WO2024040608A1 - Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage - Google Patents

Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage Download PDF

Info

Publication number
WO2024040608A1
WO2024040608A1 PCT/CN2022/115274 CN2022115274W WO2024040608A1 WO 2024040608 A1 WO2024040608 A1 WO 2024040608A1 CN 2022115274 W CN2022115274 W CN 2022115274W WO 2024040608 A1 WO2024040608 A1 WO 2024040608A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
data
management system
energy management
neural network
Prior art date
Application number
PCT/CN2022/115274
Other languages
English (en)
Chinese (zh)
Inventor
王德慧
张拓
江宁
王刚
王丹
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 PCT/CN2022/115274 priority Critical patent/WO2024040608A1/fr
Publication of WO2024040608A1 publication Critical patent/WO2024040608A1/fr

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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks

Definitions

  • the invention relates to the technical field of energy management systems, in particular to model training methods, devices and storage media of energy management systems.
  • the traditional centralized energy supply system uses large-capacity equipment and centralized production methods to transport various energies to many users in a large range through specialized transmission facilities (large power grids, large heating networks, etc.).
  • specialized transmission facilities large power grids, large heating networks, etc.
  • distributed power sources such as photovoltaics, wind power, and natural gas combined heat and power
  • users have put forward further requirements for the economy, reliability, and flexibility of energy systems.
  • industrial technology is moving from electrification, automation, and digitalization. Digitalization in the energy sector is an important part of industrial digitalization.
  • the energy management system takes cold and heat balance as the core, integrating geothermal energy, solar energy, air energy, water energy, natural gas, city tap water, sewage, industrial wastewater waste heat and other renewable energy sources, using cold and heat recovery, energy storage, heat balance, Intelligent control and other technologies carry out intelligent balance control of various energy flows to achieve reciprocal utilization of energy, thereby integrating refrigeration and heating, hot water, refrigeration and freezing, drying and heating, breeding and planting, snow removal and ice removal, steam, power generation, etc. a required function.
  • Artificial intelligence is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.
  • the embodiment of the present invention provides a model training method, device and storage medium for an energy management system.
  • a model training method for an energy management system includes:
  • the mechanism model of the energy management system includes mechanism models of subsystems and connection relationships between the mechanism models of the subsystems;
  • an artificial neural network model is trained as an AI model of the energy management system.
  • the embodiment of the present invention uses the simulation data of the energy management system to enrich the training data of the AI model, because the simulation data already reflects the mechanism relationship of the mechanism model and includes the intrinsic physical connection relationship of the actual service object and the physical hierarchy of the subsystem. , therefore, the embodiment of the present invention realizes the combination of mechanism and AI, introduces AI capabilities, enriches training data, and improves the performance of the AI model.
  • the AI model includes at least one of the following:
  • Performance monitoring model of subsystem Performance monitoring model of subsystem; performance prediction model of subsystem; performance diagnosis model of subsystem; optimized operation model of subsystem; performance monitoring model of energy management system; performance prediction model of energy management system; performance diagnosis of energy management system Model; optimal operation model of energy management system.
  • the AI model of the embodiment of the present invention can have multiple implementation modes, which improves the applicability.
  • training the artificial neural network model as the AI model of the energy management system based on the training data including the measurement data and the simulation data includes:
  • the training data is input into the artificial neural network model, so that the artificial neural network model outputs a predicted value corresponding to the simulation data; based on the difference between the measured data in the training data and the predicted value, Determine the loss function value of the artificial neural network model; configure the model parameters of the artificial neural network model so that the loss function value is lower than a preset threshold; determine the configured artificial neural network model as the Describe the AI model.
  • the accuracy of model training is improved by using simulation data with abundant data and diverse types as input to generate predicted values, and using measurement data with reliable results as actual values for comparison with measured values.
  • training the artificial neural network model as the AI model of the energy management system based on the training data including the measurement data and the simulation data includes:
  • the training data is input into the artificial neural network model, so that the artificial neural network model outputs a predicted value corresponding to the measured data; based on the difference between the simulation data in the training data and the predicted value, Determine the loss function value of the artificial neural network model; configure the model parameters of the artificial neural network model so that the loss function value is lower than a preset threshold; determine the configured artificial neural network model as the Describe the AI model.
