Icing weather time series prediction method and system based on spatiotemporal generative network

Abstract

The application relates to the technical field of icing prediction, in particular to an icing weather time series prediction method and system based on a space-time generation network. The method comprises the following steps: an airborne terminal receives in-situ data; the airborne terminal performs a light compression and packaging operation on the in-situ data to generate a light file, and sends the light file to a ground terminal; wherein the following steps are included: denoising processing is performed on the in-situ data to obtain denoised data; the denoised data is compressed by using a target compression scale to make a light file; the ground terminal receives non-in-situ data; the ground terminal predicts a first icing probability of an airplane in a future period of time by using the non-in-situ data and the light file through a time series prediction model; and the ground terminal sends the icing probability to the airborne terminal. The application can greatly reduce the data fusion difficulty of a model input end.

Description

Technical Field

[0001] This invention relates to the field of icing prediction technology, and in particular to a method and system for predicting icing meteorological time series based on spatiotemporal generation networks. Background Technology

[0002] Drones are increasingly used in low-altitude cloud penetration operations, aerial surveying, logistics transportation, and power line inspection. However, wing icing is a core hidden danger threatening drone flight safety, which can lead to reduced lift, loss of control, and even crashes in severe cases. Therefore, accurate and real-time forecasting of drone icing weather is of great significance for ensuring the safety of drones flying at low altitudes.

[0003] Currently, UAV icing weather forecasting mainly relies on three types of technologies: First, traditional numerical weather models, such as the WRF (Weather Research and Forecasting Model), which achieve weather forecasting by solving atmospheric dynamic equations, but suffer from low resolution and long computation time, making them unsuitable for the real-time flight decision-making needs of UAVs; second, conventional machine learning or deep learning models, which are mostly based on discriminative classification and can only determine whether icing has occurred, lacking the ability to predict the generation and evolution of icing weather spatiotemporal sequences, and failing to provide key information such as icing intensity, location, and evolution path; and third, single-source data prediction, which uses only radar, satellite, or numerical model data without integrating UAV onboard in-situ sensing data, such as onboard cloud particle detection data, real-time images of wing icing, and flight status parameters, resulting in insufficient prediction accuracy, especially in small sample scenarios such as extreme icing, where the model has poor robustness.

[0004] In response, traditional technologies have proposed some solutions that utilize models to improve the efficiency of aircraft icing prediction.

[0005] For example, patent application CN121743870A proposes a training method, prediction method, product, device, and medium for an aircraft icing model. The method includes: acquiring aircraft icing observation data and historical weather data; performing spatiotemporal matching of the aircraft icing observation data and grid points based on the spatial and temporal information of the aircraft icing observation data, as well as the horizontal grid, vertical layering, and temporal dimension of the grid points, to obtain spatiotemporally matched grid points; calculating the feature data of the spatiotemporally matched grid points using an icing scheme to obtain the icing index of the spatiotemporally matched grid points; constructing a training dataset using the feature data corresponding to multiple spatiotemporally matched grid points, the icing index, and the observation labels of the matched aircraft icing observation data; and training the model using the training dataset to obtain an aircraft icing model.

[0006] For example, patent application CN121561579A proposes a fixed-wing aircraft natural icing prediction system based on prediction models and big data analysis. This system includes a multi-source data acquisition module, a data processing module, a feature extraction module, an icing case database, a prediction model module, a prediction analysis module, and an early warning result output module. The data processing module preprocesses and cleans the collected raw data. The feature extraction module extracts key feature variables that significantly influence aircraft natural icing from the processed data. The prediction model module uses the extracted key feature variables as input and combines them with big data analysis to construct a prediction model. The prediction analysis module inputs the currently collected multi-source data into the trained prediction model for analysis.

[0007] For example, patent application CN117648850A proposes an aircraft icing simulation system based on a neural network training method. This system can predict the icing pattern of the wing / tail based on icing environment parameters and flight parameters, and predict the actual icing state of the aircraft in real time. Taking experimental / calculated icing patterns as the research object, the system first processes the icing pattern to obtain a single-value icing height curve distributed along the airfoil, and then parameterizes the curve using a Fourier series expansion. Next, a BP neural network prediction model is established, with icing environment parameters and flight state parameters as inputs, and Fourier coefficients of the icing pattern as outputs, obtaining the weights and biases between the input and output. Based on the predicted coefficients, the icing height curve can be obtained. Finally, interpolation is used to obtain the icing height at each position on the airfoil, thus reconstructing the actual icing pattern.

[0008] However, the above methods still suffer from very limited application scenarios. Therefore, there is an urgent need for a more scenario-adaptable aircraft icing prediction method. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for predicting icing weather time series based on spatiotemporal generation networks, which partially solves or alleviates the above-mentioned shortcomings in the prior art. It can solve the technical problems of low accuracy, poor real-time performance, single output, insufficient spatiotemporal modeling capability, and poor adaptability to small sample scenarios in existing UAV icing weather prediction. The invention provides a method and system for predicting icing weather time series based on spatiotemporal generation networks, which realizes high-precision, real-time, generative, and multi-dimensional UAV icing risk prediction, and provides reliable decision support for UAV flight safety.

