WO2021062755A1 - 非线性模型的线性化处理方法、装置及存储介质 - Google Patents

非线性模型的线性化处理方法、装置及存储介质 Download PDF

Info

Publication number
WO2021062755A1
WO2021062755A1 PCT/CN2019/109675 CN2019109675W WO2021062755A1 WO 2021062755 A1 WO2021062755 A1 WO 2021062755A1 CN 2019109675 W CN2019109675 W CN 2019109675W WO 2021062755 A1 WO2021062755 A1 WO 2021062755A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
nonlinear
linear
nonlinear model
input
Prior art date
Application number
PCT/CN2019/109675
Other languages
English (en)
French (fr)
Inventor
王德慧
江宁
张拓
田中伟
Original Assignee
西门子股份公司
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 西门子股份公司, 西门子（中国）有限公司 filed Critical 西门子股份公司
Priority to EP19947704.3A priority Critical patent/EP4024259A4/en
Priority to US17/765,109 priority patent/US20220350947A1/en
Priority to PCT/CN2019/109675 priority patent/WO2021062755A1/zh
Priority to CN201980100540.8A priority patent/CN114424196A/zh
Publication of WO2021062755A1 publication Critical patent/WO2021062755A1/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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation

Definitions

  • the present invention relates to the industrial field, in particular to a linearization processing method, device and computer-readable storage medium of a nonlinear model in an integrated energy system.
  • DES Distributed Energy System
  • the process of realizing the above-mentioned integrated energy system requires the establishment of many equipment models, but these equipment usually include many non-linear physical processes (also called basic processes), such as the flow and pressure related processes in the compressor of a gas turbine, and mechanical energy.
  • the process of transforming into pressure energy, etc. so the model of this equipment is usually a non-linear model.
  • Non-linear models are usually more complex in modeling process, but the corresponding accuracy is also relatively high.
  • the non-linear model is directly used to run the simulation of the integrated energy system, it may affect the entire synthesis due to the relatively slow running speed of the non-linear model. Real-time simulation of energy system.
  • one aspect of the embodiments of the present invention proposes a linearization processing method for a nonlinear model, and on the other hand, a linearization processing device for a nonlinear model and a computer-readable storage medium are proposed to target some
  • the model including the non-linear basic process realizes the linearization of the non-linear model, and then is used for the construction of the integrated energy system.
  • a method for linearization processing of a nonlinear model proposed in an embodiment of the present invention includes: determining the value range of each input parameter of the model for the nonlinear model of each device; and dividing the value range of each input parameter Divide into a plurality of sub-intervals based on a plurality of interpolation points; determine a plurality of input sample values in each sub-interval; traverse the input sample value combinations of each input parameter of the model, and use the nonlinear model to obtain each input sample value combination Corresponding output sample value combinations; use all input sample value combinations and their corresponding output sample value combinations to generate a tensor table.
  • the dividing the value range of each input parameter into a plurality of sub-intervals based on a plurality of interpolation points is: based on an equalization criterion, dividing the value range of each input parameter into a complex number based on the plurality of interpolation points Subintervals.
  • the plurality of input sample values are determined by equalization in each sub-interval: based on the equalization criterion, the plurality of input sample values are determined in each sub-interval in a balanced manner.
  • the tensor table is searched according to the current value of each input parameter, and the corresponding data found from the tensor table is used for interpolation processing to obtain the corresponding output value.
  • the nonlinear model of each device is modeled using the following method: for each target nonlinear basic process of each device, the complete design point data is determined; the similarity number supported by the similarity criterion is used. The ratio of the similarity number based on the design point data is established to establish the description formula of the non-linear basis process, and the general model of the non-linear basis process is obtained; the general model includes the non-linear changes with the actual working condition parameters.
  • variable parameters constructing a machine learning algorithm between the actual working condition parameters and the variable parameters, and establishing an association relationship between the machine learning algorithm and the general model; all targets of each type of equipment
  • the general model of the non-linear general process and its associated machine learning algorithm constitute the general model of this kind of equipment; for each target of a specific device of this kind of equipment, the target of the specific device is obtained by the non-linear basic principle process
  • the historical data of actual working condition parameters and variable parameters corresponding to the nonlinear basic theory process are used to train the machine learning algorithm to obtain the variable parameter training model of the target nonlinear basic theory process;
  • the variable parameter training model of the target nonlinear basic process is substituted into the general model of the target nonlinear basic process to obtain the trained model of the target nonlinear basic process of the specific device;
  • the specific The post-training model of the non-linear basic process of all the targets of the equipment constitutes the post-training model of the specific equipment.
  • variable parameter has a preset default value.
  • the equipment includes: gas turbines, heat pumps, internal combustion engines, steam turbines, waste heat boilers, absorption refrigerators, heat generators, multi-effect evaporators, hydrogen production by electrolysis of water, hydrogen production equipment, reverse osmosis, fuel cells , Boiler;
  • the target non-linear basic process of each type of equipment includes one or more of the following processes: heat transfer process, process of converting thermal energy into kinetic energy, pipeline resistance process, flow and pressure related process, thermal energy conversion Mechanical energy process, electrical energy conversion to cooling, thermal energy process, distillation process, evaporation process and filtration process.
  • a linearization processing device for a nonlinear model proposed in an embodiment of the present invention includes: a first processing module for determining the value range of each input parameter of the model for the nonlinear model of each device; second The processing module divides the value range of each input parameter into a plurality of sub-intervals based on a plurality of interpolation points; the third processing module is used to determine a plurality of input sample values in each sub-interval; the fourth processing module is used to Traverse the input sample value combinations of each input parameter of the model, and use the nonlinear model to obtain the output sample value combination corresponding to each input sample value combination; the fifth processing module is used to use all the input sample value combinations and their corresponding The output sample value combination generates a tensor table.
  • the second processing module divides the value range of each input parameter into a plurality of sub-intervals based on a plurality of interpolation points based on an equalization criterion.
  • the third processing module equalizes and determines a plurality of input sample values in each sub-interval based on an equalization criterion.
  • system further includes: a sixth processing module, configured to perform interpolation processing on the tensor table according to the current value of each input parameter during simulation to obtain the corresponding output value.
  • it further includes: a first modeling module, which is used to determine the complete design point data for each target non-linear basis process of each device; the similarity number supported by the similarity criterion is compared with the design-based The ratio of the similarity numbers of point data is established to establish the description formula of the nonlinear basic principle process, and the general model of the nonlinear basic principle process is obtained; the general model includes the non-linear changes with the actual working condition parameters.
  • a first modeling module which is used to determine the complete design point data for each target non-linear basis process of each device; the similarity number supported by the similarity criterion is compared with the design-based The ratio of the similarity numbers of point data is established to establish the description formula of the nonlinear basic principle process, and the general model of the nonlinear basic principle process is obtained; the general model includes the non-linear changes with the actual working condition parameters.
  • Variable parameters construct a machine learning algorithm between the actual working condition parameters and the variable parameters, and establish an association relationship between the machine learning algorithm and the general model; all targets of each type of equipment are non-linear and universal
  • the general model of the process and its associated machine learning algorithms constitute the general model of this kind of equipment; and the second modeling module is used to obtain the non-linear basic process for each target of a specific device of this kind of equipment
  • the historical data of actual working condition parameters and variable parameters corresponding to the target non-linear basis process of a specific device are used to train the machine learning algorithm to obtain the performance of the target non-linear basis process Variable parameter training model; substituting the variable parameter training model of the target non-linear basis process into the general model of the target non-linear basis process to obtain the training of the target non-linear basis process of the specific device Post-model; the post-training model of all target non-linear basic processes of the specific device constitutes the post-training model of the specific device.
  • Another linearization processing device for a nonlinear model proposed in an embodiment of the present invention includes: at least one memory and at least one processor, wherein: the at least one memory is used for storing a computer program; the at least one processor is used for The computer program stored in the at least one memory is invoked to execute the linearization processing method of the nonlinear model as described in any of the above embodiments.