  • constructing training data based on the measurement data and the simulation data includes:
  • process connection relationship includes at least one of the following: electrical connection; energy transfer connection; Liquid flow connection; gas flow connection; information transmission connection; value connection;
  • Training data is constructed that includes measurement data of the target device, measurement data of the process-related device, simulation data of the target device, and simulation data of the process-related device.
  • the process-related equipment is determined based on the process connection relationship, which enriches the training data, and the training data contains the process-related logic of the mechanism, which improves the model accuracy of the target equipment.
  • a model training device for an energy management system including:
  • the establishment module is configured to establish a mechanism model of the energy management system, where the mechanism model of the energy management system includes a mechanism model of a subsystem and a connection relationship between the mechanism models of the subsystems;
  • An input module configured to input measurement data into the mechanism model of the energy management system, wherein the measurement data is measured at a measurement point of the energy management system;
  • a simulation module configured to simulate and run the mechanism model of the energy management system based on the measurement data to obtain simulation data
  • a building module configured to construct training data based on the measurement data and the simulation data
  • a training module configured to train an artificial neural network model as an AI model of the energy management system based on the training data.
  • the embodiment of the present invention uses the simulation data of the energy management system to enrich the training data of the AI model, because the simulation data already reflects the mechanism relationship of the mechanism model and includes the intrinsic physical connection relationship of the actual service object and the physical hierarchy of the subsystem. , therefore, the embodiment of the present invention realizes the combination of mechanism and AI, introduces AI capabilities, enriches training data, and improves the performance of the AI model.
  • the AI model includes at least one of the following:
  • Performance monitoring model of subsystem Performance monitoring model of subsystem; performance prediction model of subsystem; performance diagnosis model of subsystem; optimized operation model of subsystem; performance monitoring model of energy management system; performance prediction model of energy management system; performance diagnosis of energy management system Model; optimal operation model of energy management system.
  • the AI model of the embodiment of the present invention can have multiple implementation modes, which improves the applicability.
  • the training module is configured to input the training data into the artificial neural network model, so that the artificial neural network model outputs a predicted value corresponding to the simulation data; based on the training data The difference between the measured data and the predicted value in determines the loss function value of the artificial neural network model; configures the model parameters of the artificial neural network model so that the loss function value is lower than the preset threshold ; Determine the configured artificial neural network model as the AI model.
  • the accuracy of model training is improved by using simulation data with abundant data and diverse types as input to generate predicted values, and using measurement data with reliable results as actual values for comparison with measured values.
  • the training module is configured to input the training data into the artificial neural network model, so that the artificial neural network model outputs a predicted value corresponding to the measurement data; based on the training data The difference between the simulation data in and the predicted value is used to determine the loss function value of the artificial neural network model; configure the model parameters of the artificial neural network model so that the loss function value is lower than the preset threshold ; Determine the configured artificial neural network model as the AI model.
  • the building module is configured to determine the target device of the AI model; based on the process connection relationship of the target device in the mechanism model of the energy management system, determine the target device's Process-related equipment, wherein the process connection relationship includes at least one of the following: electrical connection; energy transfer connection; liquid flow connection; gas flow connection; information transfer connection; value connection; extracting the target device from the measurement data
  • the measurement data and the measurement data of the process-related equipment extract the simulation data of the target equipment and the simulation data of the process-related equipment from the simulation data; construct the measurement data of the target equipment, the process Measurement data of relevant equipment, simulation data of the target equipment, and training data of simulation data of the process-related equipment.
  • the process-related equipment is determined based on the process connection relationship, which enriches the training data, and the training data contains the process-related logic of the mechanism, which improves the model accuracy of the target equipment.
  • An electronic device including:
  • Memory for storing executable instructions for the processor
  • the processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the model training method of the energy management system as described in any one of the above.
  • a computer-readable storage medium has computer instructions stored thereon. When the computer instructions are executed by a processor, the model training method of the energy management system as described in any one of the above items is implemented.
  • a computer program product includes a computer program that, when executed by a processor, implements the model training method for an energy management system as described in any one of the above items.