[0010] To solve the aforementioned technical problems, the present invention specifically adopts the following technical solution: A first aspect of the present invention provides a method for predicting icing weather time series based on a spatiotemporal generation network, the aircraft comprising: an airborne terminal, wherein the airborne terminal is capable of communicating with a ground terminal located at a control center, the method comprising the steps of: S101, the airborne terminal receives in-situ data, which is monitoring data directly collected by the airborne terminal; S102, the airborne terminal performs a lightweight compression and packaging operation on the in-situ data to generate a lightweight file, and sends the lightweight file to the ground terminal; wherein, S102 includes the following steps: The in-situ data is subjected to denoising processing to obtain denoised data; The denoised data is compressed using a target compression scale to create the lightweight file; S103, the ground terminal receives non-in-situ data, the non-in-situ data including monitoring data acquired by other measuring devices located outside the airborne terminal; S104, the ground terminal uses the non-in-situ data and the lightweight file to predict the first icing probability of the aircraft in the future period through a time-series prediction model; S105, the ground terminal sends the icing probability to the airborne terminal.

[0011] In some embodiments, the steps further include: The airborne terminal selects at least one type of data from the in-situ data as pre-inspection data; Predict the pre-inspection risk of the aircraft in the future based on the pre-inspection data; The target compression scale is updated or adjusted based on the pre-inspection risks.

[0012] In some embodiments, the steps further include: When the pre-detection risk exceeds the set first risk threshold, the following steps are executed: Set the upper limit of the target compression scale to the first scale; The airborne terminal is used as the main decision-making terminal. At this time, the airborne terminal outputs the second icing probability for the future time period through the local model based on the corresponding lightweight file. Alternatively, the method may further include the steps of: When the pre-detection risk is less than the set first risk threshold, the following steps are executed: Set the upper limit of the target compression scale to the second scale, and the second scale is greater than the first scale; The ground terminal will still be used as the main decision-making terminal, and S103-S105 will be executed at this time.

[0013] In some embodiments, the steps further include: Identify the weather category of the aircraft's flight area; Determine whether the meteorological category is a rare category, wherein the rare category includes at least one type: freezing rain, mixed phase ice crystals; If so, the ground terminal will still be used as the primary decision-making terminal, and the following steps will be performed: The observation period is set based on the rare category and the current communication pressure. The observation period is used to limit the amount of data that the lightweight file is allowed to be transmitted in a single transmission window.

[0014] In some embodiments, the in-situ data includes at least one of the following: cloud particle data, flight status parameters, and wing icing images.

[0015] In some embodiments, the other measuring devices include at least one: a satellite or a probe aircraft.

[0016] In some embodiments, the off-site data includes at least one of the following: cloud top temperature, liquid water content, and median volume diameter.

[0017] A second aspect of the present invention provides an icing weather time series prediction system based on a spatiotemporal generation network, the aircraft comprising: an airborne terminal, wherein the airborne terminal is capable of communicating with a ground terminal located at a control center, including: The in-situ module receives in-situ data from the airborne terminal, and the in-situ data is monitoring data directly collected by the airborne terminal; A compression module, performed by the airborne terminal, performs lightweight compression and packaging operations on the in-situ data to generate a lightweight file, and sends the lightweight file to the ground terminal; wherein, the compression module includes: A denoising unit is used to perform denoising processing on the in-situ data to obtain denoised data; A compression unit is used to compress the denoised data using a target compression scale to create the lightweight file; The off-site module receives off-site data from the ground terminal, and the off-site data includes monitoring data acquired by other measuring devices located outside the airborne terminal; The prediction module, constructed by the ground terminal, uses the non-in-situ data and the lightweight file to predict the first icing probability of the aircraft in the future time period through a time-series prediction model. The feedback module sends the icing probability from the ground terminal to the airborne terminal.

[0018] In some embodiments, it also includes: The pre-inspection type selection module allows the airborne terminal to select at least one type of data from the in-situ data as pre-inspection data. The prediction module is used to predict the pre-inspection risk of the aircraft in the future based on the pre-inspection data. The compression scale update module is used to update or adjust the target compression scale based on the pre-inspection risk.

[0019] In some embodiments, it also includes: The first execution module is used to execute the following when the pre-detection risk is greater than a set first risk threshold: Set the upper limit of the target compression scale to the first scale; The airborne terminal is used as the main decision-making terminal. At this time, the airborne terminal outputs the second icing probability for the future time period through the local model based on the corresponding lightweight file. Alternatively, the system may also include: The second execution module is used to execute the following when the pre-detection risk is less than a set first risk threshold: Set the upper limit of the target compression scale to the second scale, and the second scale is greater than the first scale; The ground terminal will still be used as the main decision-making terminal, and the system will return to the non-in-situ module.

[0020] Beneficial technical effects: (1) The airborne end only performs two simple operations: denoising and compression, without involving complex model reasoning. Instead, the high-intensity non-in-situ data analysis and time series prediction tasks are distributed to the ground end. With the help of this invention, the contradiction between data processing and security can be coordinated through the collaboration between the airborne end and the ground end. This avoids the icing prediction calculation from putting a huge pressure on the aircraft's limited embedded computing resources and even affecting the main task of the flight control system.

[0021] (2) When the pre-inspection risk is high, in order to make decisions as soon as possible and avoid further expansion of the risk, a smaller target compression scale can be applied to retain more details, so as to ensure that the airborne end can make more accurate judgments directly based on the limited in-situ data more quickly. When the pre-inspection risk is low, a larger compression scale can be allowed to save the computing power of the airborne end. At this time, the icing risk prediction can be mainly performed by the ground end based on non-in-situ data and lightweight files.

[0022] In other words, this application proposes a multi-terminal collaborative data processing strategy based on actual icing risk, which can quickly determine the leading party in data processing with lower pre-inspection costs, thereby achieving a better balance between resource efficiency and security.

