TWI798583B - Electronic device and method for accelerating generation of simulation result of simulation software - Google Patents

Abstract

An electronic device and a method for accelerating a generation of a simulation result of a simulation software are provided. The method includes: obtaining input data; generating fake data according to the input data; inputting a first fake feature of the fake data to the simulation software to generate a first fake simulation result; obtaining a user-defined feature; calculating a first similarity between the first fake feature and the user-defined feature; generating, based on the first similarity, an approximate simulation result matching with the user-defined feature according to the first fake simulation result; and outputting the approximate simulation result.

Description

Electronic device and method for accelerating generation of simulation results of simulation software

The present invention relates to an electronic device and method for accelerating the generation of simulation results of simulation software.

Numerical simulation is a combination of theoretical calculations and computer programs, and then the computer performs large and complex calculations. In the process of simulation calculation, it is necessary to use the correct approximation method, so that the obtained simulation results can conform to the real situation. Therefore, the generation of simulation results must rely on hardware devices with high computing power. As the complexity of the simulation model increases, the time to generate simulation results also increases. Therefore, simulation software often cannot provide simulation results to users in real time. Accordingly, how to speed up the generation of simulation results without increasing the cost of hardware construction is one of the goals that people in the field are committed to.

The invention provides an electronic device and method for accelerating the generation of simulation results of simulation software, which can reduce the generation time of simulation results.

An electronic device for accelerating the generation of simulation results of simulation software in the present invention includes a processor, a storage medium, and a transceiver. The transceiver obtains input data. The storage medium stores multiple modules. The processor is coupled to the storage medium and the transceiver, and accesses and executes multiple modules, wherein the multiple modules include machine learning models and computing modules. Machine learning models generate fake data based on input data. The calculation module inputs the first imitation feature of the imitation data to the simulation software through the transceiver to generate the first imitation simulation result, receives the user-defined feature through the transceiver, and calculates the first imitation feature between the first imitation feature and the user-defined feature. Based on the first degree of similarity, an approximate simulation result matching the user-defined feature is generated according to the first imitation simulation result, and the approximate simulation result is output through the transceiver.

In an embodiment of the present invention, the above-mentioned computing module generates an approximate simulation result according to the first simulated feature, the first simulated simulation result and the user-defined feature based on one of interpolation and extrapolation.

In an embodiment of the present invention, the above-mentioned computing module inputs the user-defined features into the simulation software through the transceiver to generate a final simulation result, and outputs the final simulation result through the transceiver.

In an embodiment of the present invention, the above-mentioned modules further include a database of imitation features and a database of simulation results. The counterfeit feature database stores a plurality of counterfeit features including the first counterfeit feature. The simulation result database stores a plurality of imitation simulation results respectively corresponding to a plurality of imitation features, wherein the computing module calculates the a degree of similarity between each of the features and a user-defined feature, and in response to the first degree of similarity being the highest degree of similarity, selecting a first simulated simulation result corresponding to the first degree of similarity from the plurality of simulated simulation results to generate an approximate simulation result.

In an embodiment of the present invention, the above-mentioned modules further include a history database. The historical database stores historical data, wherein the machine learning model includes a generator and a discriminator, wherein the discriminator is trained according to the historical data, wherein the generator generates candidate counterfeit data according to the input data, and the discriminator determines the candidate counterfeit data to be counterfeit data.

In an embodiment of the present invention, the aforementioned machine learning model is a generative adversarial network.

A method for accelerating the generation of simulation results of simulation software according to the present invention includes: obtaining input data; generating counterfeit data according to the input data; inputting the first counterfeit feature of the counterfeit data into the simulation software to generate the first counterfeit simulation result; obtaining and using calculating a first similarity between the first simulated feature and the user-defined feature; generating an approximate simulation result matching the user-defined feature from the first simulated simulation result based on the first similarity; and outputting the approximate simulation result.

Based on the above, the present invention can generate approximate simulation results in real time for the user's reference before the simulation software takes a long time to generate the simulation results.

