Artificial influence weather operation benefit evaluation method and system based on meteorological parameter inversion

Abstract

The invention relates to the technical field of meteorological big data analysis, and discloses an artificial influence weather operation benefit evaluation method and system based on meteorological parameter inversion, and the method comprises the steps: constructing a structure causal model comprising a meteorological background, catalytic operation, cloud microphysics and ground precipitation nodes, and recognizing a mixed path; training a physical information neural network with embedded physical residual constraint by using historical data; connection of meteorological background pointing to catalytic operation is cut off in the causal model, operation nodes are set to be in a non-operation state, an actual meteorological background input network is combined for anti-fact inversion, and an anti-fact ground precipitation field is obtained; and performing difference calculation on the actual observation value and the anti-fact field to obtain an evaluation result. According to the method, through causal anti-fact deduction under physical constraints, natural meteorological fluctuation interference is effectively stripped, and objective, quantitative and physical interpretability evaluation of operation benefits is realized.

Description

Technical Field

[0001] This invention relates to the field of meteorological big data analysis and weather modification assessment technology, and in particular to a method and system for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion. Background Technology

[0002] my country is one of the earliest countries to carry out weather modification work, which plays an important role in serving agricultural production, ensuring water resource supply, and disaster prevention and mitigation. However, the scientific, objective, and quantitative evaluation of the operational benefits has always been a challenge for the industry.

[0003] Existing assessment methods are mainly divided into statistical and physical tests. Statistical tests rely primarily on ground-based rainfall observation data, resulting in a single data source and difficulty in reflecting the spatial distribution characteristics of cloud precipitation. While physical tests utilize equipment such as radar to detect cloud parameters, they often lean towards qualitative analysis. More importantly, the core difficulty of existing assessments lies in estimating the "background field," namely, accurately answering the counterfactual question of "how the weather would have naturally evolved if operations had not been carried out." Traditional methods often employ regional comparison or historical regression methods, which are easily influenced by subjective factors such as topographical and climatic differences when selecting "comparison areas," and struggle to completely eliminate the confounding effects of natural meteorological fluctuations (such as changes in natural precipitation itself) on the assessment results. Furthermore, purely data-driven AI models lack physical constraints, easily generating false predictions that violate atmospheric physical conservation laws, leading to insufficient credibility and interpretability of the assessment results. Summary of the Invention

[0004] This invention provides a method and system for evaluating the benefits of weather modification operations based on meteorological parameter inversion. By using causal graph reverse fact deduction under physical constraints, the interference of natural weather fluctuations is effectively eliminated, achieving an objective, quantitative, and physically interpretable accurate evaluation of the benefits of weather modification operations.

[0005] This invention provides a method for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion, comprising: S1. Construct a structural causal model of the weather modification operation process. The structural causal model is a directed acyclic graph, including meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes. The meteorological background nodes point to the catalyst operation nodes and surface precipitation nodes to form hybrid paths. The catalyst operation nodes point to the cloud microphysical parameter nodes, and the cloud microphysical parameter nodes point to the surface precipitation nodes. S2. Acquire historical multi-source data of the target operation area and train a physical information neural network to establish a mapping relationship from meteorological background nodes and catalyst operation nodes to cloud microphysical parameter nodes and ground precipitation nodes; wherein, the loss function of the physical information neural network is composed of a weighted average of a data fitting loss term and a physical residual constraint term. S3. Obtain the current actual meteorological background data of the operation impact area within the target operation area, cut off the connection between the meteorological background node and the catalyst operation node in the structural causal model, set the value of the catalyst operation node to the no-operation state value, and input it together with the actual meteorological background data into the physical information neural network to invert and obtain the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, as well as the counterfactual surface precipitation field calculated based on the cloud microphysical parameter evolution field. S4. Obtain the actual ground precipitation observation value of the operation impact area during the current operation period, perform differential calculation between the actual ground precipitation observation value and the counterfactual ground precipitation field, and obtain the artificial weather modification operation benefit evaluation result after eliminating the interference of natural meteorological fluctuations.

[0006] Furthermore, S1 specifically includes: S101. Define the vertex set V in the graph theory model G=(V,E), including meteorological background nodes X, catalyst operation nodes T, cloud microphysical parameter nodes M, and surface precipitation nodes Y; wherein, the meteorological background nodes X include dynamic factors, thermal factors, and water vapor factors; the catalyst operation nodes T include catalyst application dosage, application height, application time, and operation trajectory coordinates; the cloud microphysical parameter nodes M include radar echo characteristic parameters inside the cloud and cloud top temperature; the surface precipitation nodes Y include the cumulative precipitation observed by ground rain gauges or the quantitatively estimated precipitation QPE value by radar; S102. Based on the decision-making logic of artificial weather modification operations and the laws of natural precipitation, a directed edge from meteorological background node X to catalyst operation node T and a directed edge from meteorological background node X to ground precipitation node Y are established, thus forming a hybrid path with the structure T←X→Y. S103. Based on the principle of cloud physics catalysis and the cloud precipitation conversion mechanism, a directed edge is established from the catalyst operation node T to the cloud microphysical parameter node M, and a directed edge is established from the cloud microphysical parameter node M to the ground precipitation node Y, thus forming a causal transmission path with the structure T→M→Y. S104. Combine the meteorological background node X, catalyst operation node T, cloud microphysical parameter node M, and ground precipitation node Y with directed edges to generate the structural causal model.