  • the computer-readable storage medium proposed in the embodiment of the present invention has a computer program stored thereon; the computer program can be executed by a processor and implement the linearization processing method described in any of the above embodiments.
  • the piecewise linearization technology is used to process the non-linear model to obtain a tensor table.
  • an interpolation operation is performed based on the tensor table to obtain the required simulation data.
  • the linearized equipment model runs faster and can meet the real-time requirements of simulation.
  • the non-linear basis is established by using the ratio of the similarity number supported by the similarity criterion to the similarity number based on the design point data for each target non-linear basis process of each equipment.
  • the description formula of the management process is used to obtain a general model of the nonlinear basic process, so that the model can be used as a general model and applicable to a class of equipment.
  • the general model includes a variable parameter that changes nonlinearly with the change of actual working condition parameters, and by constructing a machine learning algorithm between the actual working condition parameter and the variable parameter, the variable parameter It can be obtained through machine learning, so that the general model has self-learning ability.
  • the actual working condition parameters and the historical data of the variable parameters corresponding to the target non-linear basis process of the specific equipment are obtained.
  • the historical data use the historical data to train the machine learning algorithm to obtain the variable parameter training model of the target nonlinear basic process, and substitute the variable parameter training model of the target nonlinear basic process into the
  • a trained model of the target nonlinear basic process of the specific device can be obtained, that is, an instantiated model that meets the characteristics of the specific device.
  • the nonlinear model can be made available when the training variable parameters are not available on site, for example, when there is insufficient historical data.
  • modeling method in the embodiment of the present invention can be applied to various non-linear processes of various devices, which is not only convenient for implementation, but also has high accuracy.
  • Fig. 1 is an exemplary flow chart of a method for linearization processing of a nonlinear model in an embodiment of the present invention.
  • Fig. 2 is an exemplary flowchart of a method for modeling a nonlinear model in an embodiment of the present invention.
  • FIG. 3 is an exemplary structure diagram of a linearization processing device of a nonlinear model in an embodiment of the present invention.
  • FIG. 4 is an exemplary structure diagram of another linearization processing device of a nonlinear model in an embodiment of the present invention.
  • FIG. 5 is an exemplary structure diagram of another linearization processing device of a nonlinear model in an embodiment of the present invention.
  • the first processing module 302 Second processing module 303
  • the third processing module 304 Fourth processing module 305
  • Fifth processing module 306 Sixth processing module 307
  • the first modeling module 308 The second modeling module 51 Memory 52 processor 53 bus
  • Fig. 1 is an exemplary flow chart of a method for linearization processing of a nonlinear model in an embodiment of the present invention. As shown in Figure 1, the method may include the following steps:
  • Step 101 For the nonlinear model of each device, determine the value range of each input parameter of the model.
  • the effective power range is 50% to 110% of the rated operating conditions.
  • Step 102 Divide the value range of each input parameter into a plurality of subintervals based on a plurality of interpolation points.
  • the value range of each input parameter can be divided into a plurality of sub-intervals based on a plurality of interpolation points.
  • the interpolation points can be determined by setting more interpolation points for regions with severe nonlinear changes, and setting fewer interpolation points for regions with slow nonlinear changes.
  • 40 points are inserted for the power range, 20 points for the ambient temperature, 5 points for the ambient pressure, etc.
  • Step 103 equalize and determine a plurality of input sample values in each sub-interval.
  • a plurality of input sample values can be equalized and determined in each sub-interval.
  • power and ambient temperature can be divided into equal parts in the actual domain.
  • Step 104 Traverse the input sample value combination of each input parameter of the model, and use the nonlinear model to obtain the output sample value combination corresponding to each input sample value combination.
  • Step 105 Generate a tensor table using all the input sample value combinations and their corresponding output sample value combinations.
  • the efficiency value can be obtained by interpolation.
  • the tensor table can be searched according to the current value of each input parameter, and the corresponding value found from the tensor table can be used for interpolation processing to obtain the corresponding output value.
  • the current value of each input parameter may be a real value or a hypothetical value.
  • the interpolation algorithm can be selected according to the actual situation, for example, a linear interpolation method or a nonlinear interpolation method can be selected.
  • linear interpolation may be used for points that are close to each other, and nonlinear interpolation may be used for points that are far apart.
  • the value of other output variables corresponding to the required temperature, pressure, and performance can be obtained through methods such as three-dimensional temperature, pressure, and power spline interpolation.
  • This general method uses a general program, regardless of any specific model, such as heat pump, internal combustion engine, heat exchanger, etc., can be processed by this piece of code tool.
  • Fig. 2 is an exemplary flowchart of a method for modeling a nonlinear model in an embodiment of the present invention. As shown in Figure 2, the method may include the following steps:
  • Step 201 Determine the complete design point data for each target non-linear basis process of each type of equipment.
  • the basic process can sometimes be called a physical process, such as the heat transfer process, the electrical energy conversion process, and the aforementioned flow and pressure related processes.
  • the basic process of interest can be determined, that is, the basic process that needs to be modeled.
  • These basic processes that need to be modeled are called target basic processes, and the non-linear target basic processes can be called It is the target nonlinear basis process.
  • the target non-linear basic process can include: the process of flow and pressure in the expansion turbine, the process of energy conversion between thermal energy and mechanical energy, etc.; for the heat pump, the target non-linear basic process can include: Thermal process, the process of converting electrical energy into thermal energy, the chemical process that uses high-temperature thermal energy to separate substances in solution, electrochemical process, pipeline resistance process, flow and pressure related processes, etc.
  • the basic process can include: flow and pressure related process, thermal energy conversion process, electrical energy conversion process, thermal energy process, pipeline resistance process, heat transfer process, distillation process, evaporation process, filtration process, chemical reaction process, electrochemical process One or more of etc.
  • the complete design point data can be restored based on the basic publicly published design parameters and the general design point information provided by the manufacturer to the user.
  • its design point data can include pressure ratio, air flow, etc. Based on these design point data, it can derive its efficiency, inlet resistance, and air extraction volume and other related designs that are not available to users. parameter.
  • Step 202 Use the ratio of the similarity number supported by the similarity criterion to the similarity number based on the design point data to establish a description formula of the non-linear basis process to obtain a general model of the non-linear basis process; the general model It includes variable parameters that change non-linearly with changes in actual operating conditions.
  • the actual working condition parameters refer to the parameters of a specific equipment that change with the actual working condition parameters.
  • the variable parameter may have a preset default value.
  • G1 is the flow rate
  • T1 is the temperature
  • P1 is the pressure
  • G0 is the corresponding design point flow
  • T0 is the corresponding design point temperature
  • P0 is the corresponding design point pressure
  • f() is a function
  • coefficients a and b are variable parameters that change non-linearly with the actual working condition parameters.
  • a default value can also be set for the variable parameters a and b.
  • IGV is the angle of the imported transmissible vane.
  • Step 203 Construct a machine learning algorithm between the actual working condition parameters and the variable parameters.
  • a machine learning algorithm between the actual working condition parameter and the variable parameter can be constructed based on a machine learning big data analysis method such as a neural intelligent network or a support vector machine.
  • step 204 the general model of the non-linear basic process of all targets of each type of equipment and its associated machine learning algorithm constitute a general model of the type of equipment.
  • the method may further include the following steps:
  • Step 205 For each target non-linear basis process of a specific device of this type of equipment, obtain historical data of actual working condition parameters and variable parameters corresponding to the target non-linear basis process of the specific equipment, and use The historical data trains the corresponding machine learning algorithm to obtain the variable parameter training model of the target nonlinear basic process.
  • a set of historical data of actual working condition parameters are used as input sample values, and the historical data of variable parameters corresponding to the set of historical data of actual working condition parameters are used as output sample values.
