Forecast data processing method, system, device and medium based on phased array radar

Abstract

The present invention discloses a forecast data processing method, system, device and medium based on phased array radar, and the method includes: an evolutionary network generates a preliminary forecast sequence based on a historical observation sequence; a generator outputs a first forecast sequence based on the historical observation sequence and the preliminary forecast sequence; a true value sequence is generated according to the observation data of the forecast target period; a first loss function of the generator and a second loss function of the discriminator are calculated based on the true value sequence and the first forecast sequence; the generator and the discriminator are alternately trained according to the first loss function and the second loss function until a Nash equilibrium is reached; a real-time radar observation data sequence in a forecast area is input into the generator to obtain an original forecast result; the original forecast result is corrected by probability density matching, and a final forecast result is output. The forecast result generated according to the method of the present invention takes into account the evolution process of the meteorological system in a preset area, and has a strong timeliness and richer details.

Description

Technical Field

[0001] The present invention relates to the field of weather forecasting, and in particular to a forecast data processing method, system, device and medium based on phased array radar. Background Art

[0002] Weather radar is a microwave remote sensing device and one of the main means of space exploration in the new generation. Radar echoes can penetrate clouds, fog, rain and snow, and have the ability to work all day and all weather, playing an irreplaceable role in disastrous weather monitoring, early warning and forecasting. Rainfall is a key factor in triggering many meteorological disasters and has an important impact on human life and property. Extrapolating radar echo data is one of the main methods for short-term rainfall forecasting.

[0003] Traditional morphology-based echo extrapolation methods, such as the single centroid method, cross-correlation method and optical flow method, perform well when the echo is stable. However, when the radar echo has obvious generation and disappearance evolution characteristics, the intensity invariance assumption cannot be met and the forecast effect will be significantly reduced. The reason is that the traditional extrapolation methods do not fully explore the laws of historical observation data in terms of data utilization, and only rely on the existing data of several nearby radars for forecasting. In terms of forecast effect, since the evolution process of the occurrence, development and extinction of the meteorological system is not taken into account, the regional demand for rainfall products cannot be met. Moreover, most traditional extrapolation methods are linear and have a short timeliness. In addition, most deep learning radar extrapolation models on the market have problems with fuzzy forecast results and discontinuous changes, which are not conducive to forecasters' analysis and judgment. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides a forecast data processing method based on phased array radar, which can make full use of the laws of historical observation data and consider the evolution process of the occurrence, development, and extinction of the meteorological system, and provide a radar echo extrapolation result that is more timely and close to the real one.

[0005] In a first aspect, an embodiment of the present invention provides a forecast data processing method based on a phased array radar, comprising:

[0006] Acquire historical observation data, and generate a historical observation sequence according to the historical observation data;

[0007] Inputting the historical observation sequence into an evolutionary network, wherein the evolutionary network generates a preliminary prediction sequence according to the historical observation sequence;

[0008] Inputting the historical observation sequence and the preliminary prediction sequence into a generator so that the generator outputs a first forecast sequence;

[0009] Acquire observation data for a forecast target period, and generate a true value sequence according to the observation data for the forecast target period;

[0010] Calculating a first loss function of the generator according to the true value sequence and the first prediction sequence;

[0011] Inputting the true value sequence and the first prediction sequence into a discriminator, and calculating a second loss function of the discriminator;

[0012] According to the first loss function and the second loss function, alternately training the generator and the discriminator until a Nash equilibrium is reached;

[0013] During real-time forecasting, a real-time radar observation data sequence within the forecast area is obtained, and then input into the generator to obtain an original forecast result;

[0014] The original forecast result is corrected by probability density matching, and the final forecast result is output.

[0015] In some embodiments of the present invention, the evolutionary network is constructed based on a two-dimensional continuity equation; the step of inputting the historical observation sequence into the evolutionary network, and the evolutionary network generating a preliminary prediction sequence according to the historical observation sequence, comprises:

[0016] The evolutionary network simulates the meteorological generation law based on the historical observation sequence through a neural evolution operator to obtain an initial prediction sequence;

[0017] The evolutionary network simultaneously learns the echo motion field and echo intensity residual of the historical observation sequence, and optimizes the initial prediction sequence through gradient back propagation to obtain the preliminary prediction sequence.