[0007] Furthermore, S2 specifically includes: S201. Determine the target operation area and historical backtracking period, extract historical observation data corresponding to the meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes, perform spatiotemporal interpolation and normalization processing on the historical observation data, and construct a tensor set containing input features. and label feature tensor set The training dataset; S202, Constructing a physical information neural network model It includes an encoder for extracting spatiotemporal features, an attention module for feature fusion, and a decoder for outputting a prediction field. The physical information neural network model has an embedded automatic differentiation module for calculating the partial derivatives of the network output variables with respect to spatiotemporal coordinates. S203. Define the total loss function of the physical information neural network model. The total loss function consists of a data fitting loss term that measures the prediction error. and physical residual constraint terms that measure physical conservation Weighted composition; S204. Input the training dataset into the physical information neural network model, and minimize the total loss function using a gradient-based optimization algorithm. Update the network weight parameters until the model converges.

[0008] Furthermore, in S202, the physical information neural network model The architecture employs a hybrid approach of 3D-ConvLSTM and a physical constraint layer, specifically including: The encoder employs a multi-layer 3D convolutional long short-term memory network (3D-ConvLSTM) to extract meteorological background tensors. and catalyst working tensor The spatiotemporal evolution characteristics; The fusion layer uses an attention mechanism to weightedly fuse meteorological background features and catalyst operation features to generate a latent state vector. The decoder maps the latent state vectors to the cloud microphysical parameter evolution field. and surface precipitation field The predicted output.

[0009] Furthermore, in S203, the total loss function The calculation formula is: ; ; ; in, is the hyperparameter for the physical constraint weights; N is the number of training samples; These are the cloud microphysical parameter predictions and surface precipitation predictions for the i-th sample output by the network, respectively. These are the historical true observations of the i-th sample; The number of physical constraint collocation points sampled in the spatiotemporal domain; To describe the partial differential equation operator for the physical conservation relationship of cloud water, it is constructed based on the cloud water resource continuity equation, and its specific physical constraint equation form is as follows: ; in, Cloud microphysical parameters output from the network The predicted value of cloud water content extracted from it; Surface precipitation field output by the network Precipitation rate forecast in the data; t is the time coordinate. V represents the local temporal variation of cloud water content; V is the wind field vector extracted from the input meteorological background data. The cloud water flux divergence term represents the cloud water transport process driven by the wind field. The source term parameters for water vapor condensation to form cloud water; These are the aggregate parameters for cloud water evaporation and dissipation.

[0010] Furthermore, S3 specifically includes: S301. Based on the current operation implementation parameters and upper-level wind field data, determine the operation impact area in the target operation area, and obtain the actual meteorological background data of the operation impact area during the current operation period; wherein, the operation impact area is the coverage area in the target operation area where the catalyst concentration is greater than a preset threshold; S302. Based on the intervention logic of the structural causal model, the connection between the meteorological background node and the catalyst operation node is cut off, and a counterfactual operation tensor is constructed to force the value of the catalyst operation node to be a no-operation state value. This tensor is combined with the actual meteorological background data to form a counterfactual intervention input tensor. The no-operation state value is a zero vector or null value encoding that matches the dimension of the actual meteorological background data. S303. Input the counterfactual intervention input tensor into the trained physical information neural network model to perform forward propagation operation, and use the network decoder to deduce the natural evolution process of the cloud system under the constraints of the physical equation. S304. Output the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, and the counterfactual surface precipitation field calculated based on the evolution field; wherein, the counterfactual surface precipitation field is obtained by the decoder of the physical information neural network model based on the output cloud microphysical parameter evolution field, through the cloud water-precipitation transformation mechanism learned within the network.

[0011] Furthermore, S4 specifically includes: S401. Obtain the ground rain gauge data and weather radar estimated precipitation data of the operation impact area during the current operation period. Through fusion calibration and spatiotemporal resampling, construct and generate the actual ground precipitation observation tensor consistent with the output grid of the physical information neural network. Specifically, using the Kalman filter method or optimal interpolation method, with the ground rain gauge observation value as the benchmark, perform deviation correction on the quantitative estimated precipitation product (QPE) of the weather radar to generate a high-precision gridded real precipitation field. Map the real precipitation field onto a preset four-dimensional grid to obtain the actual ground precipitation observation tensor. S402. Perform grid-by-grid and time-by-time difference operations on the actual surface precipitation observation tensor and the counterfactual surface precipitation field to obtain the initial net operational influence field containing positive and negative increments. S403. Using the physical residual distribution learned by the physical information neural network during the training phase, construct the upper bound of the confidence interval of the counterfactual reference field, and retain only the grid point values ​​where the actual ground precipitation observation values ​​are significantly higher than the upper bound of the confidence interval to obtain the corrected effective net rainfall field. S404. Perform spatiotemporal integration on the effective net rainfall field to calculate the absolute rainfall increase and relative rainfall rate of the weather modification operation; perform cumulative calculation on the effective net rainfall field along the time dimension to obtain a two-dimensional matrix, use color mapping technology to convert it into an isosurface layer, and overlay it with the geographic base map, catalyst operation trajectory and wind field streamline layer to generate a spatial distribution map of the rainfall enhancement effect; output the weather modification operation benefit evaluation result.

