WO2022052068A1 - 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备 - Google Patents

一种基于目标可用模型的环境预测方法、装置、程序及其电子设备 Download PDF

Info

Publication number
WO2022052068A1
WO2022052068A1 PCT/CN2020/114896 CN2020114896W WO2022052068A1 WO 2022052068 A1 WO2022052068 A1 WO 2022052068A1 CN 2020114896 W CN2020114896 W CN 2020114896W WO 2022052068 A1 WO2022052068 A1 WO 2022052068A1
Authority
WO
WIPO (PCT)
Prior art keywords
training
model
pollution
data
time
Prior art date
Application number
PCT/CN2020/114896
Other languages
English (en)
French (fr)
Inventor
周晓舟
孙天瑞
梁潇
施尼盖斯·丹尼尔
Original Assignee
西门子（中国）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 西门子（中国）有限公司 filed Critical 西门子（中国）有限公司
Priority to PCT/CN2020/114896 priority Critical patent/WO2022052068A1/zh
Priority to CN202080103886.6A priority patent/CN116157813A/zh
Priority to EP20952856.1A priority patent/EP4195091A4/en
Priority to US18/044,402 priority patent/US20230325563A1/en
Publication of WO2022052068A1 publication Critical patent/WO2022052068A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]

Definitions

  • Embodiments of the present invention relate to the field of computers, and in particular, to a method, device, program, and electronic equipment for training a pollution diffusion model.
  • the embodiments of the present invention provide an environment prediction method, apparatus, program and electronic device based on an available target model, so as to at least partially solve the above problem.
  • an environment prediction method based on an available target model comprising the following steps:
  • the actual environmental predicted value of the time-dependent pollution concentration series at the calibrated location is determined.
  • environmental data for training is determined, the environmental data includes the location of the pollution source in the pollution diffusion area, the leakage intensity of the pollution source, and the meteorological data of the pollution diffusion area;
  • the time-related pollution concentration sequence of the calibration position is generated; the training samples are generated which are characterized by the calibration position and the environmental data used for training, and the time-related pollution concentration sequence of the calibration position is used as the label, In this way, enough training samples can be obtained through simulation, so as to cover various possible leakage situations in practice and improve the adaptability of the target available model.
  • determine the position of the sensor uses computational fluid dynamics algorithm and/or Gaussian simulation algorithm to determine the pollution concentration data of the sensor position under the pollution source data and meteorological data;
  • the pollution concentration data at the sensor location is determined as the environmental data for training. Thereby improving the prediction accuracy of the available model of the target through the sensor data.
  • the training sample is used to perform data fusion based on a fluid mechanics model and a Gaussian simulation model to obtain a time-dependent pollution concentration sequence of the calibrated position,
  • a fluid mechanics model and a Gaussian simulation model to obtain a time-dependent pollution concentration sequence of the calibrated position
  • model training is performed on the initial model by using the features and labels of the training samples, and when the training prediction value of the initial model for the labels of the training samples satisfies a preset condition , it is determined that the initial model at this time is the target available model, wherein the preset condition includes that the difference between the training predicted value of the label and the actual value does not exceed a threshold. , thereby improving the prediction accuracy of the target available model.
  • an evacuation speed is determined; an evacuation path is determined from the actual predicted values of the time-related pollution concentration series of multiple calibration positions according to the evacuation speed, wherein the evacuation path is A time-ordered array containing a plurality of elements, the elements belong to the time-related pollution concentration sequences of the plurality of calibration positions, and the time difference between two adjacent elements in the array does not exceed the distance between the two adjacent elements and The ratio of the evacuation speed, and the contamination concentration of the elements in the array does not exceed a preset contamination threshold. Therefore, an evacuation path suitable for the evacuation speed can be determined according to the prediction result of the diffusion, so as to guide the crowd to evacuate safely and quickly.
  • an environment prediction apparatus based on an available target model comprising:
  • a training module generating training samples based on predetermined environmental data, and using the training samples to perform training based on the fluid mechanics model and the Gaussian simulation model to obtain a target usable model
  • the prediction module uses the target available model to determine the actual environment prediction value of the time-related pollution concentration sequence at the calibrated position.
  • a computer program comprising computer-executable instructions, the computer-executable instructions, when executed, cause at least one processor to perform any one of the methods described above.
  • an electronic device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the In the program, the method described in any one of the foregoing embodiments of the present invention is implemented.
  • a storage medium includes a stored program, wherein when the program runs, a device including the storage medium is controlled to execute any one of the embodiments of the present invention the method described.
  • a training sample is generated by using predetermined environmental data, and the training sample is used for training based on a fluid mechanics model and a Gaussian simulation model, and a target usable model is obtained through training.
  • pollution diffusion occurs, it is only necessary to directly input the current actual environmental data in the target available model to determine the actual predicted value of the time-dependent pollution concentration sequence at the calibrated position, so as to achieve rapid and accurate pollution diffusion. predict.
  • 1a is a schematic flowchart of a training method for a pollution diffusion model provided by an embodiment of the present invention
  • FIG. 1b is a schematic diagram of a calibration position provided by an embodiment of the present invention.
  • FIG. 2 is a schematic structural diagram of an apparatus for environment prediction based on a target availability model provided by an embodiment of the present invention
  • FIG. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
  • S101 Generate a training sample based on predetermined environmental data, use the training sample to perform training based on a fluid mechanics model and a Gaussian simulation model, and obtain a target usable model;
  • S102 Based on the actual environmental data, using the target available model, determine the actual environmental predicted value of the time-related pollution concentration sequence at the calibrated position;
  • the embodiments of the present invention provide a fast and accurate pollution diffusion prediction scheme. Specifically, it includes two aspects, the first aspect: the training of the target available model; the second aspect, the prediction based on the target available model.
  • FIG. 1a is a schematic flowchart of a training method for a pollution diffusion model provided by an embodiment of the present invention, which is applied to pollution diffusion including surface buildings and walkable passages scenarios, including:
  • the environmental data used for training is not the real environmental data. It can be simulated and set by the computer according to the actual needs, and the obtained training samples are no longer the data obtained in the real pollution case, but the simulation data. Simulation data. Since the current meteorological science is relatively mature, this simulation data is closer to the real, so it can also be used as a training sample.
  • the basic way to obtain training samples through simulation is to pre-determine environmental data (including the location, intensity, meteorological data, etc. of pollution sources), that is, to obtain training samples through atmospheric motion equations.
  • the location of the pollution source and the pollution intensity can be set in the pre-divided area grid.
  • the location and pollution intensity of several pollution sources can be set according to the location and possible pollution intensity of potential pollution sources.
  • Meteorological data should include wind direction and speed.
  • the meteorological data may also include other meteorological data such as atmospheric temperature, atmospheric humidity, and the like.
  • atmospheric temperature can be added as a type of environmental data to the environmental data used for training of the pollution source.
  • the pollutant data may also include a pollutant type, and in this case, the meteorological data may also include a meteorological parameter that can affect the pollution diffusion of the pollutant type.
