A false precipitation data identification method and system, an electronic device, and a storage medium

Abstract

The application discloses a false precipitation data identification method and system, electronic equipment and a storage medium. The method comprises the following steps: determining an original observation sequence; based on the original observation sequence, extracting liquid precipitation, air temperature, and snow depth corresponding to the precipitation difference, air temperature difference, and snow depth difference; based on the air temperature at the current time and the snow depth difference at the current time, calculating the snowmelt index; by accumulating all snow depth differences that meet the snowmelt condition in the historical time window, the accumulated snowmelt equivalent is obtained; the precipitation difference, the air temperature difference, the snow depth difference, the snowmelt index, and the accumulated snowmelt equivalent are constructed into a difference physical feature sequence; a false precipitation probability prediction model is constructed; the original observation sequence and the difference physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability; and the quality control code mark is marked on the precipitation data according to the false precipitation probability. The application can improve the accuracy of identifying snowmelt false precipitation data.

Description

Technical Field

[0001] This application relates to the field of data quality control technology, and in particular to a method, system, electronic device and storage medium for identifying false precipitation data. Background Technology

[0002] Traditional precipitation data quality control primarily relies on spatial consistency and temporal consistency checks. Spatial consistency checks compare precipitation observations at a target station with those at neighboring stations to determine if the data is abnormal; temporal consistency checks analyze the continuity of precipitation time series at a single station and the range of climatic extremes to assess the data's reasonableness. However, traditional methods have significant limitations: spatial consistency checks are difficult to implement effectively in sparsely populated areas (such as mountainous or high-altitude regions) and have a high misclassification rate for highly localized precipitation processes (such as convective precipitation); temporal consistency checks struggle to identify systematic errors with "smooth changes"—false precipitation generated by snowmelt processes exhibits slow, continuous changes in its time series, highly similar to real precipitation processes in temporal morphology, making it prone to being missed by traditional thresholding methods.

[0003] Although existing technologies have studied precipitation data quality control methods based on machine learning or deep learning, these methods still have problems such as simple feature fusion methods, failure to consider the dynamic weights of physical features, lack of organic integration between physical constraints and data-driven approaches, and insufficient utilization of multi-timescale features of the snowmelt process.

[0004] In summary, existing technologies still suffer from relatively low accuracy in identifying false snowmelt precipitation data. Summary of the Invention

[0005] This application aims to provide a method, system, electronic device, and storage medium for identifying false precipitation data, which can improve the accuracy of identifying false precipitation data related to snowmelt.

[0006] In a first aspect, embodiments of this application provide a method for identifying false precipitation data, the method comprising: Acquire multi-source meteorological observation data including liquid precipitation, air temperature, and snow depth; preprocess the multi-source meteorological observation data to obtain the original observation sequence; Based on the original observation sequence, the precipitation difference, temperature difference, and snow depth difference corresponding to the liquid precipitation, the air temperature, and the snow depth are extracted respectively; The snowmelt index is calculated based on the current temperature and the current snow depth difference; the cumulative snowmelt equivalent is obtained by accumulating the snow depth differences of all snowmelt conditions within the historical time window. The precipitation difference, the temperature difference, the snow depth difference, the snowmelt index, and the cumulative snowmelt equivalent are constructed into a differential physical feature sequence; A false precipitation probability prediction model is constructed, which includes a temporal encoder, a lightweight context encoding network, and a classification head; the original observation sequence and the differential physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context encoding network includes one-dimensional convolution and global average pooling; Based on the false precipitation probability, the precipitation data is marked with a quality control code.

[0007] In some implementations, extracting the precipitation difference, temperature difference, and snow depth difference corresponding to the liquid precipitation, air temperature, and snow depth based on the original observation sequence includes: The difference between the liquid precipitation at two adjacent moments in the original observation sequence is calculated to obtain the precipitation difference. The temperature difference is obtained by calculating the difference between the temperatures of two adjacent moments in the original observation sequence; The snow depth difference is obtained by calculating the difference between two adjacent times in the original observation sequence.

[0008] In some implementations, calculating the snowmelt index based on the difference between the current air temperature and the current snow depth includes: The temperature difference between the current temperature and the preset snow melting threshold temperature is calculated to obtain the temperature difference result. The temperature difference result is compared with zero, and the first maximum value is selected; Take the inverse of the snow depth difference at the current moment to obtain the inverse value of the snow depth difference; The opposite value of the snow depth difference is compared with zero, and the second maximum value is selected; Multiply the first maximum value and the second maximum value to obtain the snow melting index.

[0009] In some implementations, obtaining the cumulative snowmelt equivalent by accumulating the snow depth differences of all snowmelt conditions within a historical time window includes: Construct a snow melting condition where the temperature is higher than the snow melting threshold temperature, and the value is 1. The cumulative snowmelt equivalent is obtained by taking the negative of the snow depth difference for all snowmelt conditions that meet the conditions within the cumulative historical time window.