100: Electronic device

110: Processor

120: storage media

121: Operation module

122:Machine Learning Models

1221: generator

1222: discriminator

123: Historical database

124:Imitation feature database

125:Simulation result database

126: User-defined feature database

130: Transceiver

S210, S220, S221, S222, S223, S230, S240, S410, S420, S430, S440, S510, S520, S530, S540, S550, S560, S570: steps

Fig. 1 shows a simulation result of an acceleration simulation software according to an embodiment of the present invention Schematic of the resulting electronics.

FIG. 2 illustrates a flow chart of generating simulation simulation results according to an embodiment of the present invention.

FIG. 3 shows a detailed flow chart of the steps of generating counterfeit data according to the input data according to an embodiment of the present invention.

FIG. 4 illustrates a flowchart for generating approximate simulation results and final simulation results according to an embodiment of the present invention.

FIG. 5 is a flowchart of a method for accelerating the generation of simulation results of simulation software according to an embodiment of the present invention.

FIG. 1 is a schematic diagram of an electronic device 100 for accelerating the generation of simulation results of simulation software according to an embodiment of the present invention. The electronic device 100 may include a processor 110 , a storage medium 120 and a transceiver 130 .

The processor 110 is, for example, a central processing unit (central processing unit, CPU), or other programmable general purpose or special purpose micro control unit (micro control unit, MCU), microprocessor (microprocessor), digital signal processing Digital Signal Processor (DSP), Programmable Controller, Application Specific Integrated Circuit (ASIC), Graphics Processing Unit (GPU), Image Signal Processor (ISP) ), image processing unit (image processing unit, IPU), arithmetic logic unit (arithmetic logic unit, ALU), complex programmable logic device (complex programmable logic device, CPLD), field programmable logic Gate array (field programmable gate array, FPGA) or other similar components or a combination of the above components. The processor 110 can be coupled to the storage medium 120 and the transceiver 130 , and access and execute multiple modules and various application programs stored in the storage medium 120 .

The storage medium 120 is, for example, any type of fixed or removable random access memory (random access memory, RAM), read-only memory (read-only memory, ROM), flash memory (flash memory) , hard disk drive (hard disk drive, HDD), solid state drive (solid state drive, SSD) or similar components or a combination of the above components, and are used to store multiple modules or various application programs that can be executed by the processor 110 . In this embodiment, the storage medium 120 can store multiple models including the computing module 121, the machine learning model 122, the historical database 123, the imitation feature database 124, the simulation result database 125, and the user-defined feature database 126. group, its function will be explained later.

The transceiver 130 transmits and receives signals in a wireless or wired manner. The transceiver 130 may also perform operations such as low noise amplification, impedance matching, frequency mixing, up or down frequency conversion, filtering, amplification, and the like.

The electronic device 100 can be used to generate fake simulation results. FIG. 2 illustrates a flow chart of generating simulation simulation results according to an embodiment of the present invention. In step S210 , the computing module 121 can obtain input data through the transceiver 130 . The input data may be image data or data curves used to generate simulation results. For example, the input data may include weather-related images or curves such as satellite cloud images, radar echo images, temperature curves, humidity curves, rainfall accumulation curves, or air quality curves, but the invention is not limited thereto.

In step S220, the machine learning model 122 can generate imitation data according to the input data. The counterfeit data may be counterfeit graphics or data curves. For example, the fake data may include fake satellite cloud images, fake radar echo charts, fake temperature curves, fake humidity curves, fake rainfall accumulation curves or fake air quality curves, etc., but the present invention does not limited to this. The machine learning model 122 is, for example, a generative adversarial network (GAN). The machine learning model 122 may include a generator (generator) 1221 and a discriminator (discriminator) 1222, wherein the discriminator 1222 is trained by the machine learning model 122 according to the historical data of the historical database 123, wherein the historical data includes historical satellite cloud images, Images such as historical radar echo graphs, historical temperature curves, historical humidity curves, historical rainfall accumulation curves, or historical air quality curves.

