, the original value retention ratio is set to zero, wherein the preset lower limit is not higher than the preset validity threshold.

4. The method according to claim 3, characterized in that, The supplementary source data includes at least one of the following: historical remote sensing features of the corresponding economic activity spatial unit, remote sensing features of other economic activity spatial units belonging to the same region and industry as the corresponding economic activity spatial unit, other remote sensing modal features of the corresponding economic activity spatial unit, and historical economic activity features. The process of determining the completion result corresponding to the remote sensing feature to be corrected and the completion reliability index corresponding to the completion result includes: Candidate completion values ​​are generated based on the data from each completion source; Based on the observation validity, historical completion accuracy, and inverse score of cross-validation error of each completion source data, the source score corresponding to each completion source data is determined; The corresponding combined weight is determined based on the scores of each source. When the supplementary source data includes one source, the combined weight corresponding to that source is set to 1. When the supplementary source data includes multiple sources, the scores of each source are normalized to obtain the combined weight corresponding to each source. The candidate completion values ​​are weighted and fused according to the combined weights to obtain the completion result, and the scores from each source are weighted according to the combined weights to obtain the completion reliability index.

5. The method according to claim 4, characterized in that, Determining the effective quality of each economic activity spatial unit at each point in time includes: We weighted and aggregated the original observation effectiveness indices of each economic activity spatial unit at each time point and under each remote sensing mode to obtain the basic effectiveness index of the corresponding economic activity spatial unit at the corresponding time point. The completion reliability index corresponding to each remote sensing mode with completion results is weighted and aggregated to obtain the comprehensive completion reliability index. The completion reliability index corresponding to the remote sensing mode without completion results is set to zero. If at least one completion result exists, the sum of the comprehensive completion reliability index scaled by the completion gain coefficient and the basic effectiveness index is calculated, and the smaller value between the sum and the preset quality upper limit is determined as the effective quality; if no completion result exists, the basic effectiveness index is determined as the effective quality.

6. The method according to claim 5, characterized in that, The step of determining the aggregation weight of each economic activity spatial unit based on the effective quality, and performing hierarchical aggregation of the modified remote sensing features and the effective quality according to the aggregation weight, includes: The modal fusion weights are determined based on the modal effective quality and historical prediction contribution of each remote sensing modality. The modified remote sensing features of each economic activity spatial unit under each remote sensing modality are then fused according to the modal fusion weights to obtain the remote sensing activity features of each economic activity spatial unit. The modal effective quality is determined based on the original observation effectiveness index of the corresponding remote sensing modality and the completion reliability index when completion results exist. Based on the scale weight, sample representativeness weight, historical predictability weight, and effective quality of each economic activity spatial unit, the aggregate score corresponding to each economic activity spatial unit is determined. Within the industry group or regional group to which each economic activity spatial unit belongs, the aggregated score is normalized to obtain the aggregated weight corresponding to each economic activity spatial unit. The remote sensing activity characteristics of each economic activity spatial unit are weighted and aggregated according to the aggregation weight to obtain the economic activity index. The effective quality of each economic activity spatial unit is weighted and aggregated according to the aggregation weight to obtain the hierarchical aggregation effectiveness.

7. The method according to claim 6, characterized in that, The frequency mixing current prediction model is a frequency mixing data sampling model; The step of adjusting the weights of the high-frequency remote sensing inputs based on the effectiveness of the hierarchical aggregation includes: Obtain the preset baseline input weights of the mixed-frequency data sampling model for each high-frequency time point in the economic activity index; The product of each baseline input weight and the corresponding high-frequency time point hierarchical aggregation effectiveness is determined, and each product is normalized to obtain the effective input weights corresponding to each high-frequency time point. The daily, weekly, ten-day, or monthly cumulative economic activity indices are input into the mixed-frequency data sampling model according to the effective input weights to predict the target economic indicators for the month or quarter, and to obtain the current prediction results. Based on the effective input weights corresponding to each high-frequency time point, the hierarchical aggregation effectiveness corresponding to each high-frequency time point is weighted and aggregated to obtain the aggregation effectiveness for the prediction period. When the aggregation validity during the prediction period is not lower than the first prediction output threshold, the current prediction result is output normally. When the aggregation validity during the prediction period is lower than the first prediction output threshold and the pause output condition is not met, the current prediction result is output according to the low confidence level. When the aggregation validity during the prediction period is lower than the second prediction output threshold and the number of effective economic activity spatial units is lower than a preset number, the output of the current prediction result is paused, wherein the second prediction output threshold is lower than the first prediction output threshold.

8. A real-time economic activity forecasting system based on the effectiveness of remote sensing observation, characterized in that, include: The validity determination module is used to acquire multi-source remote sensing data and corresponding remote sensing quality data of multiple economic activity spatial units at multiple time points, extract the original remote sensing features of each economic activity spatial unit at each time point and in each remote sensing mode from the multi-source remote sensing data, and determine the original observation validity index corresponding to each original remote sensing feature based on the remote sensing quality data. The feature correction module is used to determine the remote sensing feature to be corrected based on the original observation effectiveness index, determine the completion result corresponding to the remote sensing feature to be corrected, and fuse the remote sensing feature to be corrected and the completion result based on the original observation effectiveness index to obtain the corrected remote sensing feature, wherein the original observation effectiveness index is not updated with the corrected remote sensing feature. The reliability determination module is used to determine the completion reliability index corresponding to the completion result based on the observation validity and completion performance of the completion source data used to generate the completion result; The quality determination module is used to aggregate the original observation effectiveness index of each economic activity spatial unit at each time point and in each remote sensing mode, and combine it with the corresponding completion reliability index to determine the effective quality of each economic activity spatial unit at each time point, and constrain the effective quality by using a preset quality upper limit when the completion result exists. The hierarchical aggregation module is used to determine the aggregation weight of each economic activity spatial unit based on the effective quality, and to perform hierarchical aggregation on the modified remote sensing features and the effective quality according to the aggregation weight to obtain the economic activity index and the hierarchical aggregation effectiveness. The prediction control module is used to take the economic activity index at each high-frequency time point as the high-frequency remote sensing input of the mixed-frequency real-time prediction model, adjust the weight of the corresponding high-frequency remote sensing input according to the hierarchical aggregation effectiveness corresponding to each high-frequency time point, and obtain the real-time prediction result of the target economic indicator; determine the aggregation effectiveness of the prediction period based on the hierarchical aggregation effectiveness corresponding to each high-frequency time point, and control the output state of the real-time prediction result based on the comparison result of the aggregation effectiveness of the prediction period and the preset prediction output threshold.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images