[0012] Furthermore, in S403, the upper bound of the confidence interval The calculation formula is: ; in, The predicted value in the counterfactual surface precipitation field; is the standard deviation of the physical residuals during the training process of the physical information neural network; k is the confidence coefficient, with a value of 1.96, corresponding to a 95% confidence level; is the actual ground precipitation observation value at a certain time point if and only if... If the rain enhancement effect of the grid point is effective, then the initial net operational impact field value corresponding to the grid point is set to 0. In S404, the formula for calculating the absolute rainfall increase is: ; The formula for calculating the relative rainfall increase rate is: ; in, This is the corrected effective net rainfall field; Representing coordinates and time t.

[0013] This invention also provides a system for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion. Based on the aforementioned method for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion, the system includes: A construction module is used to construct a structural causal model of the weather modification operation process. The structural causal model is a directed acyclic graph, including meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes. The meteorological background nodes point to the catalyst operation nodes and surface precipitation nodes to form hybrid paths. The catalyst operation nodes point to the cloud microphysical parameter nodes, and the cloud microphysical parameter nodes point to the surface precipitation nodes. The acquisition module is used to acquire historical multi-source data of the target operation area and train a physical information neural network to establish a mapping relationship from meteorological background nodes and catalyst operation nodes to cloud microphysical parameter nodes and ground precipitation nodes; wherein, the loss function of the physical information neural network is composed of a weighted average of a data fitting loss term and a physical residual constraint term. The inversion module is used to obtain the current actual meteorological background data of the operation impact area within the target operation area, cut off the connection between the meteorological background node and the catalyst operation node in the structural causal model, set the value of the catalyst operation node to the no-operation state value, and input it together with the actual meteorological background data into the physical information neural network to invert and obtain the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, as well as the counterfactual surface precipitation field calculated based on the cloud microphysical parameter evolution field. The output module is used to obtain the actual ground precipitation observation values ​​of the operation impact area during the current operation period, and to perform differential calculation between the actual ground precipitation observation values ​​and the counterfactual ground precipitation field to obtain the artificial weather modification operation benefit evaluation results after eliminating the interference of natural meteorological fluctuations.

[0014] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0016] The beneficial effects of this invention are as follows: This invention constructs a structured causal model (SCM) incorporating meteorological background, catalytic operations, cloud microphysics, and surface precipitation nodes. This model identifies and severs confounding paths caused by the meteorological background, logically decoupling the causal relationship between natural fluctuations and the effects of human intervention. It utilizes a Physical Information Neural Network (PINN) with embedded physical residual constraints such as the cloud water resource continuity equation to establish a mapping relationship conforming to physical conservation laws. Furthermore, through a counterfactual inversion technique that forces intervention nodes to zero, it accurately deduces the evolution of natural cloud systems and the counterfactual precipitation field under the hypothetical no-operation scenario. This method overcomes the subjectivity and uncertainty of traditional comparison area selection, solves the problem of the uninterpretable "black box" of purely data-driven models, and achieves an objective, quantitative, and high-precision assessment of the net benefits of weather modification operations under physical mechanism constraints. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the device structure according to an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0020] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0022] like Figure 1 As shown, this invention provides a method for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion, including: S1. Construct a structural causal model of the weather modification operation process. The structural causal model is a directed acyclic graph, including meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes. The meteorological background nodes point to the catalyst operation nodes and surface precipitation nodes to form hybrid paths. The catalyst operation nodes point to the cloud microphysical parameter nodes, and the cloud microphysical parameter nodes point to the surface precipitation nodes.

[0023] In a preferred embodiment of the present invention, step S1 specifically includes the following sub-steps S101-S104: S101. Based on the principles of atmospheric physics and the procedures for artificial weather modification operations, define the vertex set V in the structural causal model G=(V,E). The vertex set V contains four types of key nodes: meteorological background nodes X, catalyst operation nodes T, cloud microphysical parameter nodes M, and surface precipitation nodes Y. Specifically, (1) Meteorological background node X: This node characterizes the environmental background field that is unaffected by human intervention but simultaneously determines operational decisions and natural precipitation processes. In this embodiment, the node contains a multi-dimensional physical feature vector, specifically including: Dynamic factors: vertical wind shear at 700 hPa and 850 hPa, and surface roughness (characterizing the effect of topographic forced uplift). Thermal factors: convective available potential energy, K index (characterizing atmospheric stratification stability); Water vapor factors: total precipitable water content in the atmosphere and specific humidity at the cloud base.

[0024] (2) Catalyst operation node T: used to characterize the intensity and spatiotemporal distribution of human intervention. This node includes the type of operation tool (aircraft / rocket / anti-aircraft gun), the application dosage of the catalyst (such as silver iodide AgI), the application altitude, the application time, and the operation trajectory coordinates.

[0025] (3) Cloud Microphysical Parameter Node M: Used to characterize the microphysical state inside the cloud directly affected by the catalyst. To accurately capture the phase changes caused by the catalytic operation, this node preferably includes characteristic parameters based on dual-polarization radar inversion, specifically: combined reflectivity factor. Differential reflectivity (Characterizing particle shape and phase state), relative differential propagation phase shift (Characterizing liquid water content) and cloud top temperature (CTT).

[0026] (4) Surface precipitation node Y: used to characterize the final operational effect, i.e. the cumulative precipitation during the period observed by the surface rain gauge or the quantitatively estimated precipitation by radar (QPE).

[0027] S102. Establish directed edges X→T and X→Y to form a mixed path.