  • Figure 1 is a flow chart of a model training method of an energy management system according to an embodiment of the present invention.
  • Figure 2 is a schematic diagram of selecting a mechanism model of a subsystem and establishing a connection relationship in a drag-and-drop manner according to the embodiment of the present invention.
  • FIG. 3 is an exemplary process diagram of model training of the energy management system according to the embodiment of the present invention.
  • FIG. 4 is a schematic diagram of the processing framework of the energy management system according to the embodiment of the present invention.
  • FIG. 5 is an exemplary structural diagram of a model training device of an energy management system according to an embodiment of the present invention.
  • FIG. 6 is an exemplary structural diagram of an electronic device according to an embodiment of the present invention.
  • Model training method for energy management systems 101 ⁇ 105 step 200
  • Mechanism model of energy management system 201
  • Mechanism model of the first subsystem 202
  • Mechanism model of the second subsystem 203
  • Mechanism model of the third subsystem 204
  • Mechanism model of the fourth subsystem 11
  • Measurement data twenty one
  • Simulation data 30
  • training data 31
  • AI model for monitoring/prediction/diagnosis 51
  • Optimize running AI models 52
  • Verification process 54
  • Training process Verification process 57 Online usage process 60
  • Mechanism model establishment process 61
  • Simulation processing 500 Model training device for energy management systems 501 Build module 502 input module 503
  • Simulation module 504 building blocks 505 training module 600
  • Electronic equipment 601 processor 602 memory
  • Energy management system is a necessary operation and management tool for various energy systems. Whether it is a regional energy management system such as an industrial park or a field-level energy management system such as a group, factory, building, or microgrid, it is necessary to provide digital services for prediction, diagnosis, and intelligent decision-making of energy system operations.
  • Advanced smart energy system software tools need to be able to implement customized services for different energy systems of different users, be able to systematically address operational problems, and use AI algorithms to solve problems based on the physical relationship between energy systems and equipment.
  • AI algorithms should be versatile, and energy management systems should be scalable to reduce the inevitable man-hour investment in customized services and provide users with the ability to develop and expand.
  • Embodiments of the present invention are methods for designing energy management systems that meet these needs.
  • the current energy management system mainly includes: (1) Systems based on such as SCADA system framework, which often cannot provide machine learning or big data services, and only provide display, basic statistics, etc. using fixed graphics and tables. Traditional energy service functions such as comprehensive reports have poor customization and service scalability. (2) Energy management systems using big data analysis and AI algorithms as means. Such systems analyze and process raw data obtained from measurements. They cannot reflect the intrinsic physical connection relationships of actual service objects and The physical hierarchy of subsystems therefore cannot provide systematic digital services for the operation of user energy systems.
  • Figure 1 is a flow chart of a model training method of an energy management system according to an embodiment of the present invention. As shown in Figure 1, the method includes:
  • Step 101 Establish a mechanism model of the energy management system.
  • the mechanism model of the energy management system includes the mechanism models of subsystems and the connection relationships between the mechanism models of the subsystems.
  • Energy management systems usually include: perception layer, network layer, platform layer and application layer.
  • the entire energy management system can adopt a B/S architecture based on the Internet technology software body, and each business function unit system adopts Java EE's lightweight SSH or a multi-layer structure of similar architecture.
  • the perception layer includes integrated data collection, measurement analysis and real-time control systems, mainly composed of instrumentation equipment and on-site PLC controllers
  • the network layer can pass various network systems (ADSL, GPRS, 3G, 4G, optical fiber, etc.) , transmit real-time data to the dispatching and management center.
  • the dispatching and management center can also issue control instructions to the on-site controller through the network system to execute control and adjustment instructions;
  • the platform layer is responsible for receiving data from each on-site monitoring device and storing real-time operating parameters. In the data, basic data is provided for subsequent management, analysis, and control, and the data is stored, analyzed, alarmed, and reported printed;
  • the application layer is the direct use level for operators, and can conduct continuous dynamic analysis of uploaded data in real time, and can Issue adjustment instructions based on analysis results.
  • the energy management system can be divided based on the above-mentioned hierarchical approach to determine the various subsystems that make up the energy management system.
  • the energy management system may also be divided in other ways, and the embodiments of the present invention are not limited to this.