[0018] In some embodiments of the present invention, the step of inputting the historical observation sequence and the preliminary prediction sequence into a generator so that the generator outputs a first forecast sequence includes:

[0019] Fix the parameters of the discriminator, and input the preliminary prediction sequence and the historical observation sequence into the generator;

[0020] The generator extracts echo features of the preliminary prediction sequence through a multi-layer network structure and nonlinear transformation, and obtains an evolution pattern from the historical observation sequence to the preliminary prediction sequence;

[0021] The first forecast sequence is generated according to the evolution pattern and the historical observation sequence.

[0022] In some embodiments of the present invention, the step of calculating the first loss function of the generator according to the true value sequence and the first prediction sequence includes:

[0023] Forward propagation calculates the weighted reconstruction loss and the perceptual similarity loss of the generator, and simultaneously calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains the first loss function according to the weighted reconstruction loss, the perceptual similarity loss, the sequence discrimination loss, and the frame discrimination loss;

[0024] The first loss function is back-propagated to the generator, and the generator adjusts its own parameters according to the first loss function.

[0025] In some embodiments of the present invention, the step of inputting the true value sequence and the first prediction sequence into a discriminator and calculating the second loss function of the discriminator includes:

[0026] Fixing the parameters of the generator, and randomly inputting the first prediction sequence and the true value sequence into the discriminator;

[0027] The discriminator discriminates the authenticity of the first prediction sequence and the true value sequence;

[0028] Forward propagation calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains the second loss function according to the sequence discrimination loss and the frame discrimination loss;

[0029] The second loss function is back-propagated to the discriminator, and the discriminator adjusts its own parameters according to the second loss function.

[0030] In some embodiments of the present invention, the step of alternately training the generator and the discriminator according to the first loss function and the second loss function until a Nash equilibrium is reached further includes:

[0031] The generator and the discriminator are trained alternately and cyclically until the distribution of the generator output fits the distribution of the true value sequence and the discriminator cannot distinguish the authenticity of the generator output.

[0032] In some embodiments of the present invention, the step of performing probability density matching correction on the original forecast result and outputting the final forecast result includes:

[0033] Inputting the real-time radar observation data sequence into the generator to obtain a second forecast sequence;

[0034] Acquire a preset number of forecast frames of the second forecast sequence, and calculate a first probability density distribution of the preset number of forecast frames;

[0035] Acquire all the forecast frames of the second forecast sequence, and calculate a second probability density distribution of all the forecast frames;

[0036] The first probability density distribution and the second probability density distribution are aligned, and the strengths of all the forecast frames are corrected to obtain the final forecast result.

[0037] In a second aspect, an embodiment of the present invention provides a forecast data processing system based on a phased array radar, comprising:

[0038] A historical data processing module, used to obtain historical observation data and generate a historical observation sequence according to the historical observation data;

[0039] An evolutionary network module, used for inputting the historical observation sequence into the evolutionary network, and for generating a preliminary prediction sequence according to the historical observation sequence;

[0040] A true value sequence generation module, used to obtain observation data of a forecast target period, and generate a true value sequence according to the observation data of the forecast target period;

[0041] A training module, used for inputting the historical observation sequence and the preliminary prediction sequence into a generator, so that the generator outputs a first forecast sequence; the training module is also used for calculating a first loss function of the generator according to the true value sequence and the first forecast sequence; the training module is also used for inputting the true value sequence and the first forecast sequence into a discriminator, and calculating a second loss function of the discriminator; the training module is also used for alternately training the generator and the discriminator according to the first loss function and the second loss function until a Nash equilibrium is reached;

[0042] The forecast module is used to first obtain the real-time radar observation data sequence in the forecast area during real-time forecasting, and then input it into the generator to obtain the original forecast result; and to perform probability matching correction on the original forecast result and output the final forecast result.