FIG. 3 shows a detailed flow chart of the steps of generating counterfeit data according to the input data according to an embodiment of the present invention. In step S221, the generator 1221 can generate candidate counterfeit data according to the input data. In step S222, the discriminator 1222 can compare whether the candidate counterfeit data matches the historical data. If the candidate counterfeit data does not match the historical data, the discriminator 1222 may delete the candidate counterfeit data, and step S221 may be re-executed to generate new candidate counterfeit data. If the candidate counterfeit data matches the historical data, go to step S223. In step S223 , the discriminator 1222 may determine that the candidate forgery data is a forgery data, and store the generated forgery data in the forgery feature database 124 . Next, step S221 can be re-executed to generate new candidate counterfeit data. Step S221 to step S223 in FIG. 3 can be repeatedly executed until the counterfeit feature database 124 obtains enough counterfeit information.

For example, suppose the input data is a satellite cloud image. The machine learning model 122 can generate a plurality of fake satellite cloud images according to historical data (for example: historical satellite cloud images) and satellite cloud images.

Back to Figure 2. In step S230 , the computing module 121 may perform feature extraction on a plurality of counterfeit data in the counterfeit feature database 124 to obtain a plurality of counterfeit features respectively corresponding to the plurality of counterfeit data. The computing module 121 can store a plurality of imitation features in the imitation feature database 124 .

Specifically, the computing module 121 may perform feature extraction on each of the plurality of simulated satellite cloud images to obtain simulated regional rainfall information (ie, simulated features). The computing module 121 can store the simulated area rainfall information in the simulated feature database 124 . For example, the simulated feature database 124 may store simulated area rainfall information corresponding to 400 mm, simulated area rainfall information corresponding to 500 mm, and simulated area rainfall information corresponding to 600 mm.

In step S240, the computing module 121 may input the plurality of imitation features into the simulation software to generate a plurality of imitation simulation results respectively corresponding to the plurality of imitation features. The calculation module 121 can store a plurality of imitation simulation results in the simulation result database 125 .

For example, if the simulation software is used to determine whether a flood has occurred, the computing module 121 can respectively input a plurality of simulated regional rainfall information (ie, simulated features) into the simulated software. The simulation software can generate a plurality of flood judgment results respectively corresponding to a plurality of simulated regional rainfall information (ie, simulated simulation results). The imitation feature database 124 can store a plurality of regional rainfall information and corresponding to the plurality of regional rainfall information Multiple flood judgment results. For example, the imitation feature database 124 can store the simulated area rainfall information corresponding to 400 mm and the corresponding flood judgment result, the simulated area rainfall information corresponding to 500 mm and the corresponding flood judgment result, and the simulated area corresponding to 600 mm. Imitation of regional rainfall information and corresponding flood judgment results.

After obtaining a plurality of counterfeit simulation results, the electronic device 100 can generate an approximate simulation result corresponding to the user's demand according to the plurality of counterfeit simulation results. Approximate simulation results are produced in a fraction of the time. In addition, the electronic device 100 can also use simulation software to generate a final simulation result that takes a long time but is relatively accurate. FIG. 4 illustrates a flowchart for generating approximate simulation results and final simulation results according to an embodiment of the present invention.

In step S410 , the computing module 121 may receive user-defined features through the transceiver 130 . User-defined features can be determined by users according to their needs. For example, if the user is interested in the flood judgment result corresponding to the regional rainfall information of 540 mm, the user may determine that the user-defined feature corresponds to the regional rainfall information of 540 mm. The user can input the user-defined features into the electronic device 100 through an input device such as a keyboard or a touch screen. The computing module 121 can obtain user-defined features from the input device through the transceiver 130 . The computing module 121 can store the user-defined features in the user-defined feature database 126 .

In step S420, the calculation module 121 can calculate the similarity between a plurality of imitation features and user-defined features, and respond to the first similarity between the first imitation feature and the user-defined feature among the plurality of imitation features degree is the highest similarity from A first imitation feature is selected from the plurality of imitation features. For example, the computing module 121 may select the simulated regional rainfall information corresponding to 400 mm, the simulated regional rainfall information corresponding to The simulated area rainfall information corresponding to 500 mm is selected from the simulated area rainfall information corresponding to 500 mm and the simulated area rainfall information corresponding to 600 mm.