[0028] Constructing the connection X→T: Based on the business logic of weather modification operations, the command personnel need to decide whether to carry out the operation and the dosage based on the current weather conditions (i.e., the meteorological background node X). Therefore, the meteorological background is the "cause" of the operation, forming a directed edge from the meteorological background node to the catalyst operation node.

[0029] Constructing the X→Y connection: Based on natural precipitation patterns, regardless of whether operations are conducted, meteorological background conditions (such as water vapor convergence and orographic lifting) are the direct driving force for precipitation. Therefore, a directed edge is formed from the meteorological background node to the surface precipitation node.

[0030] Formation of confounding paths: The two paths described above together constitute the structure T←X→Y. In this structure, the meteorological background node X, as a common cause, is a confounding factor that leads to a spurious correlation between the catalyst operation node T and the surface precipitation node Y.

[0031] S103. Establish directed edges T→M and M→Y to form a causal transmission chain.

[0032] Constructing the T→M connection: Based on cloud physics catalysis principles (such as the Bergeron process), catalyst seeding directly alters the distribution of condensation nuclei within the cloud, thereby changing the scale width and phase conversion efficiency of cloud droplets / ice crystals. Therefore, the catalyst operating node T directly points to the cloud microphysical parameter node M.

[0033] Constructing the connection from M to Y: Based on the precipitation formation mechanism, surface precipitation is formed by the collision, coalescence, and growth of water-forming particles (cloud water, cloud ice) in clouds before they fall. Changes in cloud microphysical states are a direct precursor to changes in surface precipitation. Therefore, the cloud microphysical parameter node M directly points to the surface precipitation node Y.

[0034] S104. Combine the nodes and directed edges defined in steps S101 to S103 to formally express them as a directed acyclic graph of the structural causal model. In this graph, there are two paths from the catalyst operation node T to the ground precipitation node Y: The front path: T→M→Y, represents the real physical causal effect of artificial rain enhancement operations; The backdoor path, T←X→Y, represents the interference effect introduced by natural meteorological background fluctuations.

[0035] By constructing this directed acyclic graph, we can clarify the specific logical links that need to be cut off in subsequent steps through counterfactual inversion (i.e., cutting off the backdoor path from X to T).

[0036] S2. Obtain historical multi-source data of the target operation area and train a physical information neural network to establish a mapping relationship from meteorological background nodes and catalyst operation nodes to cloud microphysical parameter nodes and ground precipitation nodes; wherein, the loss function of the physical information neural network is composed of a weighted average of data fitting loss term and physical residual constraint term.

[0037] In a preferred embodiment of the present invention, step S2 specifically includes the following sub-steps S201-S204: S201. Determine the target operation area (e.g., a specific watershed or windward slope of a mountainous area) and the historical backtracking period (e.g., the past 5-10 years of artificial rain enhancement operation seasons). Extract the historical observation data corresponding to each node defined in step S1 from the multi-source database, and perform spatiotemporal alignment and tensor quantization processing. Specifically: Data Acquisition: Obtain historical meteorological background data for the target work area. (such as ERA5 reanalysis data or sounding station data), historical catalytic operation records (Operation location, time, dose), cloud microphysical parameters retrieved from historical dual-polarization radar. and historical surface precipitation observations .

[0038] Preprocessing: The above data is spatiotemporally interpolated and uniformly mapped onto a preset four-dimensional grid (H×W×D×Time).

[0039] Normalization: Physical quantities with different dimensions (such as radar reflectivity dBZ and rainfall mm) are Z-score normalized to construct the input tensor set. and label tensor .

[0040] S202, Constructing a deep neural network model (Physical Information Neural Network PINN architecture) is used to approximate a nonlinear mapping function from input (background + task) to output (cloud + rain). In this embodiment, the physical information neural network adopts a hybrid architecture of 3D-ConvLSTM + physical constraint layer, specifically including: Encoder: Employs a multi-layer 3D convolutional long short-term memory network (3D-ConvLSTM) to extract meteorological background tensors. and catalyst working tensor The spatiotemporal evolution characteristics; Fusion layer: Uses an attention mechanism to weightedly fuse background features and task features to generate a latent state vector; Decoder: Outputs the predicted evolution field of cloud microphysical parameters (Including cloud water content) Cloud ice content (etc.) and surface precipitation fields (Precipitation rate R).

[0041] Automatic Differentiation Module: This network incorporates an automatic differentiation mechanism, capable of calculating the partial derivatives of the network output variables with respect to the time coordinate t and the spatial coordinates (x, y, z) (e.g., , This provides a basis for calculating physical residuals.

[0042] S203. To ensure that the model prediction results conform to atmospheric physical laws, a total loss function is defined. Loss term from data fitting and physical residual constraint terms The weighted composition, its mathematical expression is: ; in, These are the hyperparameters for the physical constraint weights.

[0043] (1) Data fitting loss term This is used to measure the error between the network's predicted output and historical actual observations, calculated using the mean squared error (MSE). ; Where N is the number of training samples; These are the cloud microphysical parameter predictions and surface precipitation predictions for the i-th sample output by the network, respectively. These are the historical true observations of the i-th sample.