[0043] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the at least one program to execute the forecast data processing method based on phased array radar described in the embodiment of the above aspect.

[0044] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to execute the forecast data processing method based on phased array radar as described in the above-mentioned embodiment.

[0045] The forecast data processing method based on phased array radar according to the present invention has at least the following beneficial effects: the present invention constructs a high-resolution extrapolation model based on a generative adversarial network by processing the radar echo data obtained in historical observations through an evolutionary network, and inputs the processed historical data and the original historical data into a generative adversarial network model with a generator and a discriminator, thereby effectively suppressing the fuzzification problem and associating the prediction sequence output by the generator with the historical data; the forecast results output by the generative adversarial network model take into account the evolutionary processes such as the occurrence, development, and extinction of the meteorological system in a preset area, and the forecast results are nonlinear and highly timely, meeting the needs for regional meteorological forecasts; the model forecast results are rich in details and better meet the needs of refined forecasts. In addition, the probability matching method is used to post-process the forecast results, which can suppress the phenomenon of discontinuity in the forecast results. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0047] Figure 1 It is a flow chart of a forecast data processing method based on phased array radar according to an embodiment of the present invention;

[0048] Figure 2 It is a schematic block diagram of a forecast data processing method based on a phased array radar according to an embodiment of the present invention;

[0049] Figure 3 Schematic diagram of the U-Net model. DETAILED DESCRIPTION

[0050] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0051] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0052] In the description of the present invention, "several" means one or more, "more" means two or more, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of the first, second or third, it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0053] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0054] In the description of this specification, the description with reference to the terms "one embodiment", "further embodiments", "some specific embodiments" or "some examples" etc. means that the specific features, structures or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0055] The present invention processes the radar echo data obtained from historical observations through an evolutionary network, and inputs the processed historical data and the original historical data into a generative adversarial network model with a generator and a discriminator, thereby constructing a high-resolution extrapolation model based on a generative adversarial network, which effectively suppresses the fuzzification problem and associates the prediction sequence output by the generator with the historical data; the forecast results output by the generative adversarial network model take into account the evolutionary processes such as the occurrence, development, and extinction of the meteorological system in the preset area, and the forecast results are nonlinear and have strong timeliness, which meets the needs of regional meteorological forecasting; the model forecast results are rich in details and are more in line with the needs of refined forecasting. In addition, the probability matching method is used to post-process the forecast results to suppress the phenomenon of discontinuity of the forecast results.

[0056] The concepts involved in the present invention are explained below:

[0057] GAN: Generative adversarial network, generative adversarial network;

[0058] U-Net: A convolutional neural network model consisting of an encoder and a decoder. The encoder gradually abstracts the input information into high-level feature representations, and the decoder gradually decodes the high-level feature representations into specific outputs.

[0059] The specific method of the embodiment of the present invention is described below with reference to the accompanying drawings.

[0060] Reference Figure 1~Figure 2 , Figure 1 This is a flow chart of a forecast data processing method based on a phased array radar according to an embodiment of the present invention. The method specifically includes:

[0061] Step S100, acquiring historical observation data, and generating a historical observation sequence according to the historical observation data;

[0062] Step S200, inputting the historical observation sequence into the evolutionary network, and the evolutionary network will generate a preliminary prediction sequence based on the historical observation sequence;

[0063] Step S300, inputting the historical observation sequence and the preliminary prediction sequence into the generator, so that the generator outputs the first forecast sequence;

[0064] Step S400, obtaining observation data of a forecast target period, and generating a true value sequence according to the observation data of the forecast target period;

[0065] Step S500, calculating a first loss function of a generator according to the true value sequence and the first prediction sequence;

[0066] Step S600, inputting the true value sequence and the first prediction sequence into the discriminator, and calculating the second loss function of the discriminator;

[0067] Step S700, according to the first loss function and the second loss function, alternately training the generator and the discriminator until a Nash equilibrium is reached;

[0068] Step S800, when making a real-time forecast, a real-time radar observation data sequence within the forecast area is obtained, and then input into a generator to obtain an original forecast result;

[0069] Step S900, performing probability density matching correction on the original forecast result, and outputting the final forecast result.