In one embodiment, after selecting the first imitation feature corresponding to the highest similarity, the operation module 121 may respond to the second similarity between the second imitation feature and the user-defined feature among the plurality of imitation features as Select the second fake feature from the remaining fake features (that is, a plurality of fake features that have not been selected) based on the second highest similarity. For example, after selecting the simulated regional rainfall information corresponding to 500 mm, the computing module 121 may respond that the similarity between the simulated regional rainfall information corresponding to 600 mm and the regional rainfall information corresponding to 540 mm is the second highest similarity The simulated area rainfall information corresponding to 600 mm is selected from the simulated area rainfall information corresponding to 400 mm and the simulated area rainfall information corresponding to 600 mm.

In step S430 , the calculation module 121 can generate an approximate simulation result matching the user-defined feature according to the first imitation simulation result based on the first similarity, and can output the approximate simulation result through the transceiver 130 . Specifically, after selecting the first imitation feature according to the first similarity, the computing module 121 can select the first imitation simulation result from the plurality of imitation simulation results stored in the simulation result database 125 according to the first similarity . Then, the calculation module 121 can be based on the first imitation feature, the first imitation simulation result corresponding to the first imitation feature and User-defined features produce approximate simulation results.

For example, the calculation module 121 can use interpolation or extrapolation based on the simulated regional rainfall information of 500 mm (ie: the first simulated feature), the flood judgment result corresponding to the simulated regional rainfall information of 500 mm (ie : the first imitation simulation result) and the regional rainfall information corresponding to 540 mm (ie: user-defined features) to generate a flood judgment result (ie: approximate simulation result) corresponding to the regional rainfall information of 540 mm. Because the approximate simulation result can be generated only by using the pre-generated first imitation simulation result, without using simulation software. Therefore, approximate simulation results can be generated very quickly.

In one embodiment, after selecting the first imitation feature and the second imitation feature according to the first similarity and the second similarity, the computing module 121 can select A first forgery simulation result is selected from the plurality of forgery simulation results, and a second forgery simulation result is selected from the plurality of forgery simulation results stored in the simulation result database 125 according to the second similarity. Then, the operation module 121 can calculate the first imitation feature, the first imitation simulation result corresponding to the first imitation feature, the second imitation feature, the second imitation feature corresponding to the second imitation feature based on the interpolation method or the extrapolation method. Simulation results as well as user-defined features produce approximate simulation results.

For example, the calculation module 121 can use interpolation or extrapolation based on the simulated regional rainfall information of 500 mm (ie: the first simulated feature), the flood judgment result corresponding to the simulated regional rainfall information of 500 mm (ie : the first imitation simulation result), the rainfall information of the imitation area of 600 mm (ie: the second imitation feature), corresponding to The flood judgment result of the simulated regional rainfall information of 600 mm (ie: the second simulated simulation result) and the corresponding regional rainfall information of 540 mm (ie: user-defined feature) to generate the flood hazard judgment corresponding to the regional rainfall information of 540 mm results (ie: approximate simulation results). Because the approximate simulation results can be generated only by using the pre-generated first simulation simulation results and second simulation simulation results, without using simulation software. Therefore, approximate simulation results can be generated very quickly.

In order to obtain a final simulation result that is more accurate than the approximate simulation result, in step S440, the calculation module 121 can input the user-defined features into the simulation software through the transceiver 130 to generate the final simulation result, and can output the final simulation result through the transceiver 130 Simulation results. For example, the calculation module 121 can input the regional rainfall information corresponding to 540 mm (ie: user-defined features) into the simulation software through the transceiver 130 to generate a flood judgment result corresponding to the regional rainfall information of 540 mm (ie : the final simulation result).