  • the types of pollutants can be, for example, harmful gases or aerosols produced by chemical leaks, or smoke, dust, various gases, or mists produced by fires.
  • the Gaussian diffusion model is suitable for uniform atmospheric conditions and open and flat areas on the ground.
  • the pollution diffusion model of a high-altitude point source can be simulated as follows:
  • C is the pollution concentration at any point in the space
  • q is the leakage intensity of the pollution source
  • x, y, and z are the distances from the three directions of the point in the coordinate system to the origin of the coordinate system
  • H is the height of the pollution source
  • u is the wind speed
  • ⁇ y is the diffusion coefficient in the y direction
  • ⁇ z is the diffusion coefficient in the z direction.
  • ⁇ y and ⁇ z are conventional coefficients that have been determined, that is, the pollution concentration of any point C (x, y, z) can be obtained based on the Gaussian simulation model.
  • the Gaussian simulation model is not very accurate.
  • the fluid mechanics model such as the finite difference method, the finite volume method, or the Lattice Boltzmann Method (LBM) can be used. And so on, to get the pollution concentration at any point in the space.
  • LBM Lattice Boltzmann Method
  • the space can be divided into multiple grids, and the determined pollution source data can be used as the initial condition of a certain point in the grid, while the meteorological data can be used as a constraint condition, using a predetermined atmospheric motion
  • the difference form of the equation is used to calculate the pollution concentration of other grid points one by one, so as to obtain the pollution concentration of any point in the space.
  • a training sample can be obtained based on the pollution concentration of each empty point obtained by the simulation.
  • the time series t1 to tn may have a fixed time interval.
  • the data obtained by the CFD simulation is more accurate than the Gaussian model but takes longer time, the data obtained by the CFD simulation can be called high-precision data, and the data obtained by the Gaussian model can be called low-precision data.
  • training samples can be obtained based on high-precision data and low-precision data according to actual needs, for example, using data fusion methods such as multi-precision methods, regression methods such as neural networks and response surface methods.
  • the specific fusion method can be that, for a designated area, it is pre-delimited into multiple parts, for example, including a complex terrain part and a simple terrain part.
  • the CFD model is used to obtain the training samples
  • the Gaussian simulation model can be used to obtain training samples.
  • training samples in different environments can be obtained. Through enough simulations, a sufficient number of training samples can be obtained.
  • sufficient training samples can cover various possible leakage situations in practice, avoid possible under-fitting or over-fitting in model training, and improve the adaptability of target available models.
  • the number of training samples under different conditions can also be adjusted based on actual needs. For example, if the actual location of the park is more east wind, then more east wind data can be set in the meteorological data, so as to generate more training samples corresponding to east wind conditions.
  • the existence of a corresponding relationship between the environmental data used for training and the training samples means that the training samples obtained from one kind of environmental data are not suitable for other environmental data.
  • the environmental parameters are (pollution source 1, intensity 1, east wind, wind speed 1m/s)
  • the training samples obtained under this condition obviously cannot adapt to the environmental parameters (pollution source 2, intensity 1, south wind, wind speed 2m/s) )conditions of.
  • the obtained training sample may be characterized by the calibration position and the environmental data used for training, and the training sample with the time-related pollution concentration sequence of the calibration position as the label.
  • the calibration position can be preset according to actual needs.
  • the calibrated locations may include, for example, surface structures and walkable passages.
  • FIG. 1b is a schematic diagram of a calibration position provided by an embodiment of the present invention.
  • the demarcated positions B1, B2, B3 and B4 respectively represent 4 surface buildings, and the demarcated positions S1 to S5 respectively correspond to five streets.
  • the training prediction value of the initial model for the label of the training sample can be obtained. Therefore, the model can be adjusted according to the training predicted value of the label and the true value of the label. For example, based on the preset loss function, the parameters in the initial model are adjusted by forward propagation and directional propagation. After several iterations, an initial model for which the training prediction value of the label meets the preset condition is finally obtained, and the initial model at this time is the target available model.
  • the preset condition may be, for example, that the accuracy of the training prediction value of the label relative to the actual value satisfies a certain condition, and so on.
  • the preset condition may be that for any training sample, any element of the difference between the training predicted value of the label of the training sample and the actual value of the label does not exceed a preset threshold .
  • the label of the training sample is a sequence with multiple elements, and the difference between the training predicted value and the true value is also a sequence. If any element of the difference does not exceed the preset difference, it means that the target available model obtained at this time is sufficiently accurate, and in this way, the prediction accuracy of the target available model for real leakage can be improved.
  • some sensors can also be placed in advance.
  • the sensor position can be predetermined, and in this way, the aforementioned simulation mode can still be used to obtain the pollution concentration data of the sensor position under the pollution source data and meteorological data.
  • the pollution concentration data at the sensor location can also be a time-correlated series.
  • the pollution concentration data at the sensor location can be directly obtained as a known parameter. Therefore, in the training sample, the pollution concentration data of the sensor location can also be used as the environmental data for training, that is, the characteristics of the training sample can also include the pollution concentration data of the sensor location. In this way, the pollution concentration data at the sensor location participates in the model prediction as an independent variable, that is, the pollution concentration data at the sensor location is an independent input parameter of the model, because the measured sensor data can be considered as more reliable data than the simulation data. source, thereby improving the accuracy of the model available for the target.
  • the target usable model F After the target usable model F is determined, once the variables are specified (ie, determined from actual environmental data), the average concentration C(Si) on the street at each calibrated location (including) is used as time t, leak intensity q, meteorological data
  • pollution source data can be obtained directly from sensors close to the pollution source, or it can be inferred from pollution-related phenomena, or it can be obtained by pre-judging potential pollution sources; meteorological data can be obtained. It can be obtained by observation, or obtained from the meteorological department, etc. This scheme does not make any specific limitation on this.
  • each location here also includes calibration.
  • a training sample is generated by using predetermined environmental data, and the training sample is used for training based on a fluid mechanics model and a Gaussian simulation model, and a target usable model is obtained through training.
  • pollution diffusion occurs, it is only necessary to directly input the current actual environmental data in the target available model to determine the actual predicted value of the time-dependent pollution concentration sequence at the calibrated position, so as to achieve rapid and accurate pollution diffusion. predict.
  • the pollution concentration data of the sensor location can also be directly obtained through the sensor, so that the pollution source data, meteorological data and the pollution concentration data of the sensor location can be determined as the actual environmental data as input, and the calibration position can be obtained.
  • the actual predicted value of the pollution concentration series can be obtained.
  • the pollution concentration at each calibration location is actually a function of time, pollution source data, meteorological data, and real-time data from sensors.
  • real pollution data can be collected when pollution spreads, which can be used as environmental data input to make the model's prediction results for each location more accurate (because first, the prediction data at the sensor location must conform to the sensor acquisition. obtained pollution concentration data).