[0010] In some implementations, the temporal encoder includes a first temporal encoder and a second temporal encoder, and the trained false precipitation probability prediction model is trained through a total loss function, which is constructed through a weighted binary cross-entropy loss and a physically guided auxiliary loss. The step of inputting the original observation sequence and the differential physical feature sequence into the trained false precipitation probability prediction model to obtain the false precipitation probability includes: The original observation sequence is input into the first time encoder to extract the first global time encoding vector of the original observation; The differential physical feature sequence is input into the second temporal encoder to extract the second global temporal encoding vector of the differential physical features; The original observation sequence and the differential physical feature sequence are concatenated to obtain the sequence concatenation result; The sequence concatenation result is input into the lightweight context coding network to extract the context vector; Based on the first global temporal coding vector, the second global temporal coding vector, and the context vector, calculate the gating weights; Based on the gating weights, the first global temporal coding vector and the second global temporal coding vector are fused to obtain a fused feature vector. The fused feature vector is input into the classification head to obtain the false precipitation probability.

[0011] In some implementations, calculating the gating weights based on the first global temporal coding vector, the second global temporal coding vector, and the context vector includes: The first global temporal coding vector, the second global temporal coding vector, and the context vector are concatenated to obtain the vector concatenation result; The concatenated vectors are multiplied by the weight matrix to obtain the multiplication result; the multiplication result is then added to the bias term to obtain the summation result. The summation result is input into the activation function to obtain the gating weights.

[0012] In some implementations, the step of fusing the first global temporal coding vector and the second global temporal coding vector based on the gating weights to obtain a fused feature vector includes: The first global temporal coding vector is multiplied element-wise with the gate weight to obtain the first element-wise multiplication result; The subtraction result is obtained by subtracting the gate weight; the second global temporal coding vector is multiplied element-wise with the subtraction result to obtain the second element-wise multiplication result. The first element-wise multiplication result and the second element-wise multiplication result are added together to obtain the fused feature vector.

[0013] Secondly, embodiments of this application also provide a false precipitation data identification system, the system comprising: The data preprocessing module is used to acquire multi-source meteorological observation data including liquid precipitation, air temperature, and snow depth; and to preprocess the multi-source meteorological observation data to obtain the original observation sequence. The feature extraction module is used to extract the precipitation difference, temperature difference, and snow depth difference corresponding to the liquid precipitation, the air temperature, and the snow depth, respectively, based on the original observation sequence. The data calculation module is used to calculate the snowmelt index based on the current temperature and the current snow depth difference; and to obtain the cumulative snowmelt equivalent by accumulating the snow depth differences of all snowmelt conditions within the historical time window. The sequence construction module is used to construct the precipitation difference, the temperature difference, the snow depth difference, the snowmelt index, and the cumulative snowmelt equivalent into a differential physical feature sequence; The model prediction module is used to construct a false precipitation probability prediction model that includes a temporal encoder, a lightweight context encoding network, and a classification head; the original observation sequence and the differential physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context encoding network includes one-dimensional convolution and global average pooling; The data quality control module is used to mark the precipitation data with quality control codes based on the false precipitation probability.

[0014] Thirdly, embodiments of this application also provide an electronic device, including at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform a false precipitation data identification method as described above.

[0015] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer-executable instructions for causing a computer to execute a false precipitation data identification method as described above.

[0016] Compared with the prior art, this application has the following beneficial effects: This application acquires multi-source meteorological observation data including liquid precipitation, temperature, and snow depth; preprocesses the multi-source meteorological observation data to obtain the original observation sequence; based on the original observation sequence, extracts the precipitation difference, temperature difference, and snow depth difference corresponding to liquid precipitation, temperature, and snow depth respectively; calculates the snowmelt index based on the current temperature and the current snow depth difference; obtains the cumulative snowmelt equivalent by accumulating all snow depth differences that meet the snowmelt conditions within the historical time window; constructs a differential physical feature sequence from the precipitation difference, temperature difference, snow depth difference, snowmelt index, and cumulative snowmelt equivalent; constructs a false precipitation probability prediction model including a temporal encoder, a lightweight context coding network, and a classification head; inputs the original observation sequence and the differential physical feature sequence into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context coding network includes one-dimensional convolution and global average pooling; and marks the precipitation data with quality control codes according to the false precipitation probability.

[0017] Thus, by extracting the precipitation difference, temperature difference, and snow depth difference corresponding to liquid precipitation, temperature, and snow depth, and then inputting the original observation sequence and the differential physical feature sequence into a trained false precipitation probability prediction model, the physical priors of the snowmelt process can be explicitly encoded into the deep learning model. This reduces the learning difficulty of the model implicitly learning physical laws from the original data, and improves the accuracy and interpretability of identifying false snowmelt precipitation data. By marking precipitation data with quality control codes based on the accurately identified false precipitation probabilities, large amounts of meteorological station data can be processed in real time without human intervention, improving the efficiency and objectivity of meteorological data quality control. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart illustrating an embodiment of the false precipitation data identification method provided in this application; Figure 2 This is a schematic diagram of the overall scheme flow in the best embodiment of the false precipitation data identification method provided in this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the false precipitation data identification system provided in this application; Figure 4 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0020] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0021] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and 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, and therefore should not be construed as a limitation of this application.

[0022] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0023] To address the issue of low accuracy in identifying false precipitation data related to snowmelt in existing technologies, this application proposes a method, system, electronic device, and storage medium for identifying false precipitation data.

[0024] Reference Figure 1 This application provides a schematic flowchart of a method for identifying false precipitation data. This method is applied to an electronic device, which may be a server or a mobile terminal, etc. Figure 1 As shown, the method for identifying false precipitation data may include the following steps S101 to S106.

[0025] Step S101: Obtain multi-source meteorological observation data including liquid precipitation, air temperature and snow depth; preprocess the multi-source meteorological observation data to obtain the original observation sequence.