  • the machine learning algorithm is trained on the input sample value and the corresponding output sample value of, and the self-learning model with variable parameters, also called training model, can be obtained.
  • the historical data of the relevant actual working condition parameters of the gas turbine on site and the historical data of the corresponding variable parameters can be obtained, and the input and output sample sets can be obtained.
  • a training model with variable parameters a and b can be obtained.
  • Step 206 Substitute the variable parameter training model of the target non-linear basis process into the general model of the target non-linear basis process to obtain the trained model of the target non-linear basis process of the specific device .
  • the trained model is a self-learning model with learning ability.
  • variable parameter training model of the target nonlinear basis process can be substituted into the general model of the target nonlinear basis process.
  • step 207 the trained models of all target non-linear basic processes of the specific equipment constitute the post-training model of the specific equipment.
  • the input parameters of the trained model may include all input parameters required by the trained model of the target nonlinear basis process.
  • the linearization method of the nonlinear model in the embodiment of the present invention and one of the modeling methods are described above in detail, and the linearization processing device of the nonlinear model in the embodiment of the present invention and one of the modeling methods are described below.
  • the device is described in detail.
  • the device in the embodiment of the present invention can be used to implement the method in the embodiment of the present invention. For details that are not disclosed in the device embodiment of the present invention, refer to the corresponding description in the method embodiment of the present invention, which will not be repeated here.
  • FIG. 3 is an exemplary structure diagram of a linearization processing device of a nonlinear model in an embodiment of the present invention. As shown in FIG. 3, the device may include: a first processing module 301, a second processing module 302, a third processing module 303, a fourth processing module 304, and a fifth processing module 305.
  • the first processing module 301 is configured to determine the value range of each input parameter of the model for the nonlinear model of each device.
  • the second processing module 302 divides the value range of each input parameter into a plurality of sub-intervals based on a plurality of interpolation points.
  • the second processing module 302 may divide the value range of each input parameter into a plurality of sub-intervals based on a plurality of interpolation points based on an equalization criterion.
  • the third processing module 303 is configured to determine a plurality of input sample values in each sub-interval equally. During specific implementation, the third processing module 303 may determine a plurality of input sample values in each sub-interval based on the equalization criterion.
  • the fourth processing module 304 is configured to traverse the input sample value combination of each input parameter of the model, and use the nonlinear model to obtain the output sample value combination corresponding to each input sample value combination.
  • the fifth processing module 305 is configured to generate a tensor table using all input sample value combinations and their corresponding output sample value combinations.
  • the linearization processing device of the non-linear model may further include a sixth processing module 306 as shown in the dotted line in FIG. 3, which is used to perform a simulation based on the current value of each input parameter.
  • the tensor table is interpolated to obtain the corresponding output value.
  • FIG. 4 is an exemplary structure diagram of another linearization processing device of a nonlinear model in an embodiment of the present invention. As shown in FIG. 4, on the basis of the device shown in FIG. 3, the device may further include: a first modeling module 307 and a second modeling module 308.
  • the first modeling module 307 is used to determine the complete design point data for each target nonlinear basis process of each device; the ratio of the similarity number supported by the similarity criterion to the similarity number based on the design point data , Establish the description formula of the non-linear basis process, and obtain the general model of the non-linear basis process; the general model includes variable parameters that change nonlinearly with the change of actual working condition parameters; construct the actual The machine learning algorithm between the working condition parameters and the variable parameters, and the association between the machine learning algorithm and the general model is established; the general model of the non-linear general process of all targets of each device and its The associated machine learning algorithms constitute a general model of this kind of equipment.
  • the second modeling module 308 is used to obtain the actual working condition parameters and variable parameters corresponding to the target nonlinear basic principle process of the specific equipment for each target non-linear basic principle process of a specific equipment of this kind of equipment Using the historical data to train the machine learning algorithm to obtain the variable parameter training model of the target nonlinear basic process; substituting the variable parameter training model of the target nonlinear basic process into In the general model of the target non-linear basis process, the trained model of the target non-linear basis process of the specific device is obtained; the post-training model of the target non-linear basis process of the specific device is constituted by Describe the post-training model of the specific equipment.
  • FIG. 5 is a schematic structural diagram of another linearization processing device of a nonlinear model in an embodiment of the present invention.
  • the system may include: at least one memory 51 and at least one processor 52.
  • some other components may also be included, such as communication ports. These components communicate via the bus 53.
  • At least one memory 51 is used to store a computer program.
  • the computer program can be understood as each module of the linearization processing device including the nonlinear model shown in FIG. 3 or FIG. 4.
  • at least one memory 51 may also store an operating system and the like.
  • Operating systems include but are not limited to: Android operating system, Symbian operating system, Windows operating system, Linux operating system and so on.
  • At least one processor 52 is configured to call a computer program stored in at least one memory 51 to execute the linearization processing method of the nonlinear model described in the embodiment of the present invention.
  • the processor 52 may be a CPU, a processing unit/module, an ASIC, a logic module or a programmable gate array, etc. It can receive and send data through the communication port.
  • a hardware module may include specially designed permanent circuits or logic devices (such as dedicated processors, such as FPGAs or ASICs) to complete specific operations.
  • the hardware module may also include programmable logic devices or circuits temporarily configured by software (for example, including general-purpose processors or other programmable processors) for performing specific operations.
  • software for example, including general-purpose processors or other programmable processors
  • the embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and the computer program can be executed by a processor and realize the linearization of the nonlinear model described in the embodiment of the present invention.
  • a system or device equipped with a storage medium may be provided, and the software program code for realizing the function of any one of the above-mentioned embodiments is stored on the storage medium, and the computer (or CPU or MPU of the system or device) ) Read and execute the program code stored in the storage medium.
  • an operating system or the like operating on the computer can also be used to complete part or all of the actual operations through instructions based on the program code.
  • Implementations of storage media used to provide program codes include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), Magnetic tape, non-volatile memory card and ROM.
  • the program code can be downloaded from the server computer via a communication network.
  • the piecewise linearization technology is used to process the non-linear model to obtain a tensor table.
  • an interpolation operation is performed based on the tensor table to obtain the required simulation data.
  • the linearized equipment model runs faster and can meet the real-time requirements of simulation.
  • the ratio of the similarity number supported by the similarity criterion to the similarity number based on the design point data is adopted for each target non-linear basis process of each equipment to establish the non-linear basis process
  • the formula is described, and a general model of the nonlinear basic process is obtained, so that the model can be applied to a class of equipment.
  • the general model includes variable parameters that change non-linearly with the actual operating condition parameters, and by constructing a machine learning algorithm between the actual operating condition parameters and the variable parameters, the variable parameters It can be obtained through machine learning, so that the passing model has self-learning ability.
  • the actual working condition parameters and the historical data of the variable parameters corresponding to the target non-linear basis process of the specific equipment are obtained.
  • the historical data use the historical data to train the machine learning algorithm to obtain the variable parameter training model of the target nonlinear basic process, and substitute the variable parameter training model of the target nonlinear basic process into the
  • a trained model of the target nonlinear basic process of the specific device can be obtained, that is, an instantiated model that meets the characteristics of the specific device.
  • the nonlinear model can be made available when the training variable parameters are not available on site, for example, when there is insufficient historical data.
  • modeling method in the embodiment of the present invention can be applied to various non-linear processes of various devices, which is not only convenient for implementation, but also has high accuracy.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Feedback Control In General (AREA)