  • an evacuation path may also be determined according to the calibration positions at this time. Specifically, first, the evacuation speed can be determined (for example, it is approximately equal to the walking speed of a person 1m/s), so that a set of time-ordered arrays containing multiple elements can be obtained when the starting point and the ending point have been determined. .
  • the elements in the ordered array are arranged in chronological order, and each element is from the pollution concentration sequence of the multiple calibration positions.
  • the calibration position can be an actual walkable channel
  • the time difference between two adjacent elements in the array can be limited to not exceed the distance between the adjacent two elements (that is, the two adjacent elements corresponding to the two adjacent elements.
  • the ratio of the actual distance of each calibration position) to the evacuation speed (so as to ensure that the evacuees can move from one calibration position to another adjacent calibration position on the premise of moving at the evacuation speed), and the array
  • the contamination concentration of the elements in the evacuation route does not exceed the preset contamination threshold (that is, to ensure that the contamination concentration at any point on the evacuation path does not exceed the standard). In this way, when a pollution accident occurs, the evacuation path can be estimated accurately and quickly, so as to realize the safe evacuation of the crowd.
  • the target model of the present application can also be applied to the scene of safe evacuation of indoor floors.
  • the use of the Gaussian simulation model can be reduced in training the target model in advance, and the CFD model can be used more to improve the accuracy of the target model .
  • the pollution source intensity q and the pollution source location caused by the fire can be determined first, and then the target model that has been trained can be used to predict the indoor pollution concentration sequence.
  • the pollutants at this time may include water vapor, carbon monoxide, carbon dioxide, etc. generated during the fire.
  • the pollution concentration sequence of each calibration position in the same floor can be obtained respectively, and the results in different floors can be obtained. (ie at different heights) pollution concentration series.
  • a safe evacuation path can be obtained based on the obtained pollution concentration sequence of the demarcated positions on each floor, so as to facilitate the evacuation of the crowd.
  • the safe evacuation path obtained in this way is also a time-ordered array, and the restrictions on the elements in the array refer to the foregoing description.
  • the array also needs to satisfy that the height difference of the calibration position corresponding to each adjacent element shall not exceed the preset value, or the number of floors corresponding to the calibration position shall not exceed 1, so as to avoid the continuity of the safe evacuation path on the floor.
  • an embodiment of the present invention further provides an environment prediction device based on a target availability model, as shown in FIG. 2 , which is an environment prediction device based on a target availability model provided by an embodiment of the present invention.
  • FIG. 2 Schematic diagram of the structure, the device includes:
  • a training module 201 generating a training sample based on predetermined environmental data, and using the training sample to perform training based on a fluid mechanics model and a Gaussian simulation model to obtain a target usable model;
  • the prediction module 202 uses the target available model to determine the actual environment prediction value of the time-dependent pollution concentration sequence at the calibrated position.
  • the training module 201 determines the environmental data used for training, the environmental data includes the location of the pollution source in the pollution diffusion area, the leakage intensity of the pollution source, and the meteorological data of the pollution diffusion area; Under the environmental data of , calibrate the time-related pollution concentration sequence of the location; generate a training sample characterized by the calibrated location and the environmental data used for training, and labelled with the time-related pollution concentration sequence of the calibrated location.
  • the training module 201 determines the position of the sensor; adopts computational fluid dynamics algorithm and/or Gaussian simulation algorithm to determine the pollution concentration data of the sensor position under the pollution source data and meteorological data; The pollution concentration data at the sensor location is determined as the environmental data for training.
  • the training module 201 uses the training sample to perform data fusion based on the fluid mechanics model and the Gaussian simulation model according to the environment data used for training, to obtain a time-dependent pollution concentration sequence of the calibrated position.
  • the training module 201 uses the features and labels of the training samples to perform model training on the initial model, and when the training prediction value of the initial model for the labels of the training samples satisfies a preset condition, It is determined that the initial model at this time is the target available model, wherein the preset condition includes that the difference between the training predicted value of the label and the actual value does not exceed a threshold.
  • the device further includes an evacuation path determination module 203, which determines an evacuation speed; according to the evacuation speed, an evacuation path is determined from the actual predicted values of the time-related pollution concentration series of a plurality of calibration positions, wherein the The evacuation path is a time-ordered array containing a plurality of elements, the elements belong to the time-related pollution concentration sequence of the plurality of calibration positions, and the time difference between two adjacent elements in the array does not exceed two adjacent ones.
  • the ratio between the distance of the elements and the evacuation speed, and the pollution concentration of the elements in the array does not exceed the preset pollution threshold.
  • a computer program including computer-executable instructions, the computer-executable instructions, when executed, cause at least one processor to perform the prediction described in any one of the embodiments of the present invention method.
  • an electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the present invention when executing the program.
  • FIG. 3 shows a schematic structural diagram of an electronic device provided by an embodiment of the present invention.
  • the electronic device includes: one or more processors 1001, a memory 1002, a display unit 1003, and one or more programs, wherein the one or more programs are stored in the memory, and Configured to be executed by the one or more processors, the one or more programs comprise for performing the method described in any of the embodiments of the present invention.
  • a storage medium is further provided, where the storage medium includes a stored program, wherein when the program runs, a device including the storage medium is controlled to execute any one of the embodiments of the present invention Prediction method described in item.
  • the computer storage medium of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two.
  • the computer readable medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above.
  • Computer readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer disks, hard disks, random access storage media (RAM), read only storage media (ROM), erasable storage media programmable read-only storage media (EPROM or flash memory), optical fiber, portable compact disk read-only storage media (CD-ROM), optical storage media devices, magnetic storage media devices, or any suitable combination of the foregoing.
  • a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
  • a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • a computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program configured for use by or in connection with an instruction execution system, apparatus, or device .
  • Program code embodied on a computer readable medium may be transmitted using any suitable medium including, but not limited to, wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Evolutionary Computation (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Algebra (AREA)
  • Mathematical Physics (AREA)
  • Fluid Mechanics (AREA)
  • Computing Systems (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Operations Research (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Architecture (AREA)
  • Civil Engineering (AREA)
  • Structural Engineering (AREA)
  • Computational Mathematics (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于目标可用模型的环境预测方法、装置、程序及其电子设备。基于预先确定的环境数据生成训练样本，采用训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型（S101）。在污染扩散发生时，只需要在目标可用模型中直接输入当前实际的环境数据，就可以确定出在标定位置（B1-B4，S1-S5）下的与时间相关的污染浓度序列的实际预测值（S102），从而实现快速且准确的污染扩散预测。