[0070] It should be noted that Figure 2The schematic block diagram of the forecast data processing method based on phased array radar in the embodiment of the present invention, in step S100, we usually regard the radar image of the study area at a certain moment as a frame of two-dimensional image, then the whole rainfall process is composed of a sequence of two-dimensional radar rainfall images, there is a strong spatiotemporal correlation between the adjacent frames of these radar images, and the spatiotemporal sequence prediction method can be used to process the historical observation data, and the historical observation sequence is generated according to the historical observation data, that is, the continuous echo data is converted into a sequence data composed of each frame of image; the historical observation data is the radar echo data obtained by the meteorological radar continuously obtained in any area at any time. In step S200, the evolutionary network used in the embodiment of the present invention can be based on the differentiable neural evolution solution operator, it simultaneously learns the echo motion field and the echo intensity residual through the neural network, in addition, it directly optimizes the prediction error in the entire time range through gradient back propagation, which can effectively prevent error accumulation. In step S400, the radar observation data of the forecast target period is first obtained, and the forecast target period refers to the adjacent period after the start time, generally 2 to 3 hours, and then the true value sequence is generated according to the observation data of the forecast target period. In steps S300 and S600, the generator generates a first forecast sequence by learning the evolution pattern of radar meteorological data in the historical observation sequence and the preliminary prediction sequence. At this time, the first forecast sequence generated has many random factors, and it will be inaccurate to use the first forecast sequence directly as the prediction result. Therefore, it is necessary to input the image of the first forecast sequence and the image of the true value sequence into the discriminator so that the discriminator can identify the two image sequences, find the image of the true value sequence, and feed it back to the generator as the basis for the generator to adjust the parameters. Repeat this process so that the first forecast sequence generated by the generator has more features of the true sequence until the discriminator cannot distinguish between the image of the first forecast sequence and the image of the true value sequence; in steps S500 to S700, the generator generates The process of generating the first forecast sequence will also calculate the first loss function according to the forward propagation. The first loss function represents the loss of the generator in the process of generating the first forecast sequence. The generator adjusts its own parameters according to the feedback of the first loss function, thereby generating a first forecast sequence that is more consistent with the real sequence; the discriminator will also calculate the second loss function according to the forward propagation in the process of distinguishing the first forecast sequence and the true value sequence. The second loss function represents the loss of the discriminator in the discrimination process. The discriminator adjusts its own parameters according to the feedback of the second loss function, thereby improving its own discrimination ability. When the discriminator cannot distinguish the first forecast sequence and the true value sequence, it can be considered that the prediction sequence generated by the generator is consistent with the real situation. At this time, the generator can be used to generate the most preferred prediction image. In step S800, the real-time radar sequence of the area to be predicted can be input into the generator to obtain the original prediction result. In step S900, the original forecast result is corrected by probability density matching, and the final forecast result is output.It is understandable that the post-processing of forecast results using the probability matching method can suppress the phenomenon of discontinuous results.

[0071] It should be noted that in steps S100 to S900, the embodiment of the present invention fully utilizes the rules of historical observation data by adopting a generative adversarial network, and with the adversarial training of the generator and the discriminator and continuous adjustment of their own parameters, the obtained prediction data is close to the truth, thereby improving the accuracy of the prediction data and meeting the needs of meteorological forecasting.

[0072] In some embodiments of the present invention, an evolutionary network is constructed based on a two-dimensional continuity equation; the steps of inputting a historical observation sequence into the evolutionary network, and the evolutionary network generating a preliminary prediction sequence based on the historical observation sequence include:

[0073] Step S210, the evolutionary network simulates the meteorological generation law through the neural evolution operator based on the historical observation sequence to obtain the initial prediction sequence;

[0074] In step S220, the evolutionary network simultaneously learns the echo motion field and echo intensity residual of the historical observation sequence, and optimizes the initial prediction sequence through gradient back propagation to obtain a preliminary prediction sequence.