Abstract

一种非线性模型的线性化处理方法、装置及存储介质。其中，方法包括：针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围（101）；将每个输入参数的取值范围基于复数个插值点划分为复数个子区间（102）；在每个子区间内均衡确定复数个输入样本值（103）；遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合（104）；利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表（105）。上述方法能够针对一些包括非线性物理过程的模型实现非线性模型的线性化处理。

Description

非线性模型的线性化处理方法、装置及存储介质 技术领域
本发明涉及工业领域，特别是一种综合能源系统中的非线性模型的线性化处理方法、装置以及计算机可读存储介质。
背景技术
分布式能源系统(DES)被认为是解决不稳定可再生能源消耗的有效途径之一。包括我国在内的世界各地都在建设DES，对运行优化所用模型和整体系统的能源生产及使用的安排的需求也在不断增加。以前的研究者已经开发出一些特定的操作优化模型，这些模型只适用于具有特定能源和组件的系统。因此，它们的模型在实际实现中并不能满足综合能源服务这一新事物的要求。
因此，有必要提供一种通用化的综合能源系统，以便充分利用可再生能源、化石燃料、余温余压、新能源等多种资源形式，使之相互配合，通过源、网、荷、储的灵活运行，建立创新的商业机制、采用智能的手段实现高质量、高效率、最经济及环境效应的区域电、热、冷、气等多种负荷的综合供给，满足终端负荷随机性变动要求。综合能源促进了可再生能源消纳能力，提高能源综合利用率。
但实现上述综合能源系统的过程中需要建立很多设备的模型，但这些设备通常包括很多非线性的物理过程(也称基理过程)，如燃气轮机的压气机中的流量与压力的相关过程，机械能转变为压能的过程等，因此这种设备的模型通常为非线性模型。非线性模型通常建模过程比较复杂，但相应的精度也比较高，但若直接采用该非线性模型运行综合能源系统的仿真，则可能会由于非线性模型的运行速度相对较慢而影响整个综合能源系统的仿真实时性。
发明内容
有鉴于此，本发明实施例中一方面提出了一种非线性模型的线性化处理方法，另一方面提出了一种非线性模型的线性化处理装置及计算机可读存储介质，用于针对一些包括非线性的基理过程的模型实现非线性模型的线性化处理，进而用于综合能源系统的构建。
本发明实施例中提出的一种非线性模型的线性化处理方法，包括：针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围；将每个输入参数的取值范围基于复数个插 值点划分为复数个子区间；在每个子区间内均衡确定复数个输入样本值；遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合；利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表。
在一个实施方式中，所述将每个输入参数的取值范围基于复数个插值点划分为复数个子区间为：基于均衡准则，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。
在一个实施方式中，所述在每个子区间内均衡确定复数个输入样本值为：基于均衡准则，在每个子区间内均衡确定复数个输入样本值。
在一个实施方式中，进行仿真时，根据各个输入参数的当前值查找所述张量表，并利用从所述张量表中查找到的对应数据进行插值处理，得到对应的输出值。
在一个实施方式中，每个设备的非线性模型采用如下方法建模得到：针对每种设备的每个目标非线性基理过程，确定其完整的设计点数据；采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型；所述通用模型中包括随实际工况参数变更而非线性变化的可变参数；构建所述实际工况参数与所述可变参数之间的机器学习算法，并建立所述机器学习算法与所述通用模型之间的关联关系；每种设备的所有目标非线性通用过程的通用模型及其所关联的机器学习算法构成该种设备的通用模型；针对该种设备的一个具体设备的每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型；将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，得到所述具体设备的所述目标非线性基理过程的训练后模型；所述具体设备的所有目标非线性基理过程的训练后模型构成所述具体设备的训练后模型。
在一个实施方式中，所述可变参数具有预先设置的默认值。
在一个实施方式中，所述设备包括：燃气轮机、热泵、内燃机、汽轮机、余热锅炉、吸收制冷机、制热机、多效蒸发器、电解水制氢、氢制化学品设备、反渗透、燃料电池、锅炉；所述每种设备的目标非线性基理过程包括下述过程中的一个或多个：传热过程、热能转换为动能的过程、管道阻力过程、流量与压力的相关过程、热能转机械能过程、电能转冷、热能过程、精馏过程、蒸发过程和过滤过程。
本发明实施例中提出的一种非线性模型的线性化处理装置，包括：第一处理模块，用于针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围；第二处理模块，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间；第三处理模块，用于在每个 子区间内均衡确定复数个输入样本值；第四处理模块，用于遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合；第五处理模块，用于利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表。
在一个实施方式中，所述第二处理模块基于均衡准则，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。
在一个实施方式中，所述第三处理模块基于均衡准则，在每个子区间内均衡确定复数个输入样本值。
在一个实施方式中，该系统进一步包括：第六处理模块，用于在进行仿真时，根据各个输入参数的当前值对所述张量表进行插值处理，得到对应的输出值。
在一个实施方式中，进一步包括：第一建模模块，用于针对每种设备的每个目标非线性基理过程，确定其完整的设计点数据；采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型；所述通用模型中包括随实际工况参数变更而非线性变化的可变参数；构建所述实际工况参数与所述可变参数之间的机器学习算法，并建立所述机器学习算法与所述通用模型之间的关联关系；每种设备的所有目标非线性通用过程的通用模型及其所关联的机器学习算法构成该种设备的通用模型；和第二建模模块，用于针对该种设备的一个具体设备的每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型；将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，得到所述具体设备的所述目标非线性基理过程的训练后模型；所述具体设备的所有目标非线性基理过程的训练后模型构成所述具体设备的训练后模型。
本发明实施例中提出的又一种非线性模型的线性化处理装置，包括：至少一个存储器和至少一个处理器，其中：所述至少一个存储器用于存储计算机程序；所述至少一个处理器用于调用所述至少一个存储器中存储的计算机程序，执行如上任一实施方式所述的非线性模型的线性化处理方法。
本发明实施例中提出的计算机可读存储介质，其上存储有计算机程序；所述计算机程序能够被一处理器执行并实现如上任一实施方式所述的线性化处理方法。
从上述方案中可以看出，由于本发明实施例中采用分段线性化技术对所述非线性化模型进行处理，得到一个张量表。并在仿真应用时，基于所述张量表进行插值运算，得到所需的仿真数据。经线性化处理后的设备模型运行速度较快，能够满足仿真的实时性 要求。
此外，对设备模型进行建模时，通过针对每种设备的每个目标非线性基理过程，采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型，使得该模型可以作为通用模型而适用于一类设备。此外，所述通用模型中包括随实际工况参数变更而非线性变化的可变参数，并且通过构建所述实际工况参数与所述可变参数之间的机器学习算法，使得该可变参数可通过机器学习获得，进而使得该通用模型具有自学习能力。