[0026] In this step, hourly surface meteorological observation data from the target meteorological station during winter / spring (usually November to March of the following year) are acquired to construct multi-source meteorological observation data. This multi-source meteorological observation data must include at least the following elements: liquid precipitation, 2-meter air temperature, and snow depth. Data exceeding physically reasonable ranges in the multi-source meteorological observation data are removed. For sequences with no more than 3 consecutive hours of missing data, linear interpolation is used to fill in the gaps. Sequences with more than 3 consecutive hours of missing data are removed and not used for model training and inference.

[0027] Step S102: Based on the original observation sequence, extract the precipitation difference, temperature difference, and snow depth difference corresponding to liquid precipitation, air temperature, and snow depth.

[0028] In this step, the difference between the liquid precipitation at two adjacent moments in the original observation sequence is calculated to obtain the precipitation difference; the difference between the air temperature at two adjacent moments in the original observation sequence is calculated to obtain the air temperature difference; and the difference between the snow depth at two adjacent moments in the original observation sequence is calculated to obtain the snow depth difference.

[0029] Step S103: Calculate the snowmelt index based on the current temperature and the current snow depth difference; obtain the cumulative snowmelt equivalent by accumulating the snow depth differences of all snowmelt conditions within the historical time window.

[0030] In this step, the difference between the current air temperature and the preset snow melting threshold temperature is calculated to obtain the air temperature difference result; the air temperature difference result is compared with zero, and the first maximum value is selected; the inverse of the current snow depth difference is taken to obtain the inverse value of the snow depth difference; the inverse value of the snow depth difference is compared with zero, and the second maximum value is selected; the first maximum value and the second maximum value are multiplied to obtain the snow melting index. That is, the snow melting index is calculated as follows: ; in, Indicates the snowmelt index. This indicates the current temperature. This indicates the preset snow melting threshold temperature. express The difference in snow depth at any given moment This indicates taking the maximum value.

[0031] A snowmelt condition with a value of 1 is constructed when the air temperature is above the snowmelt threshold temperature. The cumulative snowmelt equivalent is obtained by taking the negative of the snow depth difference for all snowmelt conditions that meet the cumulative historical time window. Specifically, the cumulative snowmelt equivalent is calculated as follows: ; in, This represents the cumulative snowmelt equivalent. express The difference in snow depth at any given moment express Temperature at any given moment Indicates an indicator function, when The value is 1 when the temperature is higher than the snow melting threshold temperature, and 0 when the temperature is lower than or equal to the snow melting threshold temperature.

[0032] Step S104: Construct a differential physical feature sequence from the precipitation difference, temperature difference, snow depth difference, snow melt index, and cumulative snow melt equivalent.

[0033] In this step, the precipitation difference, temperature difference, snow depth difference, snowmelt index, and cumulative snowmelt equivalent are combined to construct a differential physical feature sequence. This combination can be done chronologically, for example, grouping data from the same time period together; this embodiment does not impose specific limitations on this.

[0034] Step S105: Construct a false precipitation probability prediction model that includes a temporal encoder, a lightweight context encoding network, and a classification head; input the original observation sequence and the differential physical feature sequence into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context encoding network includes one-dimensional convolution and global average pooling.

[0035] In this step, a false precipitation probability prediction model is constructed, comprising a temporal encoder, a lightweight context encoder network, and a classification head; a total loss function is constructed, comprising weighted binary cross-entropy loss and physical guidance auxiliary loss; based on the total loss function, the false precipitation probability prediction model is trained to obtain a trained false precipitation probability prediction model; the original observation sequence and the differential physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability, including: The original observation sequence is input into the first time-series encoder to extract the first global time-series encoding vector of the original observations; the differential physical feature sequence is input into the second time-series encoder to extract the second global time-series encoding vector of the differential physical features; the original observation sequence and the differential physical feature sequence are concatenated to obtain the sequence concatenation result; the sequence concatenation result is input into a lightweight context encoding network to extract the context vector; based on the first global time-series encoding vector, the second global time-series encoding vector, and the context vector, the gating weights are calculated; based on the gating weights, the first global time-series encoding vector and the second global time-series encoding vector are fused to obtain the fused feature vector; the fused feature vector is input into the classification head to obtain the false precipitation probability.

[0036] The weighted binary cross-entropy loss described above is constructed as follows: ; in, This represents the weighted binary cross-entropy loss. Indicates the number of samples. For positive class weights, Indicates the number of negative samples. Indicates the number of positive samples. Indicates the predicted first The probability of false precipitation for each sample. Indicates the first The true label of each sample.

[0037] The physical-guided auxiliary loss described above is constructed as follows: ; in, Indicates the auxiliary loss of physical guidance. Indicates the positive rate of change of snow depth. This indicates the values ​​recorded by the rain gauge during the same period. This represents the snow melting conversion efficiency coefficient. This represents the square of the L2 norm.

[0038] The first and second time-series encoders mentioned above can be any of the following network models: bidirectional long short-term memory network, temporal convolutional network, and Transformer encoder. This embodiment can also use other network models known to those skilled in the art, and this embodiment does not impose any specific limitations on them.