[0075] It should be noted that in step S210 to step S220, since the common radar echo extrapolation method based on deep learning cannot comply with the inherent physical laws of the meteorological system, such as the law of conservation of advection in the rainfall system, the present application adopts an evolutionary network based on a new differentiable neural evolutionary solution operator, which simultaneously learns the echo motion field and the echo intensity residual through a neural network. In addition, it directly optimizes the prediction error within the entire time range through gradient back propagation, which can effectively prevent error accumulation. The present application adopts a pre-trained evolutionary network, uses the output of the evolutionary network as the input of the generator, and does not adjust the parameters of the pre-trained evolutionary network.

[0076] Reference Figure 1 , Figure 3 In some embodiments of the present invention, the step of inputting the historical observation sequence and the preliminary prediction sequence into the generator so that the generator outputs the first forecast sequence includes:

[0077] Step S310, fixing the parameters of the discriminator, and inputting the preliminary prediction sequence and the historical observation sequence into the generator;

[0078] Step S320, the generator extracts echo features of the preliminary prediction sequence through a multi-layer network structure and nonlinear transformation, and obtains an evolution pattern from the historical observation sequence to the preliminary prediction sequence;

[0079] Step S330, generating a first forecast sequence according to the evolution pattern and the historical observation sequence.

[0080] It should be noted that in steps S310 to S330, the present invention uses a U-Net network model as the generator of the generative adversarial network. Figure 3 As shown in , it is the framework diagram of the U-Net convolutional neural network model, which adopts a network structure including downsampling and upsampling. Downsampling is used to gradually reveal environmental information, and the upsampling process combines the information of each layer of downsampling and the input information of upsampling to restore detailed information and gradually restore the image accuracy. This model is a lightweight but very efficient model that can extract the evolution pattern of radar echoes. Further, the multi-layer network structure of the generator includes an upsampling layer and a downsampling layer; the downsampling layer obtains the echo features of the preliminary prediction sequence and the historical observation sequence through convolutional calculation by the Conv2d operator; the upsampling layer obtains the evolution pattern of the radar echoes in the preliminary prediction sequence and the historical observation sequence through transposed convolutional calculation by the ConvTranspose2d operator. In the UNet network, the downsampling and upsampling operations are respectively implemented by convolution and transposed convolution. The downsampling is implemented by the Conv2d operator, and the input is X, with the dimension (Nin, Cin, Hin, Win), where Nin represents the batch size, which indicates the number of samples input into the network at one time; Cin represents the number of input channels, which indicates the number of channels of the input data; Hin represents the input height, which indicates the height of the input data, that is, the number of rows of the image; Win represents the input width, which indicates the width of the input data, that is, the number of columns of the image. The output is V, with the dimension (Nout, Cout, Hout, Wout), and the meanings of each dimension are the same as those of the input data. The subscript out represents the output. Generally, Hout < Hin and Wout < Win. Such settings are used for the length and width of the input feature map. The specific formula is as follows:

[0081] , ①

[0082] Equation ①, where W is the parameter of the convolution operation, including the size of the convolution kernel, the size of the stride, and the size of the padding. The upsampling is implemented by the ConvTranspose2d operator, which is the inverse operation of the above convolution. Such settings are used to increase the length and width of the input feature map.

[0083] Referring to Figure 1 , in some embodiments of the present invention, the steps of calculating the first loss function of the generator according to the true value sequence and the first prediction sequence include:

[0084] Step S510, forward propagation calculates the weighted reconstruction loss and the perceptual similarity loss of the generator, and simultaneously calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains a first loss function according to the weighted reconstruction loss, the perceptual similarity loss, the sequence discrimination loss, and the frame discrimination loss;

[0085] Step S520, back-propagating the first loss function to the generator, and the generator adjusts its own parameters according to the first loss function.