此外，针对该种设备的一个具体设备，通过对其每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，并利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型，并将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，可得到所述具体设备的所述目标非线性基理过程的训练后模型，也即符合具体设备特性的实例化模型。
并且，通过为可变参数预先设置默认值，可在现场不具备训练可变参数的情况下，例如没有足够的历史数据等情况下，可以使得该非线性模型可用。
最后，本发明实施例中的建模方法可应用于各种设备的各种非线性过程，不仅方便实现，而且精度较高。
附图说明
下面将通过参照附图详细描述本发明的优选实施例，使本领域的普通技术人员更清楚本发明的上述及其它特征和优点，附图中：
图1为本发明实施例中一种非线性模型的线性化处理方法的示例性流程图。
图2为本发明实施例中一种非线性模型的建模方法的示例性流程图。
图3为本发明实施例中一种非线性模型的线性化处理装置的示例性结构图。
图4为本发明实施例中又一种非线性模型的线性化处理装置的示例性结构图。
图5为本发明实施例中又一种非线性模型的线性化处理装置的示例性结构图。
其中，附图标记如下：
标号 含义
101-105，201-207 步骤
301 第一处理模块
302 第二处理模块
303 第三处理模块
304 第四处理模块
305 第五处理模块
306 第六处理模块
307 第一建模模块
308 第二建模模块
51 存储器
52 处理器
53 总线
具体实施方式
为使本发明的目的、技术方案和优点更加清楚，以下举实施例对本发明进一步详细说明。
图1为本发明实施例中一种非线性模型的线性化处理方法的示例性流程图。如图1所示，该方法可包括如下步骤：
步骤101，针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围。
例如，有效的功率范围为额定工况的50％到110％的范围，此外，还有当地的环境温度变化的范围，环境压力变化的范围等。
步骤102，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。
本步骤中，可基于均衡准则，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。其中，插值点的确定可以是针对非线性变化剧烈的区域设置较多的插值点，针对非线性变化缓慢的区域设置较少的插值点。
例如，功率范围插入40个点，环境温度插入20个点，环境压力插入5个点等。
步骤103，在每个子区间内均衡确定复数个输入样本值。
本步骤中，可基于均衡准则，在每个子区间内均衡确定复数个输入样本值。
例如，功率及环境温度等在实际域中可采用等分方法。
步骤104，遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合。
例如，对应遍历得到的每组输入样本值，均存在一组对应的输出，例如效率输出，或燃料消耗、排放输出，运行成本输出等。
步骤105，利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表。
例如，如上述已知温度、压力、功率三维度的数值，通过查取张量表，可插值可得到效率的数值。
具体，利用该设备的模型进行仿真时，可根据各个输入参数的当前值查找所述张量表，并利用从所述张量表中查找到的对应数值进行插值处理，得到对应的输出值。其中，各个输入参数的当前值可以是真实值或者也可以是假设值。
例如，对于一个设备可以有一张或多张表，例如温度、压力、功率对应于效率的一张表，温度、压力、功率对应于排放的表，或对应于其它任何所需参数的表。其中，插值算法可根据实际情况选用，例如，可以选用线性插值法或非线性插值法等。在一个示例中，可对相临较近的点采用线性插值法，对相距较远的点采用非线性插值法等。
通过三维温度、压力、功率的样条插值等方法可得到需要的温度、压力、功绩对应的其它输出变量的数值。
这种通用方法使用通用程序，无论是任何一种具体模型，例如，热泵、内燃机，换热器等的模型都可通过这一段代码工具来进行处理。
图2为本发明实施例中一种非线性模型的建模方法的示例性流程图。如图2所示，该方法可包括如下步骤：
步骤201，针对每种设备的每个目标非线性基理过程，确定其完整的设计点数据。
本步骤中，基理过程有时也可称为物理过程，如传热过程、电能转换过程、以及前面提到的流量与压力的相关过程等。针对每种设备，可以确定感兴趣的基理过程，即需要建模的基理过程，将这些需要建模的基理过程称为目标基理过程，其中非线性的目标基理过程便可称为目标非线性基理过程。例如，针对燃气轮机，其目标非线性基理过程可包括：膨胀透平中的流量与压力的相关过程、热能与机械能能量转换的过程等；针对热泵，其目标非线性基理过程可包括：传热过程、电能转换为热能的过程、使用高温热能对溶液物质分离的化学过程、电化学过程、管道阻力过程、流量与压力的相关过程等。此外，针对内燃机、汽轮机、 余热锅炉、吸收制冷机、制热机、多效蒸发器、电解水制氢、氢制化学品设备、反渗透、燃料电池、锅炉等设备，每个设备的目标非线性基理过程可包括：流量与压力的相关过程、热能转机械能过程，电能转冷、热能过程、管道阻力过程、传热过程，精馏过程，蒸发过程，过滤过程，化学反应过程、电化学过程等中的一个或多个。
针对每个非线性的基理过程，可根据基本公开发表的设计参数及厂家提供给用户的普通设计点信息等，复原完整的设计点数据。例如，针对流量与压力的相关过程这一通用模型，其设计点数据可包括压比、空气流量等，根据这些设计点数据可推导其效率、进口阻力以及抽气量等使用者不可得的相关设计参数。
步骤202，采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型；所述通用模型中包括随实际工况参数变更而非线性变化的可变参数。
其中，实际工况参数指的是某一具体设备的随实际工况参数变化而变化的参数。例如，随着长时间的使用发生机械磨损而产生变化的尺寸，或随季节变化而变化的温度，或随不同做功情况变化的相关参数等。所述可变参数可具有预先设置的默认值。
由于针对每种设备可能存在不同的型号，例如，以压缩机为例，可能存在功率为5M、50M、500M等不同功率的压缩机，因此为了建立压缩机的通用模型，需要采用相似性准则支持的相似数来代替具体的参数值。例如，仍以上述的流量与压力的相关过程这一通用模型为例，使用流量及压力、功率的相似准则支持的相似数代替具体参数，例如：流量的相似准则可如下式(1)所示：
Figure PCTCN2019109675-appb-000001
其中，G1为流量，T1为温度，P1为压力，G0为相应设计点流量，T0为相应设计点温度，P0为相应设计点压力。
相应地，得到的流量与压力的相关过程的通用模型可如下式(2)所示：
Figure PCTCN2019109675-appb-000002
其中，f()为函数，系数a和b为随实际工况参数变更而非线性变化的可变参数，实际应 用中，也可为可变参数a和b设置一个默认值。IGV为进口可转导叶角度。
步骤203，构建所述实际工况参数与所述可变参数之间的机器学习算法。
本步骤中，可基于神经智能网络或支持向量机等机器学习大数据分析的方法来构建所述实际工况参数与所述可变参数之间的机器学习算法。
步骤204，每种设备的所有目标非线性基理过程的通用模型及其所关联的机器学习算法构成该种设备的通用模型。
可见，通过上述过程可以建立每种设备的一个非线性的通用模型。基于这些通用模型便可以构建综合能源系统平台。
实际应用中，用户购买该综合能源系统平台之后，需要搭建自己的综合能源系统，此时，每个通用模型需要与现场的具体设备相关联，因此便需要对该通用模型进行实例化。相应地，该方法可进一步包括如下步骤：
步骤205，针对该种设备的一个具体设备的每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，利用所述历史数据对所对应的机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型。
本步骤中，具体训练时，将实际工况参数的一组历史数据作为输入样本值，将实际工况参数的所述一组历史数据对应的可变参数的历史数据作为输出样本值，利用大量的输入样本值和对应的输出样本值对所述机器学习算法进行训练，便可得到所述可变参数的自学习模型，也称训练模型。
例如，仍以上述的流量与压力的相关过程为例，则可获取现场的燃气轮机的相关实际工况参数的历史数据及其对应的可变参数的历史数据，得到输入输出样本集，通过训练后可得到可变参数a和b的训练模型。