FIG. 5 shows a flow chart of a method for accelerating the generation of simulation results of simulation software according to an embodiment of the present invention, wherein the method can be implemented by the electronic device shown in FIG. 1 . In step S510, input data is acquired. In step S520, the counterfeit data is generated according to the input data. In step S530, input the first counterfeit feature of the counterfeit data into the simulation software to generate a first counterfeit simulation result. In step S540, user-defined features are acquired. In step S550, a first similarity between the first imitation feature and the user-defined feature is calculated. In step S560 , an approximate simulation result matching the user-defined feature is generated according to the first imitation simulation result based on the first similarity. In step S570, an approximate simulation result is output.

To sum up, the electronic device of the present invention can use the generative adversarial network to generate multiple counterfeit features, and input the multiple counterfeit features into the simulation software to generate multiple counterfeit simulation results in advance. When the user wants to obtain a simulation result for a specific feature, the user can input the user-defined feature into the electronic device. The electronic device can compare the similarity between the user-defined feature and a plurality of pre-stored counterfeit features, and select one or more counterfeit features highly related to the user-defined feature. The electronic device can calculate an approximate simulation result according to one or more counterfeit simulation results corresponding to the one or more counterfeit features. Since the approximate simulation results do not need to be generated by simulation software, the approximate simulation results are generated very quickly. In addition to producing approximate simulation results, the electronic device can also input user-defined features into the simulation software to produce more accurate simulation results.

S510, S520, S530, S540, S550, S560, S570: steps

Claims (7)

An electronic device for accelerating the generation of simulation results of simulation software, comprising: a transceiver for obtaining input data, the input data being weather-related image data or data curves; a storage medium for storing multiple modules; and a processor , coupling the storage medium and the transceiver, and accessing and executing the plurality of modules, wherein the plurality of modules include: a machine learning model that generates imitation data according to the input data, and the imitation The data is imitated weather-related image data or data curves; and the computing module inputs the first imitation feature of the imitation data into the simulation software through the transceiver to generate the first imitation simulation result, through The transceiver receives a user-defined feature, calculates a first similarity between the first imitation feature and the user-defined feature, and generates a result corresponding to the first imitation simulation result based on the first similarity. The user defines approximate simulation results for feature matching, and outputs the approximate simulation results via the transceiver. The electronic device according to claim 1, wherein the calculation module is based on one of interpolation and extrapolation according to the first imitation feature, the first imitation simulation result, and the user definition Features yield the approximate simulation results. The electronic device according to claim 1, wherein the computing module inputs the user-defined features into the simulation software through the transceiver to generate a final simulation result, and outputs the final simulation result through the transceiver Simulation results. The electronic device as described in claim 1, wherein the plurality of modules further include: a counterfeit feature database storing a plurality of counterfeiting features including the first counterfeiting feature; a simulation result database storing the counterfeit features corresponding to the a plurality of simulated simulation results of a plurality of simulated features, wherein the computing module calculates a degree of similarity between each of the plurality of simulated features and the user-defined feature, and responds to the first similarity The first forged simulation result corresponding to the first similarity is selected from the plurality of forged simulation results for the highest similarity to generate the approximate simulation result. The electronic device as claimed in item 1, wherein the multiple modules further include: a historical database for storing historical data, wherein the machine learning model includes a generator and a discriminator, wherein the discriminator is based on the Training on historical data, wherein the generator generates candidate forged data according to the input data, and wherein the discriminator determines that the candidate forged data is the forged data. The electronic device according to claim 5, wherein the machine learning model is a generative adversarial network. A method of accelerating the generation of simulation results of simulation software, comprising: Obtain input data, the input data is weather-related image data or data curves; generate imitation data based on the input data, and the imitation data is imitated weather-related image data or data curves; The first imitation feature of the imitation data is input into the simulation software to generate the first imitation simulation result; the user-defined feature is obtained; the first similarity between the first imitation feature and the user-defined feature is calculated; based on generating an approximate simulation result matching the user-defined feature according to the first mimic simulation result; and outputting the approximate simulation result.

Images