[0039] The above calculation of gating weights based on the first global temporal coding vector, the second global temporal coding vector, and the context vector includes: The first global temporal coding vector, the second global temporal coding vector, and the context vector are concatenated to obtain the vector concatenation result. This concatenated result is multiplied by the weight matrix to obtain the multiplication result. The multiplication result is then added to the bias term to obtain the summation result. This summation result is input into the activation function to obtain the gated weights. In other words, the gated weights can be obtained in the following way: ; in, Indicates the gating weight, This represents the Sigmoid activation function. Represents the weight matrix. This indicates a splicing operation. This represents the first global temporal coding vector. This represents the second global temporal coding vector. Represents the context vector. This indicates the bias term.

[0040] The above-mentioned method, based on gated weights, fuses the first and second global temporal coding vectors to obtain a fused feature vector, including: The first global temporal coding vector is element-wise multiplied with the gate weights to obtain the first element-wise multiplication result; the gate weights are then subtracted to obtain the subtraction result; the second global temporal coding vector is element-wise multiplied with the subtraction result to obtain the second element-wise multiplication result; the first and second element-wise multiplication results are then added to obtain the fused feature vector. In other words, the fused feature vector can be obtained in the following way: ; in, Represents the fused feature vector. This indicates element-wise multiplication.

[0041] The above process of inputting the fused feature vector into the classification head to obtain the false precipitation probability includes: Fusion features Input a classification header, output the probability of false precipitation. Specifically: ; in, Represents the weight parameters. Indicates the bias parameter. and These are the learnable parameters for the classification head. This is a randomly deactivated layer (used to prevent overfitting).

[0042] By comprehensively calculating the gating weights from the first global temporal coding vector, the second global temporal coding vector, and the context vector, and then performing vector fusion on the first and second global temporal coding vectors based on the gating weights, the model can dynamically adjust the fusion weights of the original observation features and physical difference features according to the current meteorological background. This enables adaptive feature fusion under different snowmelt scenarios, improves the robustness of the model under complex meteorological conditions, and further enhances the accuracy of identifying false snowmelt precipitation data.

[0043] Step S106: Mark the precipitation data with quality control codes based on the false precipitation probability.

[0044] In this step, precipitation data is marked with quality control codes based on the false precipitation probability, including: ; in, This represents the quality control code. A quality control code of 7 indicates that the precipitation data for that time period was identified as false precipitation due to snowmelt, while a quality control code of 0 indicates that the precipitation data for that time period is correct. This represents the decision threshold.

[0045] To facilitate understanding by those skilled in the art, a set of preferred embodiments is provided below: Precipitation observation data is a core foundation for meteorological forecasting, hydrological simulation, climate change research, and disaster prevention and mitigation decision-making. Precipitation data observed at ground rain gauges is generally considered the ground truth and is widely used in various meteorological operations and scientific research verifications. However, due to factors such as insufficient instrument maintenance, environmental interference, and technical errors, precipitation observation data suffers from systematic biases. Among these, spurious snowmelt precipitation is a special type of systematic error in winter and early spring: when temperatures rise above 0°C, surface snow melts, and the meltwater enters the rain gauge and is recorded as "precipitation," even though no actual precipitation is occurring in the atmosphere. If this type of spurious precipitation data is not effectively identified and labeled, it will directly affect the quality of real-time analysis products, thereby impacting the reliability of weather forecasts, hydrological forecasts, and climate monitoring. Existing related technologies have the following shortcomings: (1) There is a lack of a dedicated identification model for false precipitation caused by snowmelt.

[0046] Most existing precipitation data quality control methods are geared towards detecting general anomalies (such as instrument malfunctions, transmission errors, and extreme values), rather than systematic errors caused by specific physical processes. Snowmelt-induced false precipitation has a unique physical origin—a coupled process of snow depth decrease and temperature rise—and existing methods do not fully utilize this prior physical information for targeted modeling. While a meteorological information center has proposed an operational requirement to "improve the ability to identify snowmelt-induced false precipitation in winter and spring," it has not yet provided a specific and feasible deep learning technology solution.

[0047] (2) The feature fusion method is simple and does not consider the dynamic weight of physical features.

[0048] Existing deep learning-based precipitation quality control methods often employ simple concatenation in feature processing, directly concatenating feature vectors such as temperature, precipitation, and snow depth before inputting them into the neural network. This approach assumes that each feature is independent and contributes equally, failing to dynamically adjust the trust weights of different features based on meteorological background. However, in identifying false snowmelt precipitation, the importance of different features varies significantly across different scenarios. The rate of change in snow depth during early spring afternoon radiative snowmelt is the most critical, while during nighttime freezing rain transitioning to rain, there is a conflict between the original precipitation signal and the snowmelt signal, requiring dynamic balancing by the model. Simple concatenation cannot achieve this adaptive fusion.

[0049] (3) Physical constraints and data-driven approaches lack organic integration.

[0050] Existing deep learning methods primarily rely on a purely data-driven approach, learning the mapping relationship between input and output through large amounts of labeled data. While these methods can capture nonlinear features, they suffer from a "physical inconsistency" problem, meaning the model may learn spurious correlations that violate fundamental physical laws. For example, the model might misclassify "rising temperature, constant snow depth, and increased precipitation" as spurious snowmelt precipitation, when in reality, snowmelt would be impossible without a decrease in snow depth. Furthermore, existing methods do not embed physical constraints such as mass conservation into the model training process, resulting in insufficient generalization ability in extreme samples or unseen scenarios.

[0051] (4) Insufficient utilization of the multi-timescale characteristics of the snow melting process.