[0086] In step S510 to step S520, when training the generator, the embodiment of the present invention forward propagates and calculates the weighted reconstruction loss and the perceptual similarity loss, and simultaneously calculates the sequence discrimination loss and the frame discrimination loss of the discriminator. The generator parameters are optimized through back propagation of these four losses, thereby adjusting the image features of the first prediction sequence output by the generator next time.

[0087] Reference Figure 1 In some embodiments of the present invention, the step of inputting the true value sequence and the first prediction sequence into the discriminator and calculating the second loss function of the discriminator includes:

[0088] Step S610, fixing the parameters of the generator, and randomly inputting the first prediction sequence and the true value sequence into the discriminator;

[0089] Step S620, the discriminator discriminates the authenticity of the first prediction sequence and the true value sequence;

[0090] Step S630, forward propagation calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains a second loss function according to the sequence discrimination loss and the frame discrimination loss;

[0091] Step S640, back-propagating the second loss function to the discriminator, and the discriminator adjusts its own parameters according to the second loss function.

[0092] It should be noted that in step S610 to step S640, the embodiment of the present invention performs authenticity discrimination on the first prediction sequence and the true value sequence when training the discriminator, and calculates the discrimination loss of the discriminator based on the performance of the discriminator in the authenticity discrimination process, that is, forward propagation calculates the sequence discrimination loss and the frame discrimination loss, and optimizes the discriminator parameters through back propagation of these two losses, thereby improving the discrimination ability of the discriminator at the next discrimination.

[0093] Reference Figure 1 In some embodiments of the present invention, the step of alternately training the generator and the discriminator according to the first loss function and the second loss function until the Nash equilibrium is reached further includes:

[0094] Step S710, alternately and cyclically train the generator and the discriminator until the distribution of the generator output perfectly fits the distribution of the true value sequence, and the discriminator cannot determine the authenticity of the generator output.

[0095] It should be noted that the generator and the discriminator are trained alternately. When the generator and the discriminator are trained alternately to Nash equilibrium, that is, the parameters of the generator and the discriminator are adjusted to the optimal generation parameters and discrimination parameters, the discriminator cannot distinguish the difference between the first prediction sequence generated by the generator and the true sequence under the optimal discrimination parameters, and presents a straight line with a value of about 0.5 (between true and false). At this time, the back propagation training process ends.

[0096] In some embodiments of the present invention, the step of performing probability density matching correction on the original forecast result and outputting the final forecast result includes:

[0097] Step S810, inputting the real-time radar observation data sequence into the generator to obtain a second forecast sequence;

[0098] Step S820, obtaining a preset number of forecast frames of the second forecast sequence, and calculating a first probability density distribution of the preset number of forecast frames;

[0099] Step S830, obtaining all forecast frames of the second forecast sequence, and calculating the second probability density distribution of all forecast frames;

[0100] Step S840, aligning the first probability density distribution and the second probability density distribution, correcting the strength of all forecast frames, and obtaining a forecast result.

[0101] It should be noted that general GAN ​​often leads to continuity problems between forecast frames (frame skipping phenomenon) and some obviously unreasonable generation and disappearance evolution. Therefore, steps S810 to S840 post-process the forecast results through a probability matching method to suppress the frame skipping phenomenon, ensure the temporal continuity of the forecast results, and improve the model forecasting effect.

[0102] In a second aspect, an embodiment of the present invention provides a forecast data processing system based on a phased array radar, comprising:

[0103] A historical data processing module is used to obtain historical observation data and generate a historical observation sequence based on the historical observation data;

[0104] An evolutionary network module, used for inputting the historical observation sequence into the evolutionary network, and for generating a preliminary prediction sequence according to the historical observation sequence;

[0105] A true value sequence generation module is used to obtain the observation data of the forecast target period and generate a true value sequence according to the observation data of the forecast target period;

[0106] The training module is used to input the historical observation sequence and the preliminary prediction sequence into the generator, so that the generator outputs the first prediction sequence; the training module is also used to calculate the first loss function of the generator according to the true value sequence and the first prediction sequence; the training module is also used to input the true value sequence and the first prediction sequence into the discriminator, and calculate the second loss function of the discriminator; the training module is also used to alternately train the generator and the discriminator according to the first loss function and the second loss function until a Nash equilibrium is reached;

[0107] The forecast module is used to first obtain the real-time radar observation data sequence in the forecast area during real-time forecasting, and then input it into the generator to obtain the original forecast result; and to perform probability matching correction on the original forecast result and output the final forecast result.