[0023] (3) For rare weather categories, this invention proposes a restrictive end-to-end switching mechanism for icing prediction. Specifically, this invention selects to utilize the relatively abundant computing resources on the ground end (including a complete historical database, expert knowledge base, etc.), while limiting the length of a single data transmission, to ensure that the ground end can make relatively accurate predictions for rare weather conditions in a short period of time. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. The elements or parts in the drawings are not necessarily drawn to scale. Obviously, the drawings described below are some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0025] Figure 1 A flowchart illustrating the icing meteorological time series prediction method based on spatiotemporal generation network provided by the present invention; Figure 2 A schematic block diagram of the structure of a computer device provided by the present invention; Figure 3 This is a schematic diagram of the structure of the icing meteorological time series prediction system based on spatiotemporal generation network provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] In this document, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no specific meaning in themselves. Therefore, "module," "part," or "unit" may be used interchangeably.

[0028] In this document, the terms "upper," "lower," "inner," "outer," "front," "rear," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the present invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0029] In this document, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0030] In this document, "and / or" includes any and all combinations of one or more of the listed related items.

[0031] In this article, "multiple" means two or more, that is, it includes two, three, four, five, etc.

[0032] As used in this specification, the term "about" typically means + / -5% of the value, more typically + / -4% of the value, more typically + / -3% of the value, more typically + / -2% of the value, even more typically + / -1% of the value, and even more typically + / -0.5% of the value.

[0033] In this specification, certain embodiments may be disclosed in a range-bound format. It should be understood that this "range-bound" description is merely for convenience and brevity and should not be construed as a rigid limitation on the disclosed range. Therefore, the description of a range should be considered as having specifically disclosed all possible subranges and the individual numerical values ​​within those ranges. For example, a description of the range 1-6 should be considered as having specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., and the individual numbers within those ranges, such as 1, 2, 3, 4, 5, and 6. This rule applies regardless of the breadth of the range.

[0034] Example 1 The applicant noted that aircraft icing is influenced by a large number of real-world parameters, including temperature, humidity, wind speed, liquid water content, and median volume diameter (MVD) of water droplets. These data come from various data sources, such as satellites, radars, ground weather stations, radiosondes, and lidar (LiDAR), with different sampling frequencies, spatial grids, and temporal alignment methods. Unifying these data into time-series samples usable in the model requires extensive alignment and interpolation work.

[0035] Furthermore, an icing event may be caused by the slow accumulation of microclimate conditions over several hours or even longer (such as the continuous impact of supercooled water droplets). However, sudden changes in temperature or wind speed (such as a sudden temperature rise or strong wind shear) can also cause the icing to melt or detach rapidly, and these situations can lead to an explosion in the computational workload of icing prediction.

[0036] For this, please see Figure 1 This invention provides a method for predicting icing weather time series based on a spatiotemporal generation network. The aircraft includes an airborne terminal, which is capable of communicating with a ground terminal located in a control center. The method includes the following steps: S101, the airborne terminal receives in-situ data, which is monitoring data directly collected by the airborne terminal; S102, the airborne terminal performs a lightweight compression and packaging operation on the in-situ data to generate a lightweight file, and sends the lightweight file to the ground terminal; wherein, S102 includes the following steps: The in-situ data is subjected to denoising processing to obtain denoised data; The denoised data is compressed using a target compression scale to create the lightweight file; S103, the ground terminal receives non-in-situ data, the non-in-situ data including monitoring data acquired by other measuring devices located outside the airborne terminal; S104, the ground terminal uses the non-in-situ data and the lightweight file to predict the first icing probability of the aircraft in the future period through a time-series prediction model; S105, the ground terminal sends the icing probability to the airborne terminal.

[0037] In-situ data refers to data that can be directly obtained from the airborne terminal.

[0038] Among these methods, filtering algorithms can be used to denoise the in-situ data; differential coding can be used to compress the in-situ data.

[0039] In some embodiments, the time series prediction model is preferably a Transformer model. The input data of the time series prediction model consists of non-in-situ data and lightweight files, and the output data is the icing probability.

[0040] In some embodiments, the time-series prediction model can also output icing evolution data of the aircraft in future periods (such as icing evolution path, direction, etc.).

[0041] In this embodiment, the airborne terminal only performs two simple operations: denoising and compression, without involving complex model reasoning. Instead, the high-intensity off-site data analysis and time-series prediction tasks are distributed to the ground terminal. With the help of this invention, the contradiction between data processing pressure and security can be coordinated through the collaboration between the airborne and ground terminals. This avoids the icing prediction calculation from putting enormous pressure on the aircraft's limited embedded computing resources and even affecting the main tasks of the flight control system.

[0042] In some embodiments, the steps further include: The airborne terminal selects at least one type of data from the in-situ data as pre-inspection data; for example, it can randomly select a portion of the in-situ data as pre-inspection data. Predict the pre-inspection risk of the aircraft in the future based on the pre-inspection data; The target compression scale is updated or adjusted based on the pre-inspection risks.

[0043] For example, if the ice layer on a wing icing image (i.e., a type of pre-inspection data) suddenly becomes very thick, or the temperature suddenly drops sharply, it initially indicates a high risk of pre-inspection.

[0044] In some embodiments, if the pre-detection risk indicates a high risk of icing, the amount of transmitted data may be appropriately compressed to provide icing prediction feedback as quickly as possible through high-speed transmission and high-speed computation, so that the aircraft can quickly implement anti-icing measures based on the feedback.