[0052] Sham snowmelt precipitation involves physical processes at multiple time scales: hourly snowmelt pulses (short-term radiative snowmelt), daily snow ablation trends (long-term snowmelt due to sustained warming), and cumulative snowmelt effects (the correlation between total snowmelt over multiple days and total sham precipitation). Existing methods typically use fixed time windows for input and do not explicitly separate and adaptively fuse features at different time scales, making it difficult for models to simultaneously capture the discriminative signals of rapid and slow snowmelt events.

[0053] (5) Manual quality control is inefficient and highly subjective.

[0054] Traditional manual and semi-automatic quality control methods require significant time and manpower and suffer from subjectivity and inconsistency. While some research has attempted to build fully automated deep learning-based quality control systems, automated and high-precision identification solutions for the specific sub-problem of false snowmelt precipitation remain lacking.

[0055] The main objective of the technical solution in this embodiment is: 1. A method for intelligent identification of false precipitation caused by snowmelt based on physical guided differential features and gating fusion is provided. This method can automatically and accurately identify false precipitation data caused by snowmelt in rain gauge records and assign corresponding quality control mark codes (such as quality control code 7), thereby improving the quality and availability of winter / spring precipitation observation data.

[0056] 2. By constructing physical-guided differential features (including snow depth change rate, snowmelt index, and cumulative snowmelt equivalent), the physical priors of the snowmelt process are explicitly encoded into the deep learning model, reducing the learning difficulty of the model implicitly learning physical laws from the original data and improving the identification accuracy and interpretability of false snowmelt precipitation data.

[0057] 3. By designing a gating fusion mechanism, the model (i.e., the false precipitation probability prediction model) can dynamically adjust the fusion weights of the original observation features and physical difference features according to the current meteorological background, thereby achieving adaptive feature fusion under different snow melting scenarios and improving the robustness of the model under complex meteorological conditions.

[0058] 4. By introducing a physical auxiliary loss term (mass conservation constraint) into the loss function, physical laws are embedded into the model training process, ensuring that the model output conforms to basic physical consistency and improving the model's generalization ability in extreme samples and unseen scenarios.

[0059] 5. Provides an end-to-end automated identification solution that can process large amounts of meteorological station data in real time without human intervention, thereby improving the efficiency and objectivity of meteorological data quality control.

[0060] The core idea of ​​this embodiment is to explicitly encode the physical priors of the snowmelt process into differential features, then dynamically control the fusion ratio of the original observation features and the physical differential features through a gating mechanism. Finally, the classification head outputs the probability that the precipitation at that time is false snowmelt precipitation, and assigns a corresponding quality control code according to a preset threshold. (Refer to...) Figure 2 The overall technical solution of this embodiment includes the following main steps: Step 1: Acquisition and preprocessing of multi-source meteorological observation data.

[0061] (1) Data Acquisition. Hourly surface meteorological observation data from the target meteorological station during winter / spring (usually November to March of the following year) are acquired to construct multi-source meteorological observation data. This multi-source meteorological observation data must include at least the following elements: liquid precipitation (… Rainfall measured hourly by a rain gauge, in mm / h; air temperature at 2 meters ( ), in °C; snow depth ( (Unit: cm). Optional auxiliary features include: net surface radiation, relative humidity, and wind speed.

[0062] (2) Data Preprocessing. Outlier Removal: Records in multi-source meteorological observation data that clearly exceed the physically reasonable range are removed (e.g., temperatures greater than 50℃ or less than -50℃, negative snow depth, etc.). Missing Value Handling: Linear imputation is used for sequences with no more than 3 consecutive hours of missing data in multi-source meteorological observation data; sequences with more than 3 consecutive hours of missing data are removed and not used for model training and inference. Time Window Construction: Time series samples are constructed using a sliding window method, with a window length of... The optimal timeframe is 6 to 24 hours (adjustable based on business needs). Each sample includes the target time and its preceding time. Multivariate observation data at various time points.

[0063] In this embodiment, the preferred time window length is 12 hours. A window that is too short (<6 hours) will make it difficult to capture the continuous characteristics of snow melting; a window that is too long (>24 hours) will introduce too much noise and increase the computational burden. The time window length can be changed according to the actual situation, and this embodiment does not impose a specific limitation on it.

[0064] Step 2: Physically guided differential feature extraction.

[0065] Unlike the traditional method of directly inputting the original observation sequence into the model, this embodiment explicitly constructs a set of physically guided differential features, encoding the physical priors of the snowmelt process into input signals that the model can directly perceive.

[0066] (1) Basic difference characteristics. For the original observation sequence ( (For the number of variables), calculate the first difference for each feature: ; in, Represents the first-order difference. express The original observations at time [time]. express The original observation value at time.

[0067] Temperature difference This reflects temperature change trends and helps determine whether conditions are transitioning towards snow melting. Among them, express Temperature at any given moment express The temperature at that moment.

[0068] Snow depth difference This reflects the net change in snow cover; negative values ​​indicate snow melt. Among them, express The snow was deep at that moment. express The snow was deep at that moment.

[0069] Precipitation difference This reflects the instantaneous change in precipitation intensity. Among them, express Rainfall at any given time express Rainfall at any given moment.