步骤206，将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，得到所述具体设备的所述目标非线性基理过程的训练后模型。该训练后模型为具有学习能力的自学习模型。
本步骤中，可根据所述机器学习算法与所述通用模型的关联关系，将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中。
例如，仍以上述的流量与压力的相关过程为例，将可变参数a和b的当前训练模型输入上述式(2)中，便可得到现场的燃气轮机的压气机的流量与压力的相关过程的通用模型。
步骤207，所述具体设备的所有目标非线性基理过程的训练后模型构成所述具体设备的 训练后模型。
实际使用时，该训练后模型的输入参数可包括所有目标非线性基理过程的训练后模型所需的输入参数。
以上对本发明实施例中的非线性模型的线性化方法以及其中的一种建模方法进行了详细描述，下面再对本发明实施例中的非线性模型的线性化处理装置以及其中的一种建模装置进行详细描述。本发明实施例中的装置可用于实现本发明实施例中的方法。对于本发明装置实施例中未披露的细节，可参见本发明方法实施例中的相应描述，此处不再一一赘述。
图3为本发明实施例中一种非线性模型的线性化处理装置的示例性结构图。如图3所示，该装置可包括：第一处理模块301、第二处理模块302、第三处理模块303、第四处理模块304和第五处理模块305。
第一处理模块301用于针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围。
第二处理模块302将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。具体实现时，第二处理模块302可基于均衡准则，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。
第三处理模块303用于在每个子区间内均衡确定复数个输入样本值。具体实现时，第三处理模块303可基于均衡准则，在每个子区间内均衡确定复数个输入样本值。
第四处理模块304用于遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合。
第五处理模块305用于利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表。
在其他实施方式中，该非线性模型的线性化处理装置可如图3中的虚线部分所示进一步包括：第六处理模块306，用于在进行仿真时，根据各个输入参数的当前值对所述张量表进行插值处理，得到对应的输出值。
图4为本发明实施例中又一种非线性模型的线性化处理装置的示例性结构图。如图4所示，该装置可在图3所示装置的基础上，进一步包括：第一建模模块307和第二建模模块308。
其中，第一建模模块307用于针对每种设备的每个目标非线性基理过程，确定其完整的设计点数据；采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非 线性基理过程的描述公式，得到所述非线性基理过程的通用模型；所述通用模型中包括随实际工况参数变更而非线性变化的可变参数；构建所述实际工况参数与所述可变参数之间的机器学习算法，并建立所述机器学习算法与所述通用模型之间的关联关系；每种设备的所有目标非线性通用过程的通用模型及其所关联的机器学习算法构成该种设备的通用模型。
第二建模模块308用于针对该种设备的一个具体设备的每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型；将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，得到所述具体设备的所述目标非线性基理过程的训练后模型；所述具体设备的所有目标非线性基理过程的训练后模型构成所述具体设备的训练后模型。
图5为本发明实施例中又一种非线性模型的线性化处理装置的结构示意图，如图5所示，该系统可包括：至少一个存储器51和至少一个处理器52。此外，还可以包括一些其它组件，例如通信端口等。这些组件通过总线53进行通信。
其中：至少一个存储器51用于存储计算机程序。在一个实施方式中，该计算机程序可以理解为包括图3或图4所示的非线性模型的线性化处理装置的各个模块。此外，至少一个存储器51还可存储操作系统等。操作系统包括但不限于：Android操作系统、Symbian操作系统、Windows操作系统、Linux操作系统等等。
至少一个处理器52用于调用至少一个存储器51中存储的计算机程序，执行本发明实施例中所述的非线性模型的线性化处理方法。处理器52可以为CPU，处理单元/模块，ASIC，逻辑模块或可编程门阵列等。其可通过所述通信端口进行数据的接收和发送。
需要说明的是，上述各流程和各结构图中不是所有的步骤和模块都是必须的，可以根据实际的需要忽略某些步骤或模块。各步骤的执行顺序不是固定的，可以根据需要进行调整。各模块的划分仅仅是为了便于描述采用的功能上的划分，实际实现时，一个模块可以分由多个模块实现，多个模块的功能也可以由同一个模块实现，这些模块可以位于同一个设备中，也可以位于不同的设备中。
可以理解，上述各实施方式中的硬件模块可以以机械方式或电子方式实现。例如，一个硬件模块可以包括专门设计的永久性电路或逻辑器件(如专用处理器，如FPGA或ASIC)用于完成特定的操作。硬件模块也可以包括由软件临时配置的可编程逻辑器件或电路(如包括通用处理器或其它可编程处理器)用于执行特定操作。至于具体采用机械方式，或是采用专用的永久性电路，或是采用临时配置的电路(如由软件进行配置)来实现硬件模块，可以根 据成本和时间上的考虑来决定。
此外，本发明实施例中还提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序能够被一处理器执行并实现本发明实施例中所述的非线性模型的线性化处理方法。具体地，可以提供配有存储介质的系统或者装置，在该存储介质上存储着实现上述实施例中任一实施方式的功能的软件程序代码，且使该系统或者装置的计算机(或CPU或MPU)读出并执行存储在存储介质中的程序代码。此外，还可以通过基于程序代码的指令使计算机上操作的操作系统等来完成部分或者全部的实际操作。还可以将从存储介质读出的程序代码写到插入计算机内的扩展板中所设置的存储器中或者写到与计算机相连接的扩展单元中设置的存储器中，随后基于程序代码的指令使安装在扩展板或者扩展单元上的CPU等来执行部分和全部实际操作，从而实现上述实施方式中任一实施方式的功能。用于提供程序代码的存储介质实施方式包括软盘、硬盘、磁光盘、光盘(如CD-ROM、CD-R、CD-RW、DVD-ROM、DVD-RAM、DVD-RW、DVD+RW)、磁带、非易失性存储卡和ROM。可选择地，可以由通信网络从服务器计算机上下载程序代码。
从上述方案中可以看出，由于本发明实施例中采用分段线性化技术对所述非线性化模型进行处理，得到一个张量表。并在仿真应用时，基于所述张量表进行插值运算，得到所需的仿真数据。经线性化处理后的设备模型运行速度较快，能够满足仿真的实时性要求。
对设备进行建模时，通过针对每种设备的每个目标非线性基理过程，采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型，使得该模型可适用于一类设备。此外，所述通用模型中包括随实际工况参数变更而非线性变化的可变参数，并且通过构建所述实际工况参数与所述可变参数之间的机器学习算法，使得该可变参数可通过机器学习获得，进而使得该通过模型具有自学习能力。
此外，针对该种设备的一个具体设备，通过对其每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，并利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型，并将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，可得到所述具体设备的所述目标非线性基理过程的训练后模型，也即符合具体设备特性的实例化模型。
并且，通过为可变参数预先设置默认值，可在现场不具备训练可变参数的情况下， 例如没有足够的历史数据等情况下，可以使得该非线性模型可用。
最后，本发明实施例中的建模方法可应用于各种设备的各种非线性过程，不仅方便实现，而且精度较高。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (14)