  • the process of establishing a mechanism model of the energy management system includes:
  • subsystems can include: energy supply networks (such as power supply, gas supply, cooling/heating, etc. networks), energy exchange links (such as CCHP units, generator units, boilers, air conditioners, heat pumps, etc.), energy storage links (electricity storage , gas storage, heat storage, cold storage, etc.), terminal comprehensive energy supply unit (such as microgrid), etc.
  • energy supply networks such as power supply, gas supply, cooling/heating, etc. networks
  • energy exchange links such as CCHP units, generator units, boilers, air conditioners, heat pumps, etc.
  • energy storage links electricality storage , gas storage, heat storage, cold storage, etc.
  • terminal comprehensive energy supply unit such as microgrid
  • the mechanism model of a subsystem also known as the white box model of the subsystem, is an accurate mathematical model that describes the subsystem based on the objects in the subsystem, the internal mechanism of the production process, or the transfer mechanism of the material flow. It can be a mathematical model of an object or process based on the mass balance equation, energy balance equation, momentum balance equation, phase balance equation, certain physical property equations, chemical reaction laws, etc.
  • the mechanism model parameters are easy to adjust, and the resulting model has strong adaptability.
  • connection relationships include: electrical connection; energy transfer connection; liquid flow connection; gas flow connection; information transmission connection; value connection, etc.
  • Figure 2 is a schematic diagram of selecting a mechanism model of a subsystem and establishing a connection relationship in a drag-and-drop manner according to the embodiment of the present invention.
  • the first subsystem 201, the second subsystem 202, the third subsystem 203 and the fourth subsystem are selected by dragging and dropping in the interface provided by the visual modeling tool (such as Vertabelo or Apache Spark, etc.) subsystem 204, and establish the connection relationship between the first subsystem 201, the second subsystem 202, the third subsystem 203 and the fourth subsystem 204 in a drag-and-drop manner, thereby forming a system including the first subsystem 201, the second subsystem 204 and the fourth subsystem 204.
  • connection relationship between the first subsystem 201, the second subsystem 202, the third subsystem 203 and the fourth subsystem 204 can be electrical connection, energy transfer connection, liquid flow connection, and gas flow connection. , information transfer connection or value connection (for example, cost relationship), etc.
  • Step 102 Input measurement data into the mechanism model of the energy management system, where the measurement data is measured at measurement points of the energy management system.
  • the measuring points can be actually arranged in the subsystems of the energy management system, or on the connection lines between subsystems.
  • real measurement data are provided for the mechanism model of the energy management system established in step 101.
  • Step 103 Based on the measurement data, simulate and run the mechanism model of the energy management system to obtain simulation data.
  • measurement data can include:
  • step 103 simulation data of the mechanism model of the energy management system can be obtained.
  • This simulation data can reflect the operation results of the entire business logic of the entire energy management system.
  • Step 104 Construct training data based on measurement data and simulation data.
  • training data for training the AI model is constructed.
  • the training data used can have one or more data characteristics.
  • step 104 includes: determining the target equipment of the AI model; determining the process-related equipment of the target equipment based on the process connection relationship of the target equipment in the mechanism model of the energy management system, where the process connection relationship includes at least one of the following: A: Electrical connection; Energy transfer connection; Liquid flow connection; Gas flow connection; Information transfer connection; Value connection, etc.; Extract measurement data of target equipment and measurement data of process-related equipment from measurement data; Extract from simulation data Simulation data of target equipment and simulation data of process-related equipment; construct training data including measurement data of target equipment, measurement data of process-related equipment, simulation data of target equipment, and simulation data of process-related equipment.
  • the target device is: compressor.
  • the gas storage tank and the refrigeration dryer can be determined to be the process-related equipment of the compressor.
  • the flow measurement data, temperature measurement data and pressure measurement data of the compressor are extracted from the measurement data.
  • the simulation value of the compressor efficiency, the simulation value of the heat exchange capacity index of each stage, the simulation value of the second-stage inlet temperature, the simulation value of the third-stage inlet temperature, etc. are extracted from the simulation data.
  • the operating pressure measurement data of the gas tank is extracted from the measurement data
  • the simulation value of the flow rate of the gas tank is extracted from the simulation data.