[0070] (2) Melt Index. By explicitly coupling the positive zone of air temperature with the rate of snow depth decrease, a physical characteristic that can directly measure the driving force of snow melting is constructed, namely the Melt Index. for: ; in, The snowmelt threshold temperature is preferably 0℃ (this can be adjusted according to regional climate characteristics and altitude; for example, it can be set to the range of -1℃ to 1℃ in areas near the 0℃ isotherm, but this embodiment does not impose specific limitations on this). This characteristic means that the snowmelt index is positive only when the air temperature is higher than the snowmelt threshold and the snow depth decreases at the same time. Its value reflects both the temperature driving intensity and the snowmelt rate.

[0071] (3) Cumulative Melt Equivalent. To capture the cumulative effect of the snowmelt process (the correspondence between the total water released by continuous snowmelt over multiple days and the total amount of false precipitation), a feature of the cumulative melt equivalent is constructed. for: ; in, , express The snow was deep at that moment. express The snow was deep at that moment. For indicator functions, when The value is 1 when the temperature is higher than the snow melting threshold temperature, and 0 when the temperature is lower than or equal to the snow melting threshold temperature, which is used to filter the effective snow melting period.

[0072] This feature sums up the snow depth decrease for all snowmelt conditions within a historical time window, reflecting the total amount of water released by snowmelt up to the current moment, and physically corresponds to the cumulative false precipitation recorded by the rain gauge.

[0073] Step 3: Dual-channel timing feature encoding.

[0074] This embodiment employs a dual-encoder parallel architecture to extract temporal features from the original observation sequence and the differential physical feature sequence, respectively.

[0075] (1) Original feature encoder. Let the original observation sequence be... ( (The original feature dimension) is input into the first temporal encoder. : ; in, This is the first global temporal encoding vector of the original observation.

[0076] (2) Differential Feature Encoder. Let the differential physical feature sequence be... ( (For the differential feature dimension), it is input into the second temporal encoder. : ; in, This is the second global temporal coding vector for the differential physical features.

[0077] (3) Encoder implementation. Sequential encoder. and Any of the following deep learning time series models can be used: Bidirectional Long Short-Term Memory (BiLSTM) networks can capture bidirectional long-term dependencies in sequences; Temporal Convolutional Networks (TCNs) offer advantages in parallel computation and a longer effective receptive field; Transformer encoders capture dependencies between arbitrary positions in a sequence through self-attention mechanisms. The preferred approach is to use a two-layer BiLSTM with a hidden layer dimension of [missing information]. This configuration achieves a good balance between model capacity and computational efficiency, with a single-sample inference latency of less than 10 milliseconds, meeting real-time business requirements and effectively capturing the temporal dynamic evolution characteristics of the snow melting process.

[0078] Step 4: Context-aware gating fusion mechanism.

[0079] This step differs from the simple feature stitching in existing related technologies. This embodiment designs a context-aware gating fusion mechanism, which enables the model to dynamically determine the fusion ratio of the original features and the differential features based on the current meteorological background.

[0080] (1) Context vector extraction. First, the context vector of the current meteorological background is extracted from the input data. : ; in, It is a lightweight context encoding network (preferably a single-layer one-dimensional convolution and global average pooling). This is for feature concatenation operations. Context vector. It comprehensively reflects the overall meteorological status of the current time window (such as whether it is in a period of rapid warming or a period of stable warming, or whether it is in the early stage of snow melting or the late stage of snow melting, etc.).

[0081] (2) Gating weight calculation. Gating weight Dynamic generation based on dual encoder output and context vector: ; in, It is the Sigmoid activation function. and These are the learnable weight matrix and the bias term, respectively.

[0082] Gating weights Each dimension takes a value between 0 and 1, which determines the fusion ratio of the original feature encoding and the differential feature encoding in that dimension.

[0083] (3) Dynamic feature fusion. The final fused feature vector is: ; in, This represents element-wise multiplication (Hadamard product).

[0084] The physical meaning of gating mechanisms: when When the value approaches 1, the model relies more on the original observation features (suitable for scenarios where actual atmospheric precipitation dominates); when... When the value approaches 0, the model relies more on differential physical features (suitable for scenarios where the snow melting process dominates); when... When the value is around 0.5, the model assigns similar weights to the two types of features (suitable for ambiguous scenarios such as concurrent snowmelt and precipitation).

[0085] This dynamic fusion mechanism enables the model to adaptively adjust the credibility of information sources based on meteorological background, and has a stronger scene adaptability compared to feature splicing with fixed weights.

[0086] Step 5: Classification output and quality control code marking.

[0087] (1) False precipitation probability prediction. This involves fusing features... Input a classification header, output the probability of false precipitation. : ; in, and These are the learnable parameters for the classification head. This is a randomly deactivated layer (used to prevent overfitting).

[0088] (2) Quality control code marking. Based on the preset decision threshold. (Preferred value is 0.5 to 0.8, which can be adjusted according to business needs), mark the precipitation data for the target time with quality control codes: ; Among them, quality control code 7 indicates that the precipitation data at that time was identified as false precipitation due to snow melting, and it is recommended to remove it or mark it specially in subsequent business applications.

[0089] In this embodiment, the preferred decision threshold is 0.7. The decision threshold can be adjusted according to business needs. If a high recall rate is required (to minimize the omission of false precipitation), the threshold can be lowered to 0.5; if a high accuracy rate is required (to minimize false judgments), the threshold can be increased to 0.8. This embodiment does not impose specific restrictions on this.