[0045] In some embodiments, if multiple types of data are selected as pre-inspection data, the pre-inspection risk can be predicted based on the sum of the risk levels corresponding to each selected type of pre-inspection data.

[0046] For example, the airborne terminal selects three types of data (a, b, and c) from the in-situ data as pre-inspection data. Since a, b, and c correspond to risk levels 1, 2, and 4 respectively, the predicted pre-inspection risk is 7.

[0047] In other embodiments, if multiple types of data are selected as pre-inspection data, the pre-inspection risk can be predicted based on the weighted sum of the pre-inspection data selected for each type.

[0048] For example, the airborne terminal selects three types of data (a, b, and c) from the in-situ data as pre-inspection data. The risk levels corresponding to a, b, and c are 1, 2, and 4, respectively; the weighting coefficients for a, b, and c are 0.2, 0.3, and 0.5, respectively. Therefore, the final calculated pre-inspection risk is 1×0.2 + 2×0.3 + 4×0.5 = 2.8.

[0049] The weighting coefficients can be preset by engineers based on the actual situation.

[0050] For example, different types of in-situ data have varying degrees of influence on the final freezing process; for instance, the median volume diameter (MVD) of a water droplet has a significant impact on the final freezing process. The greater the influence of the in-situ data, the larger its weighting coefficient.

[0051] For example, different types of in-situ data have different confidence levels; for instance, airspeed has a higher confidence level. In-situ data with higher confidence levels have larger weighting coefficients.

[0052] In this embodiment, by selecting at least one type of in-situ data as pre-inspection data, the pre-inspection can be made more lightweight, thereby reserving valuable computing resources for critical tasks.

[0053] In some embodiments, the steps further include: When the pre-detection risk exceeds the set first risk threshold, the following steps are executed: The upper limit of the target compression scale is set to the first scale; at this time, there is non-in-situ data processed by the ground end as an auxiliary factor, and the compression degree of the lightweight file can be appropriately increased. The airborne terminal is used as the main decision-making terminal. At this time, the airborne terminal outputs the second icing probability for the future time period through the local model based on the corresponding lightweight file. Alternatively, the method may further include the steps of: When the pre-detection risk is less than the set first risk threshold, the following steps are executed: The upper limit of the target compression scale is set to the second scale, and the second scale is greater than the first scale; the ground end is still used as the main decision end, and S103-S105 will be executed at this time.

[0054] In some embodiments, when the pre-detection risk is high, a smaller target compression scale can be applied to retain more details in order to make decisions quickly and prevent the risk from escalating further, thus ensuring that the airborne end can make more accurate judgments more quickly and directly based on limited in-situ data. When the pre-detection risk is low, a larger compression scale can be allowed to save computing power on the airborne end. In this case, the icing risk prediction can be mainly performed by the ground end based on non-in-situ data and lightweight files.

[0055] In other words, this application proposes a multi-terminal collaborative data processing strategy based on actual icing risk, which can quickly determine the leading party in data processing with lower pre-inspection costs, thereby achieving a better balance between resource efficiency and security.

[0056] In some embodiments, the steps further include: Identify the weather category of the aircraft's flight area; Determine whether the meteorological category is a rare category, wherein the rare category includes at least one type: freezing rain, mixed phase ice crystals; If so, the ground terminal will still be used as the primary decision-making terminal, and the following steps will be performed: The observation period is set based on the rare category and the current communication pressure. The observation period is used to limit the amount of data that the lightweight file is allowed to be transmitted in a single transmission window.

[0057] Preferably, the icing risk corresponding to rare categories can be preset. For example, freezing rain corresponds to a level one icing risk, and mixed-phase ice crystals correspond to a level two icing risk.

[0058] The communication pressure can be determined by the computing power of communication resources. For example, the more abundant the computing power of communication resources, the lower the communication pressure.

[0059] In some embodiments, when the icing risk corresponding to a rare category is high and the current communication pressure is high, it is permissible to appropriately reduce the time length (i.e., data length) covered by data in a single transmission, so as to improve the timeliness of data feedback as much as possible through lightweight transmission. Alternatively, in some embodiments, if the current weather category corresponds to a high icing risk, but the current communication pressure is low, the set observation period can be maintained.

[0060] In some embodiments, recommended values ​​for observation periods can be preset for different combinations of rare categories and current communication pressure.

[0061] In this embodiment, for rare weather categories, the present invention proposes a restrictive end-to-end switching mechanism for icing prediction. Specifically, the present invention utilizes the relatively abundant computing resources on the ground (including a complete historical database, expert knowledge base, etc.) while limiting the length of a single data transmission, to ensure that the ground end can make relatively accurate predictions for rare weather conditions within a short period of time. This avoids the possibility of inaccurate prediction results if icing predictions are directly performed by the airborne end under rare weather conditions, due to limited training data samples or airborne equipment problems.

[0062] In some embodiments, the in-situ data includes at least one of the following: cloud particle data, flight status parameters, and wing icing images.

[0063] In some embodiments, the other measuring devices include at least one: a satellite or a probe aircraft.

[0064] In some embodiments, the off-site data includes at least one of the following: cloud top temperature, liquid water content, and median volume diameter.