  • the inlet pressure measurement data of the cold dryer is extracted from the measurement data, and the simulation value of the processing capacity of the cold dryer is extracted from the simulation data. Then, based on the flow measurement data, temperature measurement data and pressure measurement data of the compressor, the working pressure measurement data of the gas storage tank, the inlet pressure measurement data of the refrigeration dryer, the simulation value of the compressor efficiency, and the heat exchange capacity of each stage.
  • the simulation value of the indicator, the simulation value of the second-level inlet temperature, the simulation value of the third-level inlet temperature, the simulation value of the flow rate of the gas storage tank, and the simulation value of the processing capacity of the cold dryer are used to build an AI model for training the compressor ( For example, the training data of compressor life prediction model or compressor performance prediction model, etc.).
  • the embodiment of the present invention can use the simulation data of the energy management system to enrich the training data of the AI model, because the simulation data already reflects the mechanism relationship of the mechanism model and includes the intrinsic physical connection relationship of the actual service object and the physical layer of the subsystem. structure, therefore the embodiment of the present invention also realizes the combination of mechanism and AI, enriches training data, and improves the performance of the AI model.
  • Step 105 Based on the training data, train the artificial neural network model into the AI model of the energy management system.
  • the artificial neural network model can include: feedforward neural network and feedback neural network.
  • Feedforward neural networks include convolutional neural networks (CNN), fully connected neural networks (FCN) or generative adversarial networks (GAN), etc.
  • Feedback neural networks include recurrent neural networks (RNN), long short-term memory networks (LSTM), Hopfield networks or Boltzmann machines, etc.
  • training the artificial neural network model as the AI model of the energy management system based on the training data including measurement data and simulation data includes: inputting the training data into the artificial neural network model, so that the artificial neural network model outputs corresponding to Predicted value of simulation data; determine the loss function value of the artificial neural network model based on the difference between the measured data and the predicted value in the training data; configure the model parameters of the artificial neural network model so that the loss function value is lower than the preset value Threshold; determine the configured artificial neural network model as an AI model.
  • simulation data with rich data volume and diverse types are used as inputs for generating predicted values, and measurement data with reliable results are used as actual values for comparison with measured values, thereby improving model training accuracy.
  • the measurement data includes the measured value of the cooling water temperature of the refrigerator
  • the simulation data includes the simulation value of the energy efficiency ratio and the power consumption of the refrigerator.
  • the energy efficiency ratio simulation value and power consumption simulation value of the refrigerator are input into the artificial neural network model, so that the artificial neural network model outputs the predicted value of the cooling water temperature.
  • the difference between the measured value of the cooling water temperature in the measurement data i.e., the actual value in model training
  • the predicted value of the cooling water temperature output by the artificial neural network model such as mean square error or cross entropy, etc.
  • the AI model determines the loss function value of the artificial neural network model; configure the model parameters of the artificial neural network model (for example, back propagation to update the model parameters) so that the loss function value is lower than the preset threshold.
  • the AI model can be used to predict the cooling water temperature of the refrigerator.
  • training the artificial neural network model as the AI model of the energy management system based on the training data including measurement data and simulation data includes: inputting the training data into the artificial neural network model, so that the artificial neural network model outputs corresponding to Measure the predicted value of the data; determine the loss function value of the artificial neural network model based on the difference between the simulated data and the predicted value in the training data; configure the model parameters of the artificial neural network model so that the loss function value is lower than the preset Threshold; determine the configured artificial neural network model as an AI model.
  • measurement data with reliable results are used as inputs for generating predicted values
  • simulation data with rich data volume and diverse types are used as actual values for comparison with measured values, thereby enriching the types of models.
  • the measurement data includes the measured value of the cooling water temperature of the refrigerator, the measured value of the chilled water temperature and the cooling capacity of the chiller, and the simulation data includes the simulation value of the energy efficiency ratio of the refrigerator.
  • the measured values of cooling water temperature, chilled water temperature and chiller cooling capacity are input into the artificial neural network model to output a predicted value of the energy efficiency ratio of the chiller.