  1. 一种非线性模型的线性化处理方法，其特征在于，包括：
    针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围(101)；
    将每个输入参数的取值范围基于复数个插值点划分为复数个子区间(102)；
    在每个子区间内均衡确定复数个输入样本值(103)；
    遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合(104)；和
    利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表(105)。
  2. 根据权利要求1所述的非线性模型的线性化处理方法，其特征在于，所述将每个输入参数的取值范围基于复数个插值点划分为复数个子区间为：基于均衡准则，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。
  3. 根据权利要求1所述的非线性模型的线性化处理方法，其特征在于，所述在每个子区间内均衡确定复数个输入样本值为：基于均衡准则，在每个子区间内均衡确定复数个输入样本值。
  4. 根据权利要求1至3中任一项所述的非线性模型的线性化处理方法，其特征在于，
    进行仿真时，根据各个输入参数的当前值查找所述张量表，并利用从所述张量表中查找到的对应数据进行插值处理，得到对应的输出值。
  5. 根据权利要求1至3中任一项所述的非线性模型的线性化处理方法，其特征在于，每个设备的非线性模型采用如下方法建模得到：
    针对每种设备的每个目标非线性基理过程，确定其完整的设计点数据(201)；
    采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型；所述通用模型中包括随实际工况参数变更而非线性变化的可变参数(202)；
    构建所述实际工况参数与所述可变参数之间的机器学习算法，并建立所述机器学习算法与所述通用模型之间的关联关系(203)；
    每种设备的所有目标非线性通用过程的通用模型及其所关联的机器学习算法构成该种设备的通用模型(204)；
    针对该种设备的一个具体设备的每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型(205)；和
    将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，得到所述具体设备的所述目标非线性基理过程的训练后模型(206)；
    所述具体设备的所有目标非线性基理过程的训练后模型构成所述具体设备的训练后模型(207)。
  6. 根据权利要求5所述的非线性模型的建模方法，其特征在于，所述可变参数具有预先设置的默认值。
  7. 根据权利要求5或6所述的非线性模型的建模方法，其特征在于，所述设备包括：燃气轮机、热泵、内燃机、汽轮机、余热锅炉、吸收制冷机、制热机、多效蒸发器、电解水制氢、氢制化学品设备、反渗透、燃料电池、锅炉：
    所述每种设备的目标非线性基理过程包括下述过程中的一个或多个：传热过程、热能转换为动能的过程、管道阻力过程、流量与压力的相关过程、热能转机械能过程、电能转冷、热能过程、精馏过程、蒸发过程和过滤过程。
  8. 一种非线性模型的线性化处理装置，其特征在于，包括：
    第一处理模块(301)，用于针对每个设备的非线性模型，确定所述模型各个输入参数的取值范围；
    第二处理模块(302)，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间；
    第三处理模块(303)，用于在每个子区间内均衡确定复数个输入样本值；
    第四处理模块(304)，用于遍历所述模型各个输入参数的输入样本值组合，利用所述非线性模型得到每个输入样本值组合对应的输出样本值组合；和
    第五处理模块(305)，用于利用所有输入样本值组合与其所对应的输出样本值组合生成一个张量表。
  9. 根据权利要求8所述的非线性模型的线性化处理装置，其特征在于，所述第二处理模块基于均衡准则，将每个输入参数的取值范围基于复数个插值点划分为复数个子区间。
  10. 根据权利要求8所述的非线性模型的线性化处理装置，其特征在于，所述第三处理模块基于均衡准则，在每个子区间内均衡确定复数个输入样本值。
  11. 根据权利要求8至10中任一项所述的非线性模型的线性化处理装置，其特征在于，进一步包括：
    第六处理模块(306)，用于在进行仿真时，根据各个输入参数的当前值对所述张量表进 行插值处理，得到对应的输出值。
  12. 根据权利要求8至10中任一项所述的非线性模型的线性化处理装置，其特征在于，进一步包括：第一建模模块(307)，用于针对每种设备的每个目标非线性基理过程，确定其完整的设计点数据；采用相似性准则支持的相似数与基于设计点数据的相似数的比值，建立所述非线性基理过程的描述公式，得到所述非线性基理过程的通用模型；所述通用模型中包括随实际工况参数变更而非线性变化的可变参数；构建所述实际工况参数与所述可变参数之间的机器学习算法，并建立所述机器学习算法与所述通用模型之间的关联关系；每种设备的所有目标非线性通用过程的通用模型及其所关联的机器学习算法构成该种设备的通用模型；和
    第二建模模块(308)，用于针对该种设备的一个具体设备的每个目标非线性基理过程，获取所述具体设备的所述目标非线性基理过程对应的实际工况参数和可变参数的历史数据，利用所述历史数据对所述机器学习算法进行训练，得到所述目标非线性基理过程的可变参数训练模型；将所述目标非线性基理过程的可变参数训练模型代入所述目标非线性基理过程的通用模型中，得到所述具体设备的所述目标非线性基理过程的训练后模型；所述具体设备的所有目标非线性基理过程的训练后模型构成所述具体设备的训练后模型。
  13. 一种非线性模型的线性化处理装置，其特征在于，包括：至少一个存储器(51)和至少一个处理器(52)，其中：
    所述至少一个存储器(51)用于存储计算机程序；
    所述至少一个处理器(52)用于调用所述至少一个存储器(51)中存储的计算机程序，执行如权利要求1至7中任一项所述的非线性模型的线性化处理方法。
  14. 计算机可读存储介质，其上存储有计算机程序；其特征在于，所述计算机程序能够被一处理器执行并实现如权利要求1至7中任一项所述的非线性模型的线性化处理方法。
PCT/CN2019/109675 2019-09-30 2019-09-30 非线性模型的线性化处理方法、装置及存储介质 WO2021062755A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
EP19947704.3A EP4024259A4 (en) 2019-09-30 2019-09-30 LINEARIZATION PROCESSING METHOD AND APPARATUS FOR A NONLINEAR MODEL AND STORAGE MEDIA
US17/765,109 US20220350947A1 (en) 2019-09-30 2019-09-30 Linearization Processing Method and Device for Nonlinear Model, And Storage Medium
PCT/CN2019/109675 WO2021062755A1 (zh) 2019-09-30 2019-09-30 非线性模型的线性化处理方法、装置及存储介质
CN201980100540.8A CN114424196A (zh) 2019-09-30 2019-09-30 非线性模型的线性化处理方法、装置及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/109675 WO2021062755A1 (zh) 2019-09-30 2019-09-30 非线性模型的线性化处理方法、装置及存储介质