[0065] Please see Figure 3 This invention also proposes a time-series forecasting system for icing weather based on a spatiotemporal generation network. The aircraft includes an airborne terminal, which is capable of communicating with a ground terminal located in a control center, comprising: The in-situ module receives in-situ data from the airborne terminal, and the in-situ data is monitoring data directly collected by the airborne terminal; A compression module, performed by the airborne terminal, performs lightweight compression and packaging operations on the in-situ data to generate a lightweight file, and sends the lightweight file to the ground terminal; wherein, the compression module includes: A denoising unit is used to perform denoising processing on the in-situ data to obtain denoised data; A compression unit is used to compress the denoised data using a target compression scale to create the lightweight file; The off-site module receives off-site data from the ground terminal, and the off-site data includes monitoring data acquired by other measuring devices located outside the airborne terminal; The prediction module, constructed by the ground terminal, uses the non-in-situ data and the lightweight file to predict the first icing probability of the aircraft in the future time period through a time-series prediction model. The feedback module sends the icing probability from the ground terminal to the airborne terminal.

[0066] In some embodiments, it also includes: The pre-inspection type selection module allows the airborne terminal to select at least one type of data from the in-situ data as pre-inspection data. The prediction module is used to predict the pre-inspection risk of the aircraft in the future based on the pre-inspection data. The compression scale update module is used to update or adjust the target compression scale based on the pre-inspection risk.

[0067] In some embodiments, it also includes: The first execution module is used to execute the following when the pre-detection risk is greater than a set first risk threshold: Set the upper limit of the target compression scale to the first scale; The airborne terminal is used as the main decision-making terminal. At this time, the airborne terminal outputs the second icing probability for the future time period through the local model based on the corresponding lightweight file. Alternatively, the system may also include: The second execution module is used to execute the following when the pre-detection risk is less than a set first risk threshold: Set the upper limit of the target compression scale to the second scale, and the second scale is greater than the first scale; The ground terminal will still be used as the main decision-making terminal, and the system will return to the non-in-situ module.

[0068] The present invention also provides an electronic device comprising: a memory and a processor; wherein the memory is configured to store a computer program; and the processor is coupled to the memory and configured to execute the computer program to perform the steps of any of the methods described herein.

[0069] The present invention also provides a computer-readable storage medium storing computer instructions, wherein when the computer instructions are executed by one or more processors, the one or more processors cause the one or more processors to perform the steps of any of the methods described in the present invention.

[0070] The present invention also provides a computer program product, comprising a computer program that, when executed by one or more processors, causes the one or more processors to perform the steps of any of the methods described in the present invention.

[0071] In other words, this invention provides a lightweight, layered preprocessing mechanism that addresses the bottlenecks of low computing power, limited power consumption, and small storage in UAV-borne embedded units. It employs a layered architecture of simplified airborne preprocessing and deep ground-based fusion, completely offloading complex computations to ground equipment, with the airborne unit performing only zero / low-computing operations. Specifically: the airborne unit performs only threshold-based outlier removal (a noise reduction method, such as direct filtering of cloud particle data and flight state parameters beyond their range, without complex algorithms); wing icing images undergo fixed downsampling and binarization preprocessing (a compression method, compressing the resolution to low resolution, removing background noise, and without CNN feature extraction); all in-situ data is lightweight compressed and packaged to reduce wireless transmission bandwidth usage. Simultaneously, the ground unit receives lightweight airborne data and air-to-ground data, performing icing-specific deep fusion. All operations are optimized for lightweight operation using Transformer, avoiding redundant computations.

[0072] In some embodiments, this application also proposes a categorized, customized preprocessing workflow: For meteorological data (radar, satellite, numerical models): normalization is performed to eliminate the influence of dimensions, resampling is performed to unify spatial resolution, and spatial registration is performed to ensure coordinate consistency. This solves the problem of inconsistent data formats and scales from different sources. For image data (real-time images of wing icing): denoising and enhancement processing is performed, and then icing features are extracted through algorithms such as edge detection and threshold segmentation, transforming complex image information into numerical feature vectors that can be processed by the model, reducing the complexity of direct image input. For airborne time-series data (cloud particle detection, flight status): time-series interpolation is performed to fill data gaps, and outlier removal is performed to ensure data quality. This solves the problems of discontinuity and noise that may exist in sensor data.

[0073] In some embodiments, this application also proposes a unified spatiotemporal alignment and tensor construction mechanism, which synchronizes and matches all preprocessed data based on spatiotemporal stamps. Regardless of the data source and original form, all data are ultimately integrated into a unified spatiotemporal feature tensor. This step transforms multi-source, multimodal data into a standard format that the model can directly accept, fundamentally shielding the heterogeneity of the underlying data and greatly reducing the difficulty of data fusion at the model input.

[0074] In some embodiments, at the model architecture level, the feature fusion layer of the encoder utilized in this invention is specifically responsible for deeply fusing the abstract features from the spatial attention branch and the temporal attention branch with the original features of various types of data. This means that there is a dedicated mechanism within the model to process and integrate these different types of data that have already undergone preliminary cleaning and alignment, rather than simply stacking them.

[0075] The encoder employs a dual-branch design with spatial and temporal attention. The spatial attention branch extracts spatial distribution features from multi-source data (such as differences in meteorological distribution across different regions), while the temporal attention branch captures temporal dependencies (such as the evolution of icing weather patterns). This parallel design of spatial and temporal branches is a specialized adaptation to the physical characteristics of icing weather, which exhibits local spatial aggregation and continuous temporal evolution, and differs structurally from the single attention mechanism of the general Transformer.