[0044] (2) Physical residual constraint terms This system is built upon the cloud water resource continuity equation (mass conservation). The network outputs predicted cloud water content. (Source: ) and surface precipitation rate forecast (Source: The following partial differential equation (PDE) must be satisfied: ; in, Cloud microphysical parameters output from the network The predicted value of cloud water content extracted from it; Surface precipitation field output by the network Precipitation rate forecast in the data; t represents the time coordinate; The local variation term of cloud water content is calculated through automatic differentiation; The cloud water flux divergence term represents the transport of cloud water by the wind field V, where V comes from the input data. ; and These are the parameterized condensation source term and evaporation sink term, respectively. The physical residual loss is the norm of the residuals in the above equations: ; Here This represents the number of collocation points randomly sampled in the spatiotemporal domain. Even at spatiotemporal points where there is no observation data, the physical equations must hold.

[0045] S204. Input the constructed dataset into the physical information neural network model, and minimize the total loss function using a gradient-based optimization algorithm (such as the Adam optimizer combined with the L-BFGS optimizer). Specifically, during forward propagation, the network outputs predicted values ​​and calculates... Simultaneously, the physical residuals are obtained by automatically differentiating and calculating the partial derivatives of each term in the physical equations. Based on total losses Update the network weights θ. Stop training when the total loss function drops to a preset threshold or no longer decreases significantly. At this point, the trained physical information neural network model possesses the mapping ability to both conform to the characteristics of historical observation data and follow the physical conservation laws of cloud precipitation.

[0046] S3. Obtain the current actual meteorological background data of the operation impact area within the target operation area, disconnect the connection between the meteorological background node and the catalyst operation node in the structural causal model, set the value of the catalyst operation node to the no-operation state value, and input it together with the actual meteorological background data into the physical information neural network to invert and obtain the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, as well as the counterfactual surface precipitation field calculated based on the cloud microphysical parameter evolution field.

[0047] In a preferred embodiment of the present invention, step S3 specifically includes the following sub-steps S301-S304: S301. To achieve accurate assessment, it is necessary to pinpoint the specific impact area of ​​the operation from the macroscopic target operation area. Specifically, based on the implementation parameters of the current operation (operation point coordinates, time, tool type), combined with real-time upper-level wind field data, the catalyst transport and diffusion path is simulated using a catalyst diffusion model (such as a Gaussian plume model or a Lagrange particle diffusion model). The coverage area where the catalyst concentration exceeds a preset threshold is defined as the operation impact area for this assessment. The impact area during the operation period (…) is then determined. to Actual meteorological background data within) This data includes real-time dynamic factors (such as wind field), thermal factors (such as sounding temperature), and water vapor conditions, but does not include cloud and precipitation conditions affected by the operation.

[0048] S302. Based on the intervention logic of the structural causal model, construct the model input tensor for inferring the state of nature. Specifically, Logically severing the connection from the meteorological background node to the catalyst operation node means no longer considering the decision relevance of "operation is carried out because of bad weather", but instead forcibly examining the scenario of "if no operation is carried out under the same weather conditions".

[0049] Construct a system with actual meteorological background data Dimension-matched catalyst job tensor .Will All values ​​representing operational dose and dissemination intensity are forcibly set to "no-operation state values" (usually 0 vectors or null values) to simulate... Physical intervention. The actual observed meteorological background tensor... With the operation tensor that is forced to zero By splicing, a counterfactual input tensor set is formed. .

[0050] S303. Input the constructed counterfactual data into the tensor set. The input is fed into the Physics-Informed Neural Network (PINN) that has been trained in step S2, and a forward propagation operation is performed. Feature mapping, specifically: Encoder extraction of the network Due to the natural evolution characteristics of the input job features being 0, the network will ignore the forcing terms of the catalyst on the cloud physical process.

[0051] The decoder uses the trained mapping weights to deduce the natural generation and dissipation process that the cloud system should exhibit without external intervention, under the implicit constraints of the physical equation (cloud-water continuity equation).

[0052] The S304 and physical information neural network decoder ultimately output two key sets of spatiotemporal evolution fields, serving as a "baseline map" for evaluating benefits. These include: Cloud microphysical parameter inversion field : Characterizes the evolution trajectory of radar reflectivity, cloud water content, and cloud ice content of clouds under natural conditions.

[0053] Counterfactual surface precipitation field Based on The theoretical natural precipitation is calculated using the "cloud-rain conversion mechanism" learned internally by the network. This field represents the amount of rainfall that should actually fall to the ground if no operations were carried out.

[0054] S4. Obtain the actual ground precipitation observation value of the operation impact area during the current operation period, perform differential calculation between the actual ground precipitation observation value and the counterfactual ground precipitation field, and obtain the artificial weather modification operation benefit evaluation result after eliminating the interference of natural meteorological fluctuations.

[0055] In a preferred embodiment of the present invention, step S4 specifically includes the following sub-steps S401-S404: S401, Obtain the area affected by the operation during the current operation period ( to Actual surface precipitation data within ) To ensure consistency in spatial resolution with the counterfactual reference field output in step S3, a high-precision gridded real-time precipitation field is generated using a radar-rain gauge fusion calibration method. Specifically: This study integrates rain gauge data (point data) from high-density automatic weather stations with quantitative precipitation products (QPE, area data) from weather radar. Using Kalman filtering or optimal interpolation (OI), bias corrections are applied to the radar QPE field based on ground rain gauge observations. The corrected actual precipitation field is then mapped onto a four-dimensional grid (H×W×D×Time) identical to the output of the physical information neural network, yielding a standardized tensor of actual ground precipitation observations. .