[0090] Using the training dataset and the overall loss function, the false precipitation probability prediction model, which includes a temporal encoder, a lightweight context encoder network, and a classification head, is trained in this embodiment to obtain a trained false precipitation probability prediction model. The model training and overall loss function design include the following: (1) Training data labeling.

[0091] Since historical observation data typically lacks explicit labels indicating whether it is "spurious precipitation due to snowmelt," this embodiment uses a method based on physical rules and manual observation annotation to construct the training dataset: Positive sample (spurious precipitation) labeling conditions (must be met simultaneously): snow cover exists, snow depth cm; Temperature above the snowmelt threshold (Preferred temperature: 0℃); Snow depth shows a decreasing trend. (Continuous ≥2 hours); Rain gauge records precipitation, mm / h; (optional) Weather phenomenon records are no precipitation or confirmed by manual sampling as no actual precipitation process.

[0092] Negative sample (actual precipitation) labeling conditions (meeting any of the following): No snow cover ( And the temperature is above 0℃, with recorded weather phenomena as rain; snow depth is stable or increasing. The weather phenomenon was recorded as precipitation; radar / satellite data confirmed the presence of real precipitation echoes.

[0093] (2) Weighted binary cross-entropy loss.

[0094] Because false snowmelt precipitation is a sparse event in historical data (the ratio of positive to negative samples is usually...), The sample is balanced using a class-weighted strategy. ; in, This represents the weighted binary cross-entropy loss. Indicates the number of samples. For positive class weights, Indicates the number of negative samples. Indicates the number of positive samples. Indicates the predicted first The probability of false precipitation for each sample. Indicates the first The true label of each sample.

[0095] (3) Physically guided auxiliary loss.

[0096] In order to embed physical laws such as mass conservation into the model training process, this embodiment introduces a physical auxiliary loss term. : ; in, This represents the positive rate of change of snow depth (equivalent to the amount of water released during snowmelt). These are the values ​​recorded by the rain gauge during the same period. The snow melting conversion efficiency coefficient is preferably 0.85 to 0.95, which can be adjusted according to the regional snow density characteristics.

[0097] The meaning of this loss term is that, under snowmelt conditions, the amount of water released by the decrease in snow depth should be roughly equivalent to the increment recorded by the rain gauge. If the model predicts a sample as spurious precipitation, but the physical auxiliary loss is large (i.e., the amount of snow depth decrease is seriously mismatched with the rain gauge increment), the model will be penalized, thereby guiding the model to learn a physically consistent discrimination logic.

[0098] (4) Total loss function.

[0099] ; in, Represents the total loss function. The physical loss balance coefficient (preferably 0.05 to 0.2) Regularization coefficient (preferred) ), For all trainable parameters of the spurious precipitation probability prediction model, This represents the square of the L2 norm.

[0100] The original observation sequence and differential physical feature sequence corresponding to the data to be predicted are input into the trained false precipitation probability prediction model, and the model can output the false precipitation probability corresponding to the data to be predicted.

[0101] Reference Figure 3 This application also provides a false precipitation data identification system, which includes: The data preprocessing module 100 is used to acquire multi-source meteorological observation data including liquid precipitation, air temperature and snow depth; and to preprocess the multi-source meteorological observation data to obtain the original observation sequence. The feature extraction module 200 is used to extract the precipitation difference, temperature difference, and snow depth difference corresponding to liquid precipitation, air temperature, and snow depth based on the original observation sequence. The data calculation module 300 is used to calculate the snowmelt index based on the current temperature and the current snow depth difference; and to obtain the cumulative snowmelt equivalent by accumulating the snow depth differences of all snowmelt conditions within the historical time window. The sequence construction module 400 is used to construct differential physical feature sequences from precipitation difference, temperature difference, snow depth difference, snow melt index and cumulative snow melt equivalent. The model prediction module 500 is used to construct a false precipitation probability prediction model that includes a temporal encoder, a lightweight context encoding network, and a classification head; the original observation sequence and the differential physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context encoding network includes one-dimensional convolution and global average pooling; The data quality control module 600 is used to mark precipitation data with quality control codes based on the probability of false precipitation.

[0102] It should be noted that since the false precipitation data identification system in this embodiment is based on the same inventive concept as the false precipitation data identification method described above, the corresponding content in the method embodiment is also applicable to this system embodiment, and will not be described in detail here.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations. The acquisition, storage, use and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0104] Reference Figure 4 This application also provides an electronic device, which includes: At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement the false precipitation data identification method described above in this application.

[0105] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0106] The electronic devices according to embodiments of this application will now be described in detail.

[0107] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and called by the processor 1600 to execute the false precipitation data identification method of the embodiments of this application.

[0108] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0109] This application also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described method for identifying false precipitation data.

[0110] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0111] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0112] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0113] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0115] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0116] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0118] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0119] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] The embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of this application.

Claims

1. A method for identifying false precipitation data, characterized in that, The method includes: Acquire multi-source meteorological observation data including liquid precipitation, air temperature, and snow depth; preprocess the multi-source meteorological observation data to obtain the original observation sequence; Based on the original observation sequence, the precipitation difference, temperature difference, and snow depth difference corresponding to the liquid precipitation, the air temperature, and the snow depth are extracted respectively; The snowmelt index is calculated based on the current temperature and the current snow depth difference; the cumulative snowmelt equivalent is obtained by accumulating the snow depth differences of all snowmelt conditions within the historical time window. The precipitation difference, the temperature difference, the snow depth difference, the snowmelt index, and the cumulative snowmelt equivalent are constructed into a differential physical feature sequence; A false precipitation probability prediction model is constructed, which includes a temporal encoder, a lightweight context encoding network, and a classification head; the original observation sequence and the differential physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context encoding network includes one-dimensional convolution and global average pooling; Based on the false precipitation probability, the precipitation data is marked with a quality control code.