[0076] The decoder employs a timing generation module and a three-output head branch design (predicting probability, location, and path respectively). Unlike general-purpose Transformers typically used for discriminative tasks (such as classification and detection), this scheme can generate a spatiotemporal distribution sequence of icing over the next N minutes. This modification is to meet the core requirement of UAV flight safety decision-making for predicting the evolution trend of icing areas in advance.

[0077] In some embodiments, this invention proposes a method and system for predicting UAV icing weather time series based on spatiotemporal generative Transformer (or a method and system for predicting icing weather time series based on spatiotemporal generative network). The method includes: collecting six types of multi-source heterogeneous data: weather radar, satellite remote sensing, numerical weather models, UAV airborne cloud particle detection, real-time images of UAV wing icing, and UAV flight state parameters; preprocessing and spatiotemporally aligning the multi-source heterogeneous data to construct a unified spatiotemporal feature tensor; inputting the spatiotemporal feature tensor into a pre-trained spatiotemporal generative Transformer model, and obtaining future icing prediction results through generative inference; and outputting the icing probability, the strongest icing location, and the most likely icing evolution path. This invention employs generative spatiotemporal modeling, solving the problems of low prediction accuracy, single output, and insufficient spatiotemporal evolution prediction capabilities in existing technologies. By integrating multi-source in-situ and air-to-ground data, it improves adaptability to extreme icing scenarios and can provide real-time decision support for UAV flight safety. It is applicable to icing early warning and trajectory optimization for low-altitude aircraft such as UAVs and general aviation aircraft.

[0078] In some embodiments, this invention proposes a method for predicting UAV icing weather time series based on spatiotemporal generation Transformer (or a method for predicting icing weather time series based on spatiotemporal generation network), comprising the following steps: S1. Collecting multi-source heterogeneous data, including weather radar data, satellite remote sensing data, numerical weather model data, UAV airborne cloud particle detection data, real-time image data of UAV wing icing, and UAV flight status parameters; S2. Preprocessing and spatiotemporally aligning the multi-source heterogeneous data to construct a unified spatiotemporal feature tensor; S3. Inputting the spatiotemporal feature tensor into a pre-trained spatiotemporal generation Transformer model to obtain future icing prediction results through generative inference; S4. Outputting the icing prediction results, which include icing probability, strongest icing location, and most likely icing evolution path.

[0079] In some embodiments, this invention proposes a UAV icing meteorological time-series prediction system based on spatiotemporal generation Transformer (or an icing meteorological time-series prediction system based on spatiotemporal generation network), including a multi-source data acquisition module, a data preprocessing and spatiotemporal fusion module, a spatiotemporal generation Transformer prediction module, and a result output and application module; the multi-source data acquisition module is used to acquire multi-source heterogeneous data, including meteorological radar data, satellite remote sensing data, numerical meteorological model data, UAV airborne cloud particle detection data, real-time image data of UAV wing icing, and UAV data. Flight state parameters; the data preprocessing and spatiotemporal fusion module is used to preprocess and spatiotemporally align the multi-source heterogeneous data to construct a unified spatiotemporal feature tensor; the spatiotemporal generation Transformer prediction module is used to receive the spatiotemporal feature tensor, perform generative inference through a pre-trained spatiotemporal generation Transformer model, and obtain future icing prediction results; the result output and application module is used to output the icing prediction results and send the icing prediction results to the UAV flight control system or ground station, the icing prediction results including icing probability, strongest icing location, and most likely icing evolution path.

[0080] In some embodiments, the UAV-borne cloud particle detection data includes cloud particle spectrum, liquid water content, temperature, humidity, and air pressure; the UAV flight status parameters include flight altitude, speed, attitude, trajectory, and angle of attack; and the real-time image data of UAV wing icing is a high-definition visible light or infrared image.

[0081] In some embodiments, the preprocessing includes: normalizing, resampling, and spatially registering meteorological radar data, satellite remote sensing data, and numerical meteorological model data; denoising, enhancing, and extracting icing features from real-time image data of UAV wing icing; performing temporal interpolation and outlier removal on UAV airborne cloud particle detection data and flight status parameters; and the spatiotemporal alignment involves synchronizing and matching all preprocessed data based on spatiotemporal stamps to construct a unified-dimensional spatiotemporal feature tensor.

[0082] In some embodiments, the spatiotemporal generation Transformer model adopts an encoder-decoder generative architecture; the encoder includes a spatial attention branch, a temporal attention branch, and a feature fusion layer. The spatial attention branch is used to extract the spatial distribution features of multi-source heterogeneous data, the temporal attention branch is used to capture the temporal dependencies of multi-source heterogeneous data, and the feature fusion layer is used to fuse spatial features, temporal features, and the original features of various types of data; the decoder includes a temporal generation module and an output head branch. The temporal generation module is used to generate a spatiotemporal distribution sequence of icing for the next N minutes based on the fused features output by the encoder, and the output head branch is used to predict the icing probability, the strongest icing location, and the most likely icing evolution path based on the spatiotemporal distribution sequence of icing.

[0083] In some embodiments, the value of N ranges from 0 to 30 minutes.

[0084] In some embodiments, the training process of the spatiotemporal generation Transformer model includes: using historical icing meteorological data and UAV cloud penetration test data as training samples, training the model using a joint loss function consisting of icing probability classification loss, location regression loss and path fitting loss, and combining generative adversarial training to improve the authenticity and physical rationality of the spatiotemporal distribution sequence of icing.