Description

一种基于目标可用模型的环境预测方法、装置、程序及其电子设备 技术领域
本发明实施例涉及计算机领域，尤其涉及一种污染扩散模型的训练方法、装置、程序及其电子设备。
背景技术
污染物(例如工业有毒气体)泄漏是一种严重的事故，可能造成严重的人员伤亡。如果发生泄漏，需要立即查明污染物如何扩散，并确定危险区域，以便进行相应的疏散。如果在空旷无障碍的的地方发生污染，一般采用高斯模拟模型即可。
然而，现实中污染也经常性的发生在包含地表建筑和可行走通道的复杂地形区域，例如工业园区。这这种情形下如果采用高斯扩散模型则往往忽略了障碍物和建筑物，得到的扩散结果不够精确，如果进行计算流体动力学(Computational Fluid Dynamics，CFD)模拟往往很慢，不能在短时间内进行及时的预测。
基于此，需要一种更为快速且准确的环境预测方案。
发明内容
为了解决上述问题，本发明实施例提供了一种基于目标可用模型的环境预测方法、装置、程序及其电子设备，以至少部分地解决上述问题。
根据本发明实施例的第一方面，提供了一种基于目标可用模型的环境预测方法，所述方法包括以下步骤：
基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型；
基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值。
可选地，在一种实施例中，确定用于训练的环境数据，所述环境数据包括污染扩散区域的污染源位置、污染源泄露强度以及所述污染扩散区域的气象数据；确定在所述用于训练的环境数据下，标定位置的与时间相关的污染浓度序列；生成以标定位置和用于训练的环境数据为特征，以所述标定位置的与时间相关的污染浓度序列为标签的训练样本，从而 可以通过仿真获取足够的训练样本，从而覆盖实际中各种可能的泄露情形，提高目标可用模型的适应性。
可选地，在一种实施例中，确定传感器位置；采用计算流体力学算法和/或高斯模拟算法，确定在所述污染源数据和气象数据下，所述传感器位置的污染浓度数据；将所述传感器位置的污染浓度数据确定为用于训练的环境数据。从而通过传感器数据提高目标可用模型的预测准确度。
可选地，在一种实施例中，根据所述用于训练的环境数据，采用所述训练样本基于流体力学模型和高斯模拟模型进行数据融合，获得标定位置的与时间相关的污染浓度序列，从而可以获得更准确的训练样本。
可选地，在一种实施例中，采用所述训练样本的特征和标签对所述初始模型进行模型训练，当所述初始模型对于所述训练样本的标签的训练预测值满足预设条件时，确定此时的初始模型为目标可用模型，其中，所述预设条件包括，所述标签的训练预测值与真实值的差不超过阈值。，从而提高目标可用模型的预测准确度。
可选地，在一种实施例中，确定疏散速度；根据所述疏散速度从多个标定位置的与时间相关的污染浓度序列的实际预测值中确定出疏散路径，其中，所述疏散路径为包含多个元素的时间有序数组，所述元素属于所述多个标定位置的与时间相关的污染浓度序列，所述数组中相邻两个元素的时间差不超过相邻两个元素的距离和所述疏散速度的比值，且，所述数组中的元素的污染浓度不超过预设污染阈值。从而可以根据对于扩散的预测结果确定出与疏散速度相适应的疏散路径，以指导人群安全快速疏散。
根据本发明实施例的第二方面，还提供一种基于目标可用模型的环境预测装置，所述装置包括：
训练模块，基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型；
预测模块，基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值。
根据本发明实施例的第三方面，提供了一种计算机程序，包括计算机可执行指令，所述计算机可执行指令在被执行时使至少一个处理器执行前述任意一项所述的方法。
根据本发明实施例的第四方面，提供了一种电子设备，其包括：包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述程序时实现本发明实施例前述任一项所述的方法。
根据本发明实施例的第五方面，提供了一种存储介质，所述存储介质包括存储的程序， 其中，在所述程序运行时控制包括所述存储介质的设备执行本发明实施例任意一项所述的方法。
本发明实施例中，通过预先确定的环境数据生成训练样本，并采用训练样本基于流体力学模型和高斯模拟模型进行训练，训练得到目标可用模型。在污染扩散发生时，只需要在目标可用模型中直接输入当前实际的环境数据，就可以确定出在标定位置下的与时间相关的污染浓度序列的实际预测值，从而实现快速且准确的污染扩散预测。
附图说明
以下附图仅旨在于对本发明做示意性说明和解释，并不限定本发明的范围。其中，
图1a为本发明实施例所提供的一种污染扩散模型的训练方法的流程示意图；
图1b为本发明实施例所提供的一种对于标定位置的示意图；
图2为本发明实施例所提供的一种基于目标可用模型的环境预测装置的结构示意图；
图3为本发明实施例所提供的一种电子设备结构示意图；
附图标记列表：
S101：基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型；
S102：基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值；
B1、B2、B3、B4、S1、S2、S3、S4和S5：标定位置；
201：训练模块；
202：预测模块；
203：疏散路径确定模块；
1001：处理器；
1002：存储器；和
1003：显示单元。
具体实施方式
为了对本发明实施例的技术特征、目的和效果有更加清楚的理解，现对照附图说明本发明实施例的具体实施方式。
在工业园区发生污染扩散时，由于地表存在大量建筑和可行走通道，此时的高斯模拟 模型预测效果准确度太低，而采用计算流体力学(Computational Fluid Dynamics，CFD)模型进行预测效率又太差，基于此，本发明实施例提供一种快速且准确的污染扩散预测方案。具体而言，包括两个方面，第一方面：目标可用模型的训练；第二方面，基于目标可用模型的预测。
对于本发明实施例的第一方面，如图1a所示，图1a为本发明实施例所提供的一种污染扩散模型的训练方法的流程示意图，应用于包含地表建筑和可行走通道的污染扩散场景中，包括：
S101，基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型。
获取得到训练样本的方式可以是从实际污染的案例中得到。然而，这种数据实际上比较少。
因此，通常还可以采用仿真的方式来得到训练样本。此时用于训练的环境数据就并不是真实的环境数据，其可以根据实际需要使用计算机进行模拟设定，得到的训练样本也就不再是真实的污染案例中所得到的数据，而是模拟仿真数据。由于当前的气象科学较为成熟，这种仿真数据比较接近真实，因此也可以用来作为训练样本。
通过仿真得到训练样本的基本方式即为，预先确定环境数据(包括污染源的位置，强度，气象数据等等)，即可以通过大气运动方程来得到训练样本。
在实际应用中，可以在预先已经划分好的区域网格中，对污染源的位置和污染强度进行设置。通常而言，在一个已经确定的工业园区内，可以根据潜在的污染源的位置和可能污染强度设置好若干污染源的位置和污染强度。