2. The method for identifying false precipitation data according to claim 1, characterized in that, The step of extracting the precipitation difference, temperature difference, and snow depth difference corresponding to the liquid precipitation, air temperature, and snow depth based on the original observation sequence includes: The difference between the liquid precipitation at two adjacent moments in the original observation sequence is calculated to obtain the precipitation difference. The temperature difference is obtained by calculating the difference between the temperatures of two adjacent moments in the original observation sequence; The snow depth difference is obtained by calculating the difference between two adjacent times in the original observation sequence.

3. The method for identifying false precipitation data according to claim 1, characterized in that, The calculation of the snowmelt index based on the difference between the current air temperature and the current snow depth includes: The temperature difference between the current temperature and the preset snow melting threshold temperature is calculated to obtain the temperature difference result. The temperature difference result is compared with zero, and the first maximum value is selected; Take the inverse of the snow depth difference at the current moment to obtain the inverse value of the snow depth difference; The opposite value of the snow depth difference is compared with zero, and the second maximum value is selected; Multiply the first maximum value and the second maximum value to obtain the snow melting index.

4. The method for identifying false precipitation data according to claim 1, characterized in that, The cumulative snowmelt equivalent is obtained by accumulating the snow depth differences of all snowmelt conditions within a historical time window, including: Construct a snow melting condition where the temperature is higher than the snow melting threshold temperature, and the value is 1. The cumulative snowmelt equivalent is obtained by taking the negative of the snow depth difference for all snowmelt conditions that meet the conditions within the cumulative historical time window.

5. The method for identifying false precipitation data according to claim 1, characterized in that, The time encoder includes a first time encoder and a second time encoder. The trained false precipitation probability prediction model is trained through a total loss function, which is constructed by weighted binary cross-entropy loss and physical-guided auxiliary loss. The step of inputting the original observation sequence and the differential physical feature sequence into the trained false precipitation probability prediction model to obtain the false precipitation probability includes: The original observation sequence is input into the first time encoder to extract the first global time encoding vector of the original observation; The differential physical feature sequence is input into the second temporal encoder to extract the second global temporal encoding vector of the differential physical features; The original observation sequence and the differential physical feature sequence are concatenated to obtain the sequence concatenation result; The sequence concatenation result is input into the lightweight context coding network to extract the context vector; Based on the first global temporal coding vector, the second global temporal coding vector, and the context vector, calculate the gating weights; Based on the gating weights, the first global temporal coding vector and the second global temporal coding vector are fused to obtain a fused feature vector. The fused feature vector is input into the classification head to obtain the false precipitation probability.

6. The method for identifying false precipitation data according to claim 5, characterized in that, The calculation of the gating weights based on the first global temporal coding vector, the second global temporal coding vector, and the context vector includes: The first global temporal coding vector, the second global temporal coding vector, and the context vector are concatenated to obtain the vector concatenation result; The concatenated vectors are multiplied by the weight matrix to obtain the multiplication result; the multiplication result is then added to the bias term to obtain the summation result. The summation result is input into the activation function to obtain the gating weights.

7. The method for identifying false precipitation data according to claim 5, characterized in that, The step of fusing the first global temporal coding vector and the second global temporal coding vector based on the gating weights to obtain a fused feature vector includes: The first global temporal coding vector is multiplied element-wise with the gate weight to obtain the first element-wise multiplication result; The subtraction result is obtained by subtracting the gate weight; the second global temporal coding vector is multiplied element-wise with the subtraction result to obtain the second element-wise multiplication result. The first element-wise multiplication result and the second element-wise multiplication result are added together to obtain the fused feature vector.

8. A false precipitation data identification system, characterized in that, The system includes: The data preprocessing module is used to acquire multi-source meteorological observation data including liquid precipitation, air temperature, and snow depth; and to preprocess the multi-source meteorological observation data to obtain the original observation sequence. The feature extraction module is used to extract the precipitation difference, temperature difference, and snow depth difference corresponding to the liquid precipitation, the air temperature, and the snow depth, respectively, based on the original observation sequence. The data calculation module is used to calculate the snowmelt index based on the current temperature and the current snow depth difference; and to obtain the cumulative snowmelt equivalent by accumulating the snow depth differences of all snowmelt conditions within the historical time window. The sequence construction module is used to construct the precipitation difference, the temperature difference, the snow depth difference, the snowmelt index, and the cumulative snowmelt equivalent into a differential physical feature sequence; The model prediction module is used to construct a false precipitation probability prediction model that includes a temporal encoder, a lightweight context encoding network, and a classification head; the original observation sequence and the differential physical feature sequence are input into the trained false precipitation probability prediction model to obtain the false precipitation probability; the lightweight context encoding network includes one-dimensional convolution and global average pooling; The data quality control module is used to mark the precipitation data with quality control codes based on the false precipitation probability.

9. An electronic device, characterized in that, It includes at least one processor and a memory for communicatively connecting to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the false precipitation data identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the false precipitation data identification method as described in any one of claims 1 to 7.

Images