Publications (1)

Publication Number Publication Date
WO2021062755A1 true WO2021062755A1 (zh) 2021-04-08

Family

ID=75336725

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/109675 WO2021062755A1 (zh) 2019-09-30 2019-09-30 非线性模型的线性化处理方法、装置及存储介质

Country Status (4)

Country Link
US (1) US20220350947A1 (zh)
EP (1) EP4024259A4 (zh)
CN (1) CN114424196A (zh)
WO (1) WO2021062755A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118504852A (zh) * 2023-02-15 2024-08-16 华为技术有限公司 一种方案规划方法、装置及相关设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104133948A (zh) * 2014-07-16 2014-11-05 潍柴动力股份有限公司 一种仿真模型生成方法、系统及仿真模型
US20160034421A1 (en) * 2014-08-01 2016-02-04 Infineon Technologies Ag Digital pre-distortion and post-distortion based on segmentwise piecewise polynomial approximation
CN107389116A (zh) * 2017-07-29 2017-11-24 北京广利核系统工程有限公司 模拟电流变送器信号的自动化测试装置和方法
CN109614571A (zh) * 2018-10-09 2019-04-12 阿里巴巴集团控股有限公司 非线性规划问题的处理方法及装置
CN109976156A (zh) * 2019-03-13 2019-07-05 南京航空航天大学 固定翼无人机栖落机动轨迹的建模与预测控制方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109902854B (zh) * 2019-01-11 2020-11-27 重庆大学 电-气互联系统最优潮流全线性模型的构建方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104133948A (zh) * 2014-07-16 2014-11-05 潍柴动力股份有限公司 一种仿真模型生成方法、系统及仿真模型
US20160034421A1 (en) * 2014-08-01 2016-02-04 Infineon Technologies Ag Digital pre-distortion and post-distortion based on segmentwise piecewise polynomial approximation
CN107389116A (zh) * 2017-07-29 2017-11-24 北京广利核系统工程有限公司 模拟电流变送器信号的自动化测试装置和方法
CN109614571A (zh) * 2018-10-09 2019-04-12 阿里巴巴集团控股有限公司 非线性规划问题的处理方法及装置
CN109976156A (zh) * 2019-03-13 2019-07-05 南京航空航天大学 固定翼无人机栖落机动轨迹的建模与预测控制方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4024259A4 *

Also Published As

Publication number Publication date
EP4024259A4 (en) 2023-05-10
CN114424196A (zh) 2022-04-29
US20220350947A1 (en) 2022-11-03
EP4024259A1 (en) 2022-07-06

Similar Documents

Publication Publication Date Title
WO2021062748A1 (zh) 综合能源系统的优化方法、装置和计算机可读存储介质
WO2021062753A1 (zh) 综合能源系统的仿真方法、装置和计算机可读存储介质
Hipólito-Valencia et al. Optimal design of inter-plant waste energy integration
Miar Naeimi et al. Energy, exergy, exergoeconomic and exergoenvironmental analysis and optimization of a solar hybrid CCHP system
CN103917928A (zh) 联合冷却加热发电厂的热-经济学建模和优化
Zhao et al. Multi-objective optimization of a hybrid system based on combined heat and compressed air energy storage and electrical boiler for wind power penetration and heat-power decoupling purposes
CN112671028B (zh) 一种考虑热网动态特性的综合能源系统风电消纳方法
CN112560329B (zh) 不确定性下的工业装置能量系统数据驱动鲁棒优化方法
CN114450682A (zh) 综合能源系统的控制方法、装置和计算机可读存储介质
Xia et al. Method for integrating low-grade industrial waste heat into district heating network
Möhren et al. A multiperiod approach for waste heat and renewable energy integration of industrial sites
WO2021062755A1 (zh) 非线性模型的线性化处理方法、装置及存储介质
Singh et al. Exergoeconomic analysis of a Kalina cycle coupled coal-fired steam power plant
CN117852222B (zh) 一种集成有机朗肯循环的换热网络优化方法
Jahnke et al. First cycle simulations of the Honigmann process with LiBr/H2O and NaOH/H2O as working fluid pairs as a thermochemical energy storage
CN105184087A (zh) 环境温度变化对燃煤发电机组煤耗影响的计算方法及装置
WO2021062754A1 (zh) 非线性模型的建模方法、装置及存储介质
Wang et al. Modelling of the whole process of a university campus CHP power plant and dynamic performance study
Ma et al. Energy saving maximization on combined heat and power units in different scenarios
CN103983841B (zh) 热电联产机组的背压变化对功率影响的确定方法
Ghazi et al. Developing of constructal theory concept to the total site cogeneration heat and power retrofit
Zhu et al. Operation optimization of cold‐end system of direct air‐cooled units considering the effect of environmental wind
McTigue et al. Integrated Heat Pump Thermal Storage and Power Cycle for CSP (Final Technical Report)
Kecebas Thermodynamic Optimization of Turbine Lines for Maximum Exergy Efficiency in a Binary Geothermal Power Plant
CN115828038A (zh) 一种基于数据驱动的系统运行效率估算方法、装置及介质

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: 19947704

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2019947704

Country of ref document: EP

Effective date: 20220330