气象数据中应当包含风向和风速。对于某些特定类型的污染源而言，气象数据还可以包括其他的气象数据包括诸如大气温度、大气湿度等等。例如，如果一个潜在污染源可能产生的是悬浮颗粒污染物，那么在该污染源的用于训练的环境数据中，就可以加上大气湿度作为环境数据的一种。即污染物数据中还可以包括污染物类型，此时气象数据中还可以包括与能够影响该污染物类型进行污染扩散的气象参数。
在实际应用中，污染物类型可以是诸如化学泄露所产生的有害气体或者气溶胶等等，也可以是由于火灾所产生的烟尘、各类气体或者雾等等。
因此，只需要给定标定位置，就可以确定在所述用于训练的环境数据下，所述标定位置的与时间相关的污染浓度序列，从而生成以标定位置和用于训练的环境数据作为训练特征，以与时间相关的污染浓度序列作为标签的训练样本。
对于高斯模拟模型而言，高斯扩散模型适用于均一的大气条件，以及地面开阔平坦的 地区，一个高空点源的污染扩散模式可以采用如下方式进行模拟：
Figure PCTCN2020114896-appb-000001
其中，C为空间中任意点的污染浓度，q为污染源泄露强度，x、y、z分别为点在坐标系中的三个方向到坐标系原点的距离，H为污染源高度，u为风速，σ y为y方向上扩散系数，σ z为z方向上的扩散系数。
因此，只需要确定了q、u、H，而σ y和σ z是已经确定的常规系数，即可以基于高斯模拟模型得到任意点C(x，y，z)的污染浓度。
在对于复杂地形时，高斯模拟模型不太准确，在确定了环境数据时，即可以采用对于流体力学模型，诸如有限差分法、有限体积法或者格子玻尔兹曼方法(Lattice Boltzmann Method,LBM)等等，来得到空间中任一点的污染浓度。
例如，采用有限差分法时，可以将空间划分为多个网格，将已经确定的污染源数据作为网格中的某个点的初始条件，而气象数据则作为约束条件，采用预先确定的大气运动方程的差分形式来逐一计算其他网格点的污染浓度，从而得到空间中任一点的污染浓度。进而，可以基于仿真模拟得到的空点各点的污染浓度来得到训练样本。
显然，在同样的用于训练的环境数据中下，不同标定位置的污染浓度序列是不同的。污染浓度序列可以是一个有序数组的形式，例如，对于第i个标定位置，其污染浓度序列P i＝(P t1,P t2，……，P tn)。时间序列t1至tn可以存在固定的时间间隔。
具体而言，由于CFD模拟得到的数据比高斯模型更精确但是耗时更长，可以将CFD模拟得到的数据称为高精度数据，而将高斯模型得到的数据称为低精度数据。从而可以根据实际需要，例如采用数据融合方法，如多精度度法，回归方法如神经网络和响应面法等基于高精度数据和低精度数据来得到训练样本。
具体的融合方式可以是，对于一块指定的区域，预先将其划定为多个部分，例如包括复杂地形部分和简单地形部分，对于复杂地形部分即采用CFD模型来得到训练样本，而对于简单地形部分，则可以采用高斯模拟模型来得到训练样本。通过数据融合的方式，可以兼顾得到训练样本的精度和计算效率。
通过对于污染源数据和气象数据进行组合，再基于仿真模拟，即可以得到在不同环境下的训练样本。通过足够多次的仿真模拟，即可以得到数量足够的训练样本。在模型训练中，充足的训练样本可以覆盖实际中各种可能的泄露情形，避免模型训练中可能存在的欠 拟合或者过拟合现象，提高目标可用模型的适应性。
此外，还可以基于实际需要对于不同条件下的训练样本的数量进行调整。例如，如果实际中园区所处的位置东风比较多，那么即可以在气象数据中设置更多的东风数据，从而生成更多的对应于东风条件下的训练样本。
用于训练的环境数据和训练样本存在对应关系指的是，在一种环境数据所得到的训练样本，不适应于其它环境数据。例如，假设环境参数为(污染源1，强度1，东风，风速1m/s)，在这种条件下得到的训练样本显然不能适应环境参数为(污染源2，强度1，南风，风速2m/s)的条件。
得到的训练样本可以是以标定位置和用于训练的环境数据为特征，以所述标定位置的与时间相关的污染浓度序列为标签的训练样本。
标定位置可以根据实际需要预先设定。例如标定位置可以包括诸如地表建筑和可行走通道。标定位置可以是多个，一个标定位置可以与一个地标建筑或者可行走通道一一对应。如图1b所示，图1b为本发明实施例所提供的一种对于标定位置的示意图。在该示意图中标定位置B1、B2、B3和B4分别代表了4个地表建筑，而标定位置S1至S5而分别对应了五个街道。
以所述训练样本的特征(即标定位置、和用于训练的环境数据)作为初始模型的输入，即可以得到初始模型对于所述训练样本的标签的训练预测值。从而可以根据标签的训练预测值和所述标签的真实值对模型进行调整，例如基于预设的损失函数，采用正向传播、方向传播等方式来调整初始模型中的参数，经过对于初始模型的多次迭代，最终得到对于标签的训练预测值满足预设条件的初始模型，此时的初始模型即为目标可用模型。
预设条件可以是诸如标签的训练预测值相对于真实值的准确率满足一定的条件等等。在一种实施例中，预设条件可以是对于任一的训练样本，初始模型对于该训练样本的标签的训练预测值与所述标签的真实值的差中的任一元素不超过预设阈值。如前所述，训练样本的标签是一个包含多个元素的序列，其训练预测值与真实值的差也是一个序列。如果差中的任一元素都不出超过预设差值，说明此时得到的目标可用模型已经足够准确，通过该方式可以提高目标可用模型对于现实泄露的预测准确度。
在实际应用中，对于一个确定的园区，其中还可以事先放置一些传感器。此时，传感器位置是可以被预先确定，在这种方式下仍然可以采用前述的仿真模式，来得到在所述污染源数据和气象数据下，所述传感器位置的污染浓度数据。传感器位置的污染浓度数据也可以是一个和时间相关的序列。
由于在实际预测的时候，传感器位置的污染浓度数据是可以直接获取得到而作为已知 参数的。因此，在训练样本中，传感器位置的污染浓度数据也可以作为用于训练的环境数据，即训练样本的特征中也可以包含传感器位置的污染浓度数据。在这种方式下，传感器位置的污染浓度数据作为自变量参与了模型预测，即在传感器位置上的污染浓度数据是模型的独立输入参数，因为传感器实测数据可以认为是比仿真数据更可靠的数据来源，从而提高目标可用模型的准确性。
S102，基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值。
在目标可用模型F被确定以后，一旦变量被指定(即通过实际的环境数据来确定)，每个标定位置(包括)街道上的平均浓度C(Si)作为时间t、泄漏强度q、气象数据(包括风速v和风向θ)的函数C(s i)＝F(t,q,v,θ)可以很快地被预测出来。
实际的环境数据可以通过各种途径来获取，例如污染源数据可以是通过靠近污染源的传感器直接获得，也可以是通过污染相关的现象推测得到，还可以是对于潜在污染源的预先判定得到；气象数据可以是观测得到，也可以是从气象部门获取等等。本方案对此不作具体限定。
在已经被训练好的目标可用模型中，只需输入实际的环境数据，即可以得到在各个位置下的与时间相关的污染浓度序列的实际环境预测值，此处的各个位置显然也包含了标定位置。
本发明实施例中，通过预先确定的环境数据生成训练样本，并采用训练样本基于流体力学模型和高斯模拟模型进行训练，训练得到目标可用模型。在污染扩散发生时，只需要在目标可用模型中直接输入当前实际的环境数据，就可以确定出在标定位置下的与时间相关的污染浓度序列的实际预测值，从而实现快速且准确的污染扩散预测。
在一种实施例中，如果在模型训练的过程中加入了传感器的相关数据。那么在实际预测的时候，还可以直接通过传感器获取传感器位置的污染浓度数据，从而将所述污染源数据、气象数据和传感器位置的污染浓度数据确定为实际的环境数据作为输入，而得到标定位置下的污染浓度序列的实际预测值。