[0108] It should be noted that the forecast data processing system based on phased array radar provided by the embodiment of the present invention includes a high-resolution extrapolation model based on a generative adversarial network, and uses this model to effectively suppress the fuzzification problem; at the same time, as the forecast time is extended, the proportion of high-intensity echoes will not decrease significantly. The model forecast results are rich in details and better meet the needs of refined forecasts. In addition, the embodiment of the present invention uses a probability matching method to post-process the forecast results, suppressing the frame skipping phenomenon and ensuring the continuity of the forecast results.

[0109] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load at least one program to execute the forecast data processing method based on phased array radar of the embodiment of the above aspect.

[0110] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to execute the forecast data processing method based on phased array radar as described in the above-mentioned embodiment.

[0111] It should be appreciated that embodiments of the present invention may be implemented or enforced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, if desired, the program may be implemented in an assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed ASIC for this purpose.

[0112] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that may be executed by one or more processors.

[0113] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, a RAM, a ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or portions thereof, can be transmitted via a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.

[0114] The computer program can be applied to input data to perform the functions described herein, thereby converting the input data to generate output data stored in a non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the present invention, the converted data represents physical and tangible objects, including specific visual depictions of physical and tangible objects produced on the display.

[0115] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A forecast data processing method based on phased array radar, characterized in that: include: Acquire historical observation data, and generate a historical observation sequence according to the historical observation data; Inputting the historical observation sequence into an evolutionary network, wherein the evolutionary network generates a preliminary prediction sequence according to the historical observation sequence; Inputting the historical observation sequence and the preliminary prediction sequence into a generator so that the generator outputs a first forecast sequence; Acquire observation data for a forecast target period, and generate a true value sequence according to the observation data for the forecast target period; Calculating a first loss function of the generator according to the true value sequence and the first prediction sequence; Inputting the true value sequence and the first prediction sequence into a discriminator, and calculating a second loss function of the discriminator; According to the first loss function and the second loss function, alternately training the generator and the discriminator until a Nash equilibrium is reached; During real-time forecasting, a real-time radar observation data sequence within the forecast area is obtained, and then input into the generator to obtain an original forecast result; Performing probability density matching correction on the original forecast result and outputting the final forecast result; The evolutionary network is constructed based on a two-dimensional continuity equation; The step of inputting the historical observation sequence into the evolutionary network, wherein the evolutionary network generates a preliminary prediction sequence according to the historical observation sequence, comprises: The evolutionary network simulates the meteorological generation law based on the historical observation sequence through a neural evolution operator to obtain an initial prediction sequence; The evolutionary network simultaneously learns the echo motion field and echo intensity residual of the historical observation sequence, and optimizes the initial prediction sequence by gradient back propagation to obtain the preliminary prediction sequence; The step of inputting the true value sequence and the first prediction sequence into a discriminator and calculating a second loss function of the discriminator comprises: Fixing the parameters of the generator, and randomly inputting the first prediction sequence and the true value sequence into the discriminator; The discriminator discriminates the authenticity of the first prediction sequence and the true value sequence; Forward propagation calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains the second loss function according to the sequence discrimination loss and the frame discrimination loss; The second loss function is back-propagated to the discriminator, and the discriminator adjusts its own parameters according to the second loss function.

2. The forecast data processing method based on phased array radar according to claim 1, characterized in that: The step of inputting the historical observation sequence and the preliminary prediction sequence into a generator so that the generator outputs a first forecast sequence comprises: Fix the parameters of the discriminator, and input the preliminary prediction sequence and the historical observation sequence into the generator; The generator extracts echo features of the preliminary prediction sequence through a multi-layer network structure and nonlinear transformation, and obtains an evolution pattern from the historical observation sequence to the preliminary prediction sequence; The first forecast sequence is generated according to the evolution pattern and the historical observation sequence.