  • the difference between the simulated value of the energy efficiency ratio in the simulation data (that is, equivalent to the actual value in model training) and the predicted value of the energy efficiency ratio output by the artificial neural network model (such as mean square error or cross entropy, etc.
  • the AI model can be used to predict the energy efficiency ratio of the refrigerator.
  • the AI model includes at least one of the following:
  • Performance monitoring model of subsystem Performance monitoring model of subsystem; performance prediction model of subsystem; performance diagnosis model of subsystem; optimized operation model of subsystem; performance monitoring model of energy management system; performance prediction model of energy management system; performance diagnosis of energy management system Model; optimal operation model of energy management system, etc.
  • test data in addition to generating training data based on measurement data and simulation data, test data can also be generated based on measurement data and simulation data for testing the trained AI model.
  • FIG. 3 is an exemplary process diagram of model training of the energy management system according to the embodiment of the present invention.
  • the mechanism model 200 of the energy management system includes a mechanism model 201 of the first subsystem, a mechanism model 202 of the second subsystem, a mechanism model 203 of the third subsystem, and a mechanism model 204 of the fourth subsystem.
  • the measurement data 11 as input data are provided to the mechanism model 200 of the energy management system. Based on the input data, a large amount of simulation data 21 can be obtained by simulating and running the mechanism model 200 of the energy management system.
  • the simulation data 21 and the measurement data 11 are jointly constructed as training data 30, and the artificial neural network model 31 is trained using the training data 30.
  • the artificial neural network model 31 that has completed training is the AI model of the energy management system.
  • FIG. 4 is a schematic diagram of the processing framework of the energy management system according to the embodiment of the present invention.
  • a mechanism model establishment process 61 is first performed to generate a mechanism model of the energy management system.
  • the test data is input into the mechanism model, and simulation processing 61 (for example, dynamic simulation according to a certain frequency) is performed to obtain a large amount of simulation data.
  • the simulation data and test data are used to construct training data, and the training data is used to train the AI model 50 for monitoring/prediction/diagnosis and/or the AI model 51 for optimization operation.
  • the AI model 50 successively undergoes a training process 52, a verification process 53, and an online use process 54.
  • the AI model 51 successively undergoes a training process 55 , a verification process 56 and an online use process 57 .
  • the functional modularization of the embodiment of the present invention may include dragging model icons to establish human-computer interaction functions, simulation functions, energy efficiency, performance monitoring, prediction, diagnostic modules, parameter optimization operations and Scheduling module.
  • the human-computer interaction function is used to customize service objects, simulate at a certain frequency for real-time monitoring and support the other two functions. Through these functional modules and their sub-functions, systematic digital services can be provided for users by system and problem.
  • the embodiment of the present invention can customize the interface of the user system by dragging the model icon.
  • the model represented by the icon has its corresponding attributes to express its design parameters, operating parameters, economic indicator parameters, market parameters, etc., and can also customize the system for specific users. , operation problems and reduce customization man-hours.
  • the simulation function block is based on the system, equipment and their physical relationships described in the topology diagram, reading in the measurement point values, and calculating the simulation data (including other unmeasurable parameters, operating performance indicators, operating economic indicators, etc.).
  • the simulation reads data at a certain time frequency, performs calculations, and stores the results in the database in the time dimension.
  • physical relationships and data relationships are used to calculate unmeasurable parameters, implement soft measurements, and calculate other abstract parameters with physical meaning and some dimensionless similarity criterion parameters, making full use of the actual physical relationships.
  • Simulation provides the foundation for AI applications and realizes the comprehensive integration of physical relationships and AI algorithms.
  • the functions of the AI model can include: energy efficiency monitoring, performance and usage prediction, etc.
  • the AI model can directly use the accumulated data of measurement or the historical data formed by simulation. Diagnosis of operational problems is carried out by running AI algorithms and logical judgments based on monitoring and prediction results. By using simulation data, we make full use of the value provided by physical relationships and abstract parameter groups of physical science. Optimizing the operation function requires the use of performance prediction models, economic models and emission models of relevant systems and their equipment, and these models will be obtained based on accumulated simulation data, or require calibration with accumulated simulation data. Rolling optimization can be performed in real time at a certain time frequency. Each rolling calculation can obtain the latest system and equipment status from measurement points or other real-time systems, and the results based on the current status are sent to the control system.