[0085] In some embodiments, the icing probability is output in the form of an icing probability distribution map, the strongest icing location is output in the form of the region center coordinates and icing intensity, and the most likely icing evolution path is output in the form of a spatiotemporal vector trajectory.

[0086] In some embodiments, the output icing prediction results are applied to UAV icing warning, flight path planning, or emergency obstacle avoidance. Specifically, the icing prediction results are sent to the UAV flight control system or ground station, and the flight control system automatically adjusts the flight path or the ground station issues a warning.

[0087] In some embodiments, the multi-source data acquisition module includes a radar data interface, a satellite data interface, a numerical mode data interface, an UAV-borne cloud particle sensor, an image acquisition device, and a flight status sensor; the image acquisition device is installed near the UAV wing and is used to acquire real-time images of wing icing; the flight status sensor is connected to the UAV flight control system and is used to acquire flight status parameters.

[0088] In some embodiments, the spatiotemporal generation Transformer prediction module can be deployed on an UAV edge computing unit or a ground server to support real-time inference.

[0089] Example 2 This embodiment can be applied to cloud penetration detection scenarios using fixed-wing UAVs. The specific implementation steps are as follows: 1. Multi-source data acquisition: The system connects to a low-altitude meteorological radar via a radar data interface to acquire radar echo data of the target area; acquires FY-3 satellite remote sensing data via a satellite data interface; acquires meteorological data for the next hour output by the WRF numerical model via a numerical model data interface; collects parameters such as cloud particle spectrum, liquid water content, temperature, humidity, and air pressure via an onboard cloud particle probe on the UAV; acquires real-time images of wing icing via an infrared high-definition camera mounted on the wing leading edge; and collects parameters such as UAV flight altitude (500-1000 meters), speed (80-120 km / h), attitude, trajectory, and angle of attack via a flight status sensor.

[0090] 2. Data Preprocessing and Spatiotemporal Alignment: Radar echo data, satellite data, and WRF numerical model data are normalized to a spatial resolution of 100m×100m and spatially registered; infrared icing images are denoised and enhanced, and icing region features are extracted using the Canny edge detection algorithm; airborne cloud particle data and flight state parameters are temporally interpolated to remove outliers; all data are synchronized and aligned based on spatiotemporal stamps to construct a 32×32×12 (time step × spatial dimension × feature dimension) spatiotemporal feature tensor.

[0091] 3. Model Inference: The spatiotemporal feature tensor is input into the pre-trained spatiotemporal generation Transformer model. The spatial attention branch of the model encoder focuses on areas with high cloud particle concentration and temperatures below 0°C, while the temporal attention branch captures the meteorological change trend over the past 30 minutes. The feature fusion layer integrates various data features. The decoder generates the spatiotemporal distribution sequence of icing over the next 15 minutes. The output head branches predict the icing probability distribution map, the location of the strongest icing (coordinates: 30°52′N, 114°35′E, icing intensity: moderate icing), and the icing evolution path (moving northeastward at a speed of approximately 15 km / h).

[0092] 4. Results Application: The results output and application module sends the prediction results to the ground station, which issues a moderate icing warning. At the same time, the results are sent to the UAV flight control system, which automatically adjusts the flight path, shifting 2km to the southwest to avoid the icing area and ensure the safety of the UAV's cloud penetration detection.

[0093] In some embodiments, this application also provides a schematic block diagram of the structure of a computer device, please see... Figure 2 Computer programs can be used in situations such as Figure 2 It runs on the computer device shown. Figure 2 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include non-volatile storage media and internal memory. The non-volatile storage media may store an operating system and computer programs. The computer programs include program instructions that, when executed, cause the processor to perform arbitrary methods. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer programs in the non-volatile storage media; when executed by the processor, these programs cause the processor to perform arbitrary methods. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 2 The structures shown are merely block diagrams of a portion of the structure related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. It should be understood that the processor may be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0094] Example 3 This embodiment can be applied to power line inspection scenarios using multi-rotor drones. The specific implementation steps are as follows: 1. Multi-source data acquisition: The system connects to a dedicated low-altitude radar for power line inspection via a radar data interface to acquire radar echo data of the inspection area; it acquires high-resolution satellite remote sensing data via a satellite data interface; it acquires short-term numerical meteorological data of the region via a numerical model data interface; it collects cloud microphysical parameters via an airborne cloud particle sensor on the UAV; it collects real-time images of wing icing via a visible light camera near the wing; and it collects parameters such as UAV flight altitude (100-300 meters), speed (30-50 km / h), attitude, and trajectory via a flight status sensor.

[0095] 2. Data preprocessing and spatiotemporal alignment: Preprocess and spatially register various meteorological data to unify the spatial resolution to 50m×50m; denoise and enhance visible light icing images and extract icing features; remove outliers and perform temporal interpolation on airborne data and flight status parameters; align all data based on spatiotemporal stamps and construct a 24×24×10 spatiotemporal feature tensor.

[0096] 3. Model Inference: The spatiotemporal feature tensor is input into the spatiotemporal generation Transformer model. The decoder generates the spatiotemporal distribution sequence of icing for the next 5 minutes, and outputs the icing probability (0.75 in the northeast of the inspection area, <0.3 in other areas), the location of the strongest icing (near the power line tower, coordinates: 30°15′N, 114°10′E, icing intensity: light icing) and the icing evolution path (basically remaining stationary, locally spreading towards the tower).