3. The forecast data processing method based on phased array radar according to claim 1, characterized in that: The step of calculating the first loss function of the generator according to the true value sequence and the first prediction sequence comprises: Forward propagation calculates the weighted reconstruction loss and the perceptual similarity loss of the generator, and simultaneously calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains the first loss function according to the weighted reconstruction loss, the perceptual similarity loss, the sequence discrimination loss, and the frame discrimination loss; The first loss function is back-propagated to the generator, and the generator adjusts its own parameters according to the first loss function.

4. The forecast data processing method based on phased array radar according to claim 1, characterized in that: The step of alternately training the generator and the discriminator according to the first loss function and the second loss function until a Nash equilibrium is reached further includes: The generator and the discriminator are trained alternately and cyclically until the distribution of the generator output fits the distribution of the true value sequence and the discriminator cannot distinguish the authenticity of the generator output.

5. The forecast data processing method based on phased array radar according to claim 1, characterized in that: The step of performing probability density matching correction on the original forecast result and outputting the final forecast result comprises: Inputting the real-time radar observation data sequence into the generator to obtain a second forecast sequence; Acquire a preset number of forecast frames of the second forecast sequence, and calculate a first probability density distribution of the preset number of forecast frames; Acquire all the forecast frames of the second forecast sequence, and calculate a second probability density distribution of all the forecast frames; The first probability density distribution and the second probability density distribution are aligned, and the strengths of all the forecast frames are corrected to obtain the final forecast result.

6. A forecast data processing system based on phased array radar, characterized in that: include: A historical data processing module, used to obtain historical observation data and generate a historical observation sequence according to the historical observation data; An evolutionary network module, used for inputting the historical observation sequence into the evolutionary network, and for generating a preliminary prediction sequence according to the historical observation sequence; A true value sequence generation module, used to obtain observation data of a forecast target period, and generate a true value sequence according to the observation data of the forecast target period; A training module, used for inputting the historical observation sequence and the preliminary prediction sequence into a generator, so that the generator outputs a first forecast sequence; the training module is also used for calculating a first loss function of the generator according to the true value sequence and the first forecast sequence; the training module is also used for inputting the true value sequence and the first forecast sequence into a discriminator, and calculating a second loss function of the discriminator; the training module is also used for alternately training the generator and the discriminator according to the first loss function and the second loss function until a Nash equilibrium is reached; A forecast module is used to first obtain a real-time radar observation data sequence within the forecast area during real-time forecasting, and then input the sequence into the generator to obtain an original forecast result; and for performing probability matching correction on the original forecast result and outputting the final forecast result; The evolutionary network is constructed based on a two-dimensional continuity equation; The step of inputting the historical observation sequence into the evolutionary network, wherein the evolutionary network generates a preliminary prediction sequence according to the historical observation sequence, comprises: The evolutionary network simulates the meteorological generation law based on the historical observation sequence through a neural evolution operator to obtain an initial prediction sequence; The evolutionary network simultaneously learns the echo motion field and echo intensity residual of the historical observation sequence, and optimizes the initial prediction sequence by gradient back propagation to obtain the preliminary prediction sequence; The step of inputting the true value sequence and the first prediction sequence into a discriminator and calculating a second loss function of the discriminator comprises: Fixing the parameters of the generator, and randomly inputting the first prediction sequence and the true value sequence into the discriminator; The discriminator discriminates the authenticity of the first prediction sequence and the true value sequence; Forward propagation calculates the sequence discrimination loss and the frame discrimination loss of the discriminator, and obtains the second loss function according to the sequence discrimination loss and the frame discrimination loss; The second loss function is back-propagated to the discriminator, and the discriminator adjusts its own parameters according to the second loss function.

7. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the at least one program to execute the forecast data processing method based on phased array radar according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to execute the forecast data processing method based on phased array radar as described in any one of claims 1 to 5 when executed by the processor.

Images