由于在模型训练的时候即考虑了传感器因素，此时每个标定位置的的污染浓度实际上是时间、污染源数据、气象数据以及来自传感器的实时数据的函数。通过预先设置传感器，在发生污染扩散时可以采集得到真实的污染数据，作为环境数据输入可以使得模型对于每个地点的预测结果更准确(因为首先在传感器位置上的预测数据就必须要符合传感器获取得到的污染浓度数据)。
在一种实施例中，当存在多个标定位置时，此时还可以根据标定位置确定出一条疏散路径。具体而言，首先可以确定出疏散速度(例如，约等于人的行走速度1m/s)，从而可以在已经确定了从起点和终点的情形下，得到一组包含多个元素的时间有序数组。
该有序数组中的元素是按照时间先后顺序依次排列，每个元素均来自于所述多个标定位置的污染浓度序列。并且，由于标定位置可以是实际中的可行走通道，那么可以通过限定所述数组中相邻两个元素的时间差不超过相邻两个元素的距离(即相邻的两个元素所对应的两个标定位置的现实距离)和所述疏散速度的比值(从而保证疏散人员可以在以疏散速度进行移动的前提下，从一个标定位置走到相邻的另一个标定位置)，且，所述数组中的元素的污染浓度不超过预设污染阈值(即保证在疏散路径上任一点的污染浓度不会超标)。通过该方式，当污染事故发生时，即可以准确快速的估计出疏散路径，实现人群的安全疏散。
本申请的目标模型还可以应用于室内楼层的安全疏散的场景中。在这种场景中，由于高斯模型的准确性将会受到较大的倾向，因此，可以预先在训练目标模型减少高斯模拟模型的使用，而更多的采用CFD模型，以提高目标模型的准确性。
在训练得到目标模型之后，当高楼室内发生火灾时，可以首先确定因起火而导致的污染源强度q和污染源位置，然后即可以采用前述已经训练得到的目标模型进行室内污染浓度序列的预测。此时，由于火灾发生在室内，即可以认为风速v＝0，风向θ＝0，从而可以计算出任意时刻t的的标定位置的污染浓度序列C(s i)＝F(t,q,0,0)。此时的污染物可以包括火灾过程中产生的水汽、一氧化碳、二氧化碳等等。
进一步地的，由于楼层数是和楼层高度直接相关的，因此通过确定标定位置的实际高度，即可以分别得到是同一楼层内(同一高度下)各标定位置的污染浓度序列，和得到不同楼层内(即不同高度下)的污染浓度序列。进而可以基于得到的各楼层上的标定位置的污染浓度序列来得到一条安全疏散路径，以便于人群进行疏散。在这种方式下得到的安全疏散路径中同样是一个时间有序数组，数组中的元素的限制参照前述说明。此外该数组中还需要满足各相邻的元素对应的标定位置的高度差不得超过预设值，或者标定位置对应的楼层数不超过1，以避免安全疏散路径在楼层上的连续性。
与第一方面对应的，本发明实施例还提供一种基于目标可用模型的环境预测装置，如图2所示，图2为本发明实施例所提供的一种基于目标可用模型的环境预测装置的结构示意图，该装置包括：
训练模块201，基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型；
预测模块202，基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值。
可选地，所述训练模块201，确定用于训练的环境数据，所述环境数据包括污染扩散区域的污染源位置、污染源泄露强度以及所述污染扩散区域的气象数据；确定在所述用于训练的环境数据下，标定位置的与时间相关的污染浓度序列；生成以标定位置和用于训练的环境数据为特征，以所述标定位置的与时间相关的污染浓度序列为标签的训练样本。
可选地，所述训练模块201，确定传感器位置；采用计算流体力学算法和/或高斯模拟算法，确定在所述染源数据和气象数据下，所述传感器位置的污染浓度数据；将所述传感器位置的污染浓度数据确定为用于训练的环境数据。
可选地，所述训练模块201，根据所述用于训练的环境数据，采用所述训练样本基于流体力学模型和高斯模拟模型进行数据融合，获得标定位置的与时间相关的污染浓度序列。
可选地，所述训练模块201，采用所述训练样本的特征和标签对所述初始模型进行模型训练，当所述初始模型对于所述训练样本的标签的训练预测值满足预设条件时，确定此时的初始模型为目标可用模型，其中，所述预设条件包括，所述标签的训练预测值与真实值的差不超过阈值。
可选地，所述装置还包括疏散路径确定模块203，确定疏散速度；根据所述疏散速度从多个标定位置的与时间相关的污染浓度序列的实际预测值中确定出疏散路径，其中，所述疏散路径为包含多个元素的时间有序数组，所述元素属于所述多个标定位置的与时间相关的污染浓度序列，所述数组中相邻两个元素的时间差不超过相邻两个元素的距离和所述疏散速度的比值，且，所述数组中的元素的污染浓度不超过预设污染阈值。
本发明实施例的第三方面，还提供一种计算机程序，包括计算机可执行指令，所述计算机可执行指令在被执行时使至少一个处理器执行本发明实施例中任意一项所述的预测方法。
本发明实施例的第四方面，还提供一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述程序时实现本发明任一实施例中任意一项所述的预测方法。
图3示出了本发明实施例所提供的一种电子设备结构示意图。如图3所示，电子设备包括：一个或多个处理器1001、存储器1002、显示单元1003、以及一个或多个程序，其 中，所述一个或多个程序被存储在所述存储器中，并且被配置为由所述一个或多个处理器执行，所述一个或多个程序包括用于执行本发明任一实施例中所述的方法。
本发明实施例的第五方面，还提供了一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制包括所述存储介质的设备执行本发明实施例中任意一项所述的预测方法。
需要说明的是，本发明的计算机存储介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读介质例如可以但不限于是电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储介质(RAM)、只读存储介质(ROM)、可擦式可编程只读存储介质(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储介质(CD-ROM)、光存储介质件、磁存储介质件、或者上述的任意合适的组合。在本发明中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。而在本发明中，计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输配置为由指令执行系统、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于：无线、电线、光缆、RF等等，或者上述的任意合适的组合。
应当理解，虽然本发明是按照各个实施例描述的，但并非每个实施例仅包含一个独立的技术方案，说明书的这种叙述方式仅仅是为清楚起见，本领域技术人员应当将说明书作为一个整体，各实施例中的技术方案也可以经适当组合，形成本领域技术人员可以理解的其他实施方式。
以上仅为本发明实施例示意性的具体实施方式，并非用以限定本发明实施例的范围。任何本领域的技术人员，在不脱离本发明实施例的构思和原则的前提下所作的等同变化、修改与结合，均应属于本发明实施例保护的范围。