[0097] 4. Results Application: The results output and application module sends the prediction results to the UAV flight control system. The flight control system automatically adjusts the inspection route to avoid the icing area in the northeast and prioritizes the inspection of lines with no risk of icing. At the same time, the ground station receives the results and reminds the inspection personnel to pay attention to the icing situation of the towers in the northeast and conduct a focused review later.

[0098] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a computer terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0100] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for predicting icing weather time series based on spatiotemporal generation networks, characterized in that, Aircraft include: An airborne terminal, wherein the airborne terminal is capable of communicating with a ground terminal located at the control center, includes the following steps: S101, the airborne terminal receives in-situ data, which is monitoring data directly collected by the airborne terminal; S102, the airborne terminal performs a lightweight compression and packaging operation on the in-situ data to generate a lightweight file, and sends the lightweight file to the ground terminal; wherein, S102 includes the following steps: The in-situ data is subjected to denoising processing to obtain denoised data; The denoised data is compressed using a target compression scale to create the lightweight file; S103, the ground terminal receives non-in-situ data, the non-in-situ data including monitoring data acquired by other measuring devices located outside the airborne terminal; S104, the ground terminal uses the non-in-situ data and the lightweight file to predict the first icing probability of the aircraft in a future time period through a time-series prediction model; S105, the ground terminal sends the icing probability to the airborne terminal.

2. The method for predicting icing weather time series based on spatiotemporal generation networks according to claim 1, characterized in that, It also includes the following steps: The airborne terminal selects at least one type of data from the in-situ data as pre-inspection data; Predict the pre-inspection risk of the aircraft in the future based on the pre-inspection data; The target compression scale is updated or adjusted based on the pre-inspection risks.

3. The method for predicting icing weather time series based on spatiotemporal generation networks according to claim 2, characterized in that, It also includes the following steps: When the pre-detection risk exceeds the set first risk threshold, the following steps are executed: Set the upper limit of the target compression scale to the first scale; The airborne terminal is used as the main decision-making terminal. At this time, the airborne terminal outputs the second icing probability for the future time period through the local model based on the corresponding lightweight file. Alternatively, the method may further include the steps of: When the pre-detection risk is less than the set first risk threshold, the following steps are executed: Set the upper limit of the target compression scale to the second scale, and the second scale is greater than the first scale; The ground terminal will still be used as the main decision-making terminal, and S103-S105 will be executed at this time.

4. The method for predicting icing weather time series based on spatiotemporal generation networks according to claim 3, characterized in that, It also includes the following steps: Identify the weather category of the aircraft's flight area; Determine whether the meteorological category is a rare category, wherein the rare category includes at least one type: freezing rain, mixed phase ice crystals; If so, the ground terminal will still be used as the primary decision-making terminal, and the following steps will be performed: The observation period is set based on the rare category and the current communication pressure. The observation period is used to limit the amount of data that the lightweight file is allowed to be transmitted in a single transmission window.

5. The method for predicting icing weather time series based on spatiotemporal generation networks according to claim 1, characterized in that, The in-situ data includes at least one of the following: cloud particle data, flight status parameters, and wing icing images.

6. The method for predicting icing weather time series based on spatiotemporal generation networks according to claim 1, characterized in that, The other measuring equipment includes at least one of the following: a satellite or a probe aircraft.

7. The method for predicting icing weather time series based on spatiotemporal generation networks according to claim 1, characterized in that, The non-in-situ data includes at least one of the following: cloud top temperature, liquid water content, and median volume diameter.

8. A time-series forecasting system for icing weather based on a spatiotemporal generation network, characterized in that, Aircraft include: An airborne terminal, wherein the airborne terminal is capable of communicating with a ground terminal located at the control center, includes: The in-situ module receives in-situ data from the airborne terminal, and the in-situ data is monitoring data directly collected by the airborne terminal; A compression module, performed by the airborne terminal, performs lightweight compression and packaging operations on the in-situ data to generate a lightweight file, and sends the lightweight file to the ground terminal; wherein, the compression module includes: A denoising unit is used to perform denoising processing on the in-situ data to obtain denoised data; A compression unit is used to compress the denoised data using a target compression scale to create the lightweight file; The off-site module receives off-site data from the ground terminal, and the off-site data includes monitoring data acquired by other measuring devices located outside the airborne terminal; The prediction module, constructed by the ground terminal, uses the non-in-situ data and the lightweight file to predict the first icing probability of the aircraft in a future time period through a time-series prediction model. The feedback module sends the icing probability from the ground terminal to the airborne terminal.

9. The icing meteorological time series prediction system based on spatiotemporal generation network according to claim 8, characterized in that, Also includes: The pre-inspection type selection module allows the airborne terminal to select at least one type of data from the in-situ data as pre-inspection data. The prediction module is used to predict the pre-inspection risk of the aircraft in the future based on the pre-inspection data. The compression scale update module is used to update or adjust the target compression scale based on the pre-inspection risk.

10. The icing meteorological time series prediction system based on spatiotemporal generation network according to claim 9, characterized in that, Also includes: The first execution module is used to execute the following when the pre-detection risk is greater than a set first risk threshold: Set the upper limit of the target compression scale to the first scale; The airborne terminal is used as the main decision-making terminal. At this time, the airborne terminal outputs the second icing probability for the future time period through the local model based on the corresponding lightweight file. Alternatively, the system may also include: The second execution module is used to execute the following when the pre-detection risk is less than a set first risk threshold: Set the upper limit of the target compression scale to the second scale, and the second scale is greater than the first scale; The ground terminal will still be used as the main decision-making terminal, and the system will return to the non-in-situ module.

Images