[0056] S402, convert the actual surface precipitation observation tensor obtained in step S401 into... The counterfactual surface precipitation field output by step S3 (That is, assuming the natural precipitation field before any operation) perform grid-by-grid, hour-by-hour difference operations to calculate the net operational impact field. : ; Among them, if , indicating in coordinates At time t, artificial catalysis produced a positive rain enhancement effect; if This indicates no significant effect or that the catalyst did not function; this process essentially strips away natural weather fluctuations (from the physical layer) The interference of characterization was used to extract the causal effect caused purely by human intervention (T).

[0057] S403. To eliminate false rain enhancement conclusions caused by model prediction errors and random noise, a confidence interval test mechanism based on the uncertainty of PINN (Physical Neural Network Model) is introduced, specifically as follows: The physical residual distribution learned during the training phase using a physical information neural network Construct the upper bound of the 95% confidence interval for the counterfactual baseline field. Only when the actual observed value Significantly higher than the upper confidence bound Time (i.e.) Only when the rainfall enhancement effect in that area at that time is deemed physically significant is the rainfall enhancement effect considered effective. Grid points that fail the significance test are then considered physically significant. Setting it to 0 yields the corrected effective net rainfall field. This ensures the rigor of the evaluation results.

[0058] S404, Based on the corrected effective net rainfall field Through spatiotemporal integration and rendering, the final evaluation results of the artificial weather modification operation benefits are determined from three dimensions: total statistics, spatial distribution, and confidence level. Specifically, this includes: (1) Calculating core benefit indicators (total evaluation): for the effective net rainfall field By performing full-time and spatial-domain integration, the direct economic benefits of this operation are quantified. Absolute rainfall increase: Calculate the total increase in water volume within the operation's impact area.

[0059] ; This indicator represents the increase in net water resources (in ten thousand tons) resulting solely from catalytic operations, excluding fluctuations in natural precipitation.

[0060] Relative rainfall increase rate: ; This indicator represents the extent to which human intervention increases precipitation relative to the natural background precipitation.

[0061] (2) Generate a spatial distribution map of the rain enhancement effect (spatial assessment): The four-dimensional effective net rain enhancement field is generated. The dimensionality reduction mapping is transformed into a two-dimensional visualized geographic distribution map. The specific generation steps are as follows: By accumulating along the time axis, a two-dimensional cumulative rainfall matrix is ​​obtained. .

[0062] matrix The numerical values ​​are mapped to color gradients (e.g., 0 values ​​are set to transparent, and positive values ​​are mapped to light blue to dark blue levels depending on their size). The rendered rain enhancement layer is overlaid on the GIS geographic base map, and a catalyst diffusion trajectory layer and an average wind field streamline layer are simultaneously overlaid.

[0063] If the high-value rain enhancement area (dark area) in the figure is located downwind of the catalyst diffusion trajectory and is consistent with the direction of the wind field streamline, then the assessment result is determined to have physical consistency and is confirmed as the final spatial assessment result.

[0064] (3) Output the final evaluation report: Combine the absolute rainfall increase, relative rainfall increase rate and the generated spatial distribution map of the rain enhancement effect obtained above as the evaluation result of the artificial weather modification operation after removing the interference of natural meteorological fluctuations.

[0065] like Figure 2 As shown, the present invention also provides a system for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion. Based on the weather modification operation effectiveness evaluation method based on meteorological parameter inversion described above, the system includes: Module 1 is used to construct a structural causal model of the weather modification operation process. The structural causal model is a directed acyclic graph, including meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes. The meteorological background nodes point to the catalyst operation nodes and surface precipitation nodes to form hybrid paths. The catalyst operation nodes point to the cloud microphysical parameter nodes, and the cloud microphysical parameter nodes point to the surface precipitation nodes. The acquisition module 2 is used to acquire historical multi-source data of the target operation area and train a physical information neural network to establish a mapping relationship from meteorological background nodes and catalyst operation nodes to cloud microphysical parameter nodes and ground precipitation nodes; wherein, the loss function of the physical information neural network is composed of a weighted average of a data fitting loss term and a physical residual constraint term. Inversion module 3 is used to obtain the current actual meteorological background data of the operation impact area within the target operation area, cut off the connection between the meteorological background node and the catalyst operation node in the structural causal model, set the value of the catalyst operation node to the no-operation state value, and input it together with the actual meteorological background data into the physical information neural network to invert and obtain the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, as well as the counterfactual surface precipitation field calculated based on the cloud microphysical parameter evolution field. Output module 4 is used to obtain the actual ground precipitation observation value of the operation impact area during the current operation period, perform differential calculation between the actual ground precipitation observation value and the counterfactual ground precipitation field, and obtain the artificial weather modification operation benefit evaluation result after eliminating the interference of natural meteorological fluctuations.

[0066] Each of the above modules is used to perform the corresponding steps in the above-mentioned method for evaluating the benefits of artificial weather modification operations based on meteorological parameter inversion. The specific implementation methods are as described in the above-mentioned method embodiments, and will not be repeated here.

[0067] like Figure 3 As shown, the present invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores all data required for the process of evaluating the effectiveness of weather modification operations based on meteorological parameter inversion. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement the weather modification operation effectiveness evaluation method based on meteorological parameter inversion.

[0068] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0069] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion.