Claims (10)

  1. 一种基于目标可用模型的环境预测方法，所述方法包括以下步骤：
    基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型(S101)；
    基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值(S102)。
  2. 如权利要求1所述的方法，基于预先确定的环境数据生成训练样本，包括：
    确定用于训练的环境数据，所述环境数据包括污染扩散区域的污染源位置、污染源泄露强度以及所述污染扩散区域的气象数据；
    确定在所述用于训练的环境数据下，标定位置的与时间相关的污染浓度序列；
    生成以标定位置和用于训练的环境数据为特征，以所述标定位置的与时间相关的污染浓度序列为标签的训练样本。
  3. 如权利要求2所述的方法，确定用于训练的环境数据，还包括：
    确定传感器位置；
    采用计算流体力学算法和/或高斯模拟算法，确定在所述污染源数据和气象数据下，所述传感器位置的污染浓度数据；
    将所述传感器位置的污染浓度数据确定为用于训练的环境数据。
  4. 如权利要求2所述的方法，确定在所述用于训练的环境数据下，标定位置的与时间相关的污染浓度序列，包括：
    根据所述用于训练的环境数据，采用所述训练样本基于流体力学模型和高斯模拟模型进行数据融合，获得标定位置的与时间相关的污染浓度序列。
  5. 如权利要求2所述的方法，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型，包括：
    采用所述训练样本的特征和标签对所述初始模型进行模型训练，当所述初始模型对于所述训练样本的标签的训练预测值满足预设条件时，确定此时的初始模型为目标可用模型，其中，所述预设条件包括，所述标签的训练预测值与真实值的差不超过阈值。
  6. 如权利要求2所述的方法，所述方法还包括：
    确定疏散速度；
    根据所述疏散速度从多个标定位置的与时间相关的污染浓度序列的实际环境预测值中确定出疏散路径，其中，所述疏散路径为包含多个元素的时间有序数组，所述元素属于 所述多个标定位置的与时间相关的污染浓度序列，所述数组中相邻两个元素的时间差不超过相邻两个元素的距离和所述疏散速度的比值，且，所述数组中的元素的污染浓度不超过预设污染阈值。
  7. 一种基于目标可用模型的环境预测装置，所述装置包括：
    训练模块(301)，基于预先确定的环境数据生成训练样本，采用所述训练样本基于流体力学模型和高斯模拟模型进行训练，获得目标可用模型；
    预测模块(303)，基于实际的环境数据，采用所述目标可用模型，确定在标定位置下的与时间相关的污染浓度序列的实际环境预测值。
  8. 一种计算机程序，包括计算机可执行指令，所述计算机可执行指令在被执行时使至少一个处理器执行权利要求1至6任意一项所述的方法。
  9. 一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述程序时实现如权利要求1至6任一项所述的方法。
  10. 一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制包括所述存储介质的设备执行权利要求1至6任意一项所述的方法。
PCT/CN2020/114896 2020-09-11 2020-09-11 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备 WO2022052068A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
PCT/CN2020/114896 WO2022052068A1 (zh) 2020-09-11 2020-09-11 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备
CN202080103886.6A CN116157813A (zh) 2020-09-11 2020-09-11 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备
EP20952856.1A EP4195091A4 (en) 2020-09-11 2020-09-11 METHOD AND APPARATUS FOR PREDICTING A MODEL-BASED ENVIRONMENT AVAILABLE ON A TARGET, PROGRAM AND ELECTRONIC DEVICE
US18/044,402 US20230325563A1 (en) 2020-09-11 2020-09-11 Target Available Model-Based Environment Prediction Method and Apparatus, Program, and Electronic Device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/114896 WO2022052068A1 (zh) 2020-09-11 2020-09-11 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备

Publications (1)

Publication Number Publication Date
WO2022052068A1 true WO2022052068A1 (zh) 2022-03-17

Family

ID=80632613

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/114896 WO2022052068A1 (zh) 2020-09-11 2020-09-11 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备

Country Status (4)

Country Link
US (1) US20230325563A1 (zh)
EP (1) EP4195091A4 (zh)
CN (1) CN116157813A (zh)
WO (1) WO2022052068A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115081725A (zh) * 2022-07-07 2022-09-20 中国长江三峡集团有限公司 一种喀斯特地貌下污染物溶质分布的预测方法及装置
CN116306377A (zh) * 2023-04-04 2023-06-23 中国石油大学(华东) 一种加氢站泄漏事故后果快速预测方法及系统
CN117689980A (zh) * 2024-02-04 2024-03-12 青岛海尔科技有限公司 构建环境识别模型的方法、识别环境的方法及装置、设备
CN117726046A (zh) * 2024-02-07 2024-03-19 中科三清科技有限公司 点源排放预测方法、装置、存储介质及电子设备
CN116778661B (zh) * 2023-07-05 2024-06-07 深圳市华翌科技有限公司 一种烟感智能预警方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116631530B (zh) * 2023-05-29 2024-02-13 智感技术(天津)有限公司 污染物扩散风险识别方法、装置及设备
CN117457098A (zh) * 2023-10-27 2024-01-26 生态环境部南京环境科学研究所 跨界区饮用水源地污染事故预警方法、装置、介质和设备

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104008229A (zh) * 2014-04-30 2014-08-27 北京大学 一种街区污染物浓度分布模型建立方法
CN106651036A (zh) * 2016-12-26 2017-05-10 东莞理工学院 空气质量预报系统
CN106709127A (zh) * 2015-11-12 2017-05-24 国际商业机器公司 利用cfd和卫星数据提取污染排放源
CN110457829A (zh) * 2019-08-15 2019-11-15 王博 一种基于集成大气扩散模型的源项释放反演和扩散预测方法
US20200011843A1 (en) * 2016-02-08 2020-01-09 International Business Machines Corporation Leveraging air/water current variability for sensor network verification and source localization
CN111144055A (zh) * 2019-12-27 2020-05-12 苏州大学 城市环境中有毒重气泄漏浓度分布确定方法、装置及介质
CN111222685A (zh) * 2019-11-21 2020-06-02 江苏方天电力技术有限公司 基于模型迁移的锅炉宽负荷NOx排放浓度预测方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914622B (zh) * 2014-04-04 2017-07-07 清华大学 一种化学品泄漏快速预测预警应急响应决策方法
CN106650825B (zh) * 2016-12-31 2020-05-12 中国科学技术大学 一种机动车尾气排放数据融合系统
CN109146161A (zh) * 2018-08-07 2019-01-04 河海大学 融合栈式自编码和支持向量回归的pm2.5浓度预测方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104008229A (zh) * 2014-04-30 2014-08-27 北京大学 一种街区污染物浓度分布模型建立方法
CN106709127A (zh) * 2015-11-12 2017-05-24 国际商业机器公司 利用cfd和卫星数据提取污染排放源
US20200011843A1 (en) * 2016-02-08 2020-01-09 International Business Machines Corporation Leveraging air/water current variability for sensor network verification and source localization
CN106651036A (zh) * 2016-12-26 2017-05-10 东莞理工学院 空气质量预报系统
CN110457829A (zh) * 2019-08-15 2019-11-15 王博 一种基于集成大气扩散模型的源项释放反演和扩散预测方法
CN111222685A (zh) * 2019-11-21 2020-06-02 江苏方天电力技术有限公司 基于模型迁移的锅炉宽负荷NOx排放浓度预测方法
CN111144055A (zh) * 2019-12-27 2020-05-12 苏州大学 城市环境中有毒重气泄漏浓度分布确定方法、装置及介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4195091A4 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115081725A (zh) * 2022-07-07 2022-09-20 中国长江三峡集团有限公司 一种喀斯特地貌下污染物溶质分布的预测方法及装置
CN115081725B (zh) * 2022-07-07 2023-08-01 中国长江三峡集团有限公司 一种喀斯特地貌下污染物溶质分布的预测方法及装置
CN116306377A (zh) * 2023-04-04 2023-06-23 中国石油大学(华东) 一种加氢站泄漏事故后果快速预测方法及系统
CN116306377B (zh) * 2023-04-04 2024-04-05 中国石油大学(华东) 一种加氢站泄漏事故后果快速预测方法及系统
CN116778661B (zh) * 2023-07-05 2024-06-07 深圳市华翌科技有限公司 一种烟感智能预警方法
CN117689980A (zh) * 2024-02-04 2024-03-12 青岛海尔科技有限公司 构建环境识别模型的方法、识别环境的方法及装置、设备
CN117689980B (zh) * 2024-02-04 2024-05-24 青岛海尔科技有限公司 构建环境识别模型的方法、识别环境的方法及装置、设备
CN117726046A (zh) * 2024-02-07 2024-03-19 中科三清科技有限公司 点源排放预测方法、装置、存储介质及电子设备
CN117726046B (zh) * 2024-02-07 2024-04-30 中科三清科技有限公司 点源排放预测方法、装置、存储介质及电子设备

Also Published As

Publication number Publication date
EP4195091A4 (en) 2024-04-03
US20230325563A1 (en) 2023-10-12
EP4195091A1 (en) 2023-06-14
CN116157813A (zh) 2023-05-23

Similar Documents

Publication Publication Date Title
WO2022052068A1 (zh) 一种基于目标可用模型的环境预测方法、装置、程序及其电子设备
Qiu et al. Atmospheric dispersion prediction and source estimation of hazardous gas using artificial neural network, particle swarm optimization and expectation maximization
KR101885350B1 (ko) 대기오염 통합 관리 시스템
Papadopoulou et al. Optimal sensor placement for time-dependent systems: Application to wind studies around buildings
Labat et al. Assessment of the air change rate of airtight buildings under natural conditions using the tracer gas technique. Comparison with numerical modelling
Wang et al. Locating hazardous gas leaks in the atmosphere via modified genetic, MCMC and particle swarm optimization algorithms
KR20140024541A (ko) 대기 모델링 서비스 제공 시스템
Xue et al. Inverse identification of multiple outdoor pollutant sources with a mobile sensor
Ma et al. Location of natural gas leakage sources on offshore platform by a multi-robot system using particle swarm optimization algorithm
Liu et al. An adaptive particle swarm optimization algorithm for fire source identification of the utility tunnel fire
CN106777893A (zh) 一种利用可移动污染物探测器辨识城市空间多污染物源的方法
Gunawardena et al. Particle swarm optimization for source localization in realistic complex urban environments
Kopka et al. Framework for stochastic identification of atmospheric contamination source in an urban area
Wu et al. Finding the earliest arrival path through a time-varying network for evacuation planning of mine water inrush
Zhai et al. Identifying decaying contaminant source location in building HVAC system using the adjoint probability method
Shao et al. Exposure risk and emergency evacuation modeling of toxic gas leakage in urban areas under the influence of multiple meteorological factors
KR101788389B1 (ko) 가스 확산 모델링 장치 및 방법
Cybyk et al. Simulation of fluid dynamics around complex urban geometries
US8190376B2 (en) System and method for source identification for a chemical release
CN113743015B (zh) 火灾场景数据获取方法、介质及电子设备
Zhu et al. Estimation of the point source parameters by the adjoint equation in the time-varying atmospheric environment with unknown turn-on time
Walenciak et al. Extending spatial data infrastructures 3D by geoprocessing functionality-3D simulations in disaster management and environmental research
JP2014106203A (ja) 拡散物質の拡散状況予測装置、その方法およびプログラム
Cybyk et al. A detailed contaminant transport model for facility hazard assessment in urban areas
Fan et al. Improving gas dispersal simulation for mobile robot olfaction: Using robot-created occupancy maps and remote gas sensors in the simulation loop

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20952856

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020952856

Country of ref document: EP

Effective date: 20230306

NENP Non-entry into the national phase

Ref country code: DE