[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), such as dynamic RAM (used as main storage) or static RAM (commonly used as cache memory). By way of illustration and not limitation, RAM has various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and Rambus DRAM (RDRAM).

[0071] 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, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0072] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for evaluating the benefits of weather modification operations based on meteorological parameter inversion, characterized in that, include: S1. Construct a structural causal model of the weather modification operation process. The structural causal model is a directed acyclic graph, including meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes. The meteorological background nodes point to the catalyst operation nodes and surface precipitation nodes to form hybrid paths. The catalyst operation nodes point to the cloud microphysical parameter nodes, and the cloud microphysical parameter nodes point to the surface precipitation nodes. S2. Acquire historical multi-source data of the target operation area and train a physical information neural network to establish a mapping relationship from meteorological background nodes and catalyst operation nodes to cloud microphysical parameter nodes and ground precipitation nodes; wherein, the loss function of the physical information neural network is composed of a weighted average of a data fitting loss term and a physical residual constraint term. S3. Obtain the current actual meteorological background data of the operation impact area within the target operation area, cut off the connection between the meteorological background node and the catalyst operation node in the structural causal model, set the value of the catalyst operation node to the no-operation state value, and input it together with the actual meteorological background data into the physical information neural network to invert and obtain the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, as well as the counterfactual surface precipitation field calculated based on the cloud microphysical parameter evolution field. S4. Obtain the actual ground precipitation observation value of the operation impact area during the current operation period, perform differential calculation between the actual ground precipitation observation value and the counterfactual ground precipitation field, and obtain the artificial weather modification operation benefit evaluation result after eliminating the interference of natural meteorological fluctuations.

2. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 1, characterized in that, S1 specifically includes: S101. Define the vertex set V in the graph theory model G=(V,E), including meteorological background nodes X, catalyst operation nodes T, cloud microphysical parameter nodes M, and surface precipitation nodes Y; wherein, the meteorological background nodes X include dynamic factors, thermal factors, and water vapor factors; the catalyst operation nodes T include catalyst application dosage, application height, application time, and operation trajectory coordinates; the cloud microphysical parameter nodes M include radar echo characteristic parameters inside the cloud and cloud top temperature; the surface precipitation nodes Y include the cumulative precipitation observed by ground rain gauges or the quantitatively estimated precipitation QPE value by radar; S102. Based on the decision-making logic of artificial weather modification operations and the laws of natural precipitation, a directed edge from meteorological background node X to catalyst operation node T and a directed edge from meteorological background node X to ground precipitation node Y are established, thus forming a hybrid path with the structure T←X→Y. S103. Based on the principle of cloud physics catalysis and the cloud precipitation conversion mechanism, a directed edge is established from the catalyst operation node T to the cloud microphysical parameter node M, and a directed edge is established from the cloud microphysical parameter node M to the ground precipitation node Y, thus forming a causal transmission path with the structure T→M→Y. S104. Combine the meteorological background node X, catalyst operation node T, cloud microphysical parameter node M, and ground precipitation node Y with directed edges to generate the structural causal model.

3. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 1, characterized in that, S2 specifically includes: S201. Determine the target operation area and historical backtracking period, extract historical observation data corresponding to the meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes, perform spatiotemporal interpolation and normalization processing on the historical observation data, and construct a tensor set containing input features. and label feature tensor set The training dataset; S202, Constructing a physical information neural network model It includes an encoder for extracting spatiotemporal features, an attention module for feature fusion, and a decoder for outputting a prediction field. The physical information neural network model has an embedded automatic differentiation module for calculating the partial derivatives of the network output variables with respect to spatiotemporal coordinates. S203. Define the total loss function of the physical information neural network model. The total loss function consists of a data fitting loss term that measures the prediction error. and physical residual constraint terms that measure physical conservation Weighted composition; S204. Input the training dataset into the physical information neural network model, and minimize the total loss function using a gradient-based optimization algorithm. Update the network weight parameters until the model converges.

4. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 3, characterized in that, In S202, the physical information neural network model The architecture employs a hybrid approach of 3D-ConvLSTM and a physical constraint layer, specifically including: The encoder employs a multi-layer 3D convolutional long short-term memory network (3D-ConvLSTM) to extract meteorological background tensors. and catalyst working tensor The spatiotemporal evolution characteristics; The fusion layer uses an attention mechanism to weightedly fuse meteorological background features and catalyst operation features to generate a latent state vector. The decoder maps the latent state vectors to the cloud microphysical parameter evolution field. and surface precipitation field The predicted output.

5. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 3, characterized in that, In S203, the total loss function The calculation formula is: ； ； ； in, is the hyperparameter for the physical constraint weights; N is the number of training samples; These are the cloud microphysical parameter predictions and surface precipitation predictions for the i-th sample output by the network, respectively. These are the historical true observations of the i-th sample; The number of physical constraint collocation points sampled in the spatiotemporal domain; To describe the partial differential equation operator for the physical conservation relationship of cloud water, it is constructed based on the cloud water resource continuity equation, and its specific physical constraint equation form is as follows: ； in, Cloud microphysical parameters output from the network The predicted value of cloud water content extracted from it; Surface precipitation field output by the network Precipitation rate forecast in the data; t is the time coordinate. V represents the local temporal variation of cloud water content; V is the wind field vector extracted from the input meteorological background data. The cloud water flux divergence term represents the cloud water transport process driven by the wind field. The source term parameters for water vapor condensation to form cloud water; These are the aggregate parameters for cloud water evaporation and dissipation.

6. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 1, characterized in that, S3 specifically includes: S301. Based on the current operation implementation parameters and upper-level wind field data, determine the operation impact area in the target operation area, and obtain the actual meteorological background data of the operation impact area during the current operation period; wherein, the operation impact area is the coverage area in the target operation area where the catalyst concentration is greater than a preset threshold; S302. Based on the intervention logic of the structural causal model, the connection between the meteorological background node and the catalyst operation node is cut off, and a counterfactual operation tensor is constructed to force the value of the catalyst operation node to be a no-operation state value. This tensor is combined with the actual meteorological background data to form a counterfactual intervention input tensor. The no-operation state value is a zero vector or null value encoding that matches the dimension of the actual meteorological background data. S303. Input the counterfactual intervention input tensor into the trained physical information neural network model to perform forward propagation operation, and use the network decoder to deduce the natural evolution process of the cloud system under the constraints of the physical equation. S304. Output the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, and the counterfactual surface precipitation field calculated based on the evolution field; wherein, the counterfactual surface precipitation field is obtained by the decoder of the physical information neural network model based on the output cloud microphysical parameter evolution field, through the cloud water-precipitation transformation mechanism learned within the network.

7. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 1, characterized in that, S4 specifically includes: S401. Obtain the ground rain gauge data and weather radar estimated precipitation data of the operation impact area during the current operation period. Through fusion calibration and spatiotemporal resampling, construct and generate the actual ground precipitation observation tensor consistent with the output grid of the physical information neural network. Specifically, using the Kalman filter method or optimal interpolation method, with the ground rain gauge observation value as the benchmark, perform deviation correction on the quantitative estimated precipitation product (QPE) of the weather radar to generate a high-precision gridded real precipitation field. Map the real precipitation field onto a preset four-dimensional grid to obtain the actual ground precipitation observation tensor. S402. Perform grid-by-grid and time-by-time difference operations on the actual surface precipitation observation tensor and the counterfactual surface precipitation field to obtain the initial net operational influence field containing positive and negative increments. S403. Using the physical residual distribution learned by the physical information neural network during the training phase, construct the upper bound of the confidence interval of the counterfactual reference field, and retain only the grid point values ​​where the actual ground precipitation observation values ​​are significantly higher than the upper bound of the confidence interval to obtain the corrected effective net rainfall field. S404. Perform spatiotemporal integration on the effective net rainfall field to calculate the absolute rainfall increase and relative rainfall rate of the weather modification operation; perform cumulative calculation on the effective net rainfall field along the time dimension to obtain a two-dimensional matrix, use color mapping technology to convert it into an isosurface layer, and overlay it with the geographic base map, catalyst operation trajectory and wind field streamline layer to generate a spatial distribution map of the rainfall enhancement effect; output the weather modification operation benefit evaluation result.

8. The method for evaluating the benefits of weather modification operations based on meteorological parameter inversion according to claim 7, characterized in that, In S403, the upper bound of the confidence interval The calculation formula is: ； in, The predicted value in the counterfactual surface precipitation field; is the standard deviation of the physical residuals during the training process of the physical information neural network; k is the confidence coefficient, with a value of 1.96, corresponding to a 95% confidence level; is the actual ground precipitation observation value at a certain time point if and only if... If the rain enhancement effect of the grid point is effective, then the initial net operational impact field value corresponding to the grid point is set to 0. In S404, the formula for calculating the absolute rainfall increase is: ； The formula for calculating the relative rainfall increase rate is: ； in, This is the corrected effective net rainfall field; Representing coordinates and time t.

9. A system for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion, comprising the method for evaluating the effectiveness of weather modification operations based on meteorological parameter inversion as described in any one of claims 1-8, characterized in that, The system includes: A construction module is used to construct a structural causal model of the weather modification operation process. The structural causal model is a directed acyclic graph, including meteorological background nodes, catalyst operation nodes, cloud microphysical parameter nodes, and surface precipitation nodes. The meteorological background nodes point to the catalyst operation nodes and surface precipitation nodes to form hybrid paths. The catalyst operation nodes point to the cloud microphysical parameter nodes, and the cloud microphysical parameter nodes point to the surface precipitation nodes. The acquisition module is used to acquire historical multi-source data of the target operation area and train a physical information neural network to establish a mapping relationship from meteorological background nodes and catalyst operation nodes to cloud microphysical parameter nodes and ground precipitation nodes; wherein, the loss function of the physical information neural network is composed of a weighted average of a data fitting loss term and a physical residual constraint term. The inversion module is used to obtain the current actual meteorological background data of the operation impact area within the target operation area, cut off the connection between the meteorological background node and the catalyst operation node in the structural causal model, set the value of the catalyst operation node to the no-operation state value, and input it together with the actual meteorological background data into the physical information neural network to invert and obtain the cloud microphysical parameter evolution field under the assumption that no catalyst operation was implemented, as well as the counterfactual surface precipitation field calculated based on the cloud microphysical parameter evolution field. The output module is used to obtain the actual ground precipitation observation values ​​of the operation impact area during the current operation period, and to perform differential calculation between the actual ground precipitation observation values ​​and the counterfactual ground precipitation field to obtain the artificial weather modification operation benefit evaluation results after eliminating the interference of natural meteorological fluctuations.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images