  • Each AI module can include: machine learning model building tool and machine learning result verification process. Machine learning is selected in the corresponding function and thus put into online use.
  • the algorithms of each AI module are universal. These algorithms can be divided into several categories according to sub-functions, and each sub-function is a category. Each category can create multiple instances. Each instance uses a different topology map. In the configuration interaction of the corresponding instance, different configuration prediction, diagnosis, optimization goals, and different configuration impact parameters define different specific problems. .
  • Each instance in each AI module can perform rolling calculations according to their own frequency, obtain the corresponding latest input information and data from databases, measuring points, and other real-time update systems, and refresh their respective results.
  • FIG. 5 is an exemplary structural diagram of a model training device of an energy management system according to an embodiment of the present invention.
  • the model training device 500 of the energy management system includes: a building module 501 configured to establish a mechanism model of the energy management system.
  • the mechanism model of the energy management system includes the mechanism model of the subsystem and the connection relationship between the mechanism models of the subsystem;
  • input Module 502 is configured to input measurement data into the mechanism model of the energy management system, where the measurement data is measured at measurement points of the energy management system; simulation module 503 is configured to simulate and run the energy management system based on the measurement data.
  • the construction module 504 is configured to construct training data based on measurement data and simulation data;
  • the training module 505 is configured to train the artificial neural network model into an AI model of the energy management system based on the training data .
  • the AI model includes at least one of the following: a performance monitoring model of a subsystem; a performance prediction model of a subsystem; a performance diagnosis model of a subsystem; an optimized operation model of a subsystem; a performance monitoring model of an energy management system ; Performance prediction model of energy management system; Performance diagnosis model of energy management system; Optimization operation model of energy management system.
  • the training module 505 is configured to input the training data into the artificial neural network model, so that the artificial neural network model outputs a predicted value corresponding to the simulation data; based on the relationship between the measured data and the predicted value in the training data The difference value determines the loss function value of the artificial neural network model; configures the model parameters of the artificial neural network model so that the loss function value is lower than the preset threshold; determines the configured artificial neural network model as an AI model.
  • the training module 505 is configured to input the training data into the artificial neural network model, so that the artificial neural network model outputs a predicted value corresponding to the measured data; based on the relationship between the simulation data and the predicted value in the training data The difference value determines the loss function value of the artificial neural network model; configures the model parameters of the artificial neural network model so that the loss function value is lower than the preset threshold; determines the configured artificial neural network model as an AI model.
  • the building module 504 is configured to determine the target device of the AI model; determine the process-related devices of the target device based on the process connection relationship of the target device in the mechanism model of the energy management system, where the process connection relationship includes the following At least one of: electrical connection; energy transfer connection; liquid flow connection; gas flow connection; information transfer connection; value connection; extraction of measurement data of target equipment and measurement data of process-related equipment from measurement data; extraction from simulation data Simulation data of target equipment and simulation data of process-related equipment; construct training data including measurement data of target equipment, measurement data of process-related equipment, simulation data of target equipment, and simulation data of process-related equipment.
  • FIG. 6 is an exemplary structural diagram of an electronic device according to an embodiment of the present invention.
  • the electronic device 600 includes a processor 601, a memory 602, and a computer program stored on the memory 602 and executable on the processor 601.
  • the computer program is executed by the processor 601, any of the above energy management can be implemented.
  • the memory 602 can be implemented as various storage media such as electrically erasable programmable read-only memory (EEPROM), flash memory (Flash memory), programmable programmable read-only memory (PROM), etc.
  • the processor 601 may be implemented to include one or more central processing units or one or more field programmable gate arrays, where the field programmable gate array integrates one or more central processing unit cores.
  • the central processing unit or central processing unit core may be implemented as a CPU, an MCU, a DSP, or the like.
  • each step is not fixed and can be adjusted as needed.
  • the division of each module is only for the convenience of describing the functional division. In actual implementation, one module can be implemented by multiple modules, and the functions of multiple modules can also be implemented by the same module. These modules can be located on the same device. , or it can be on a different device.
  • the hardware modules in various embodiments may be implemented mechanically or electronically.
  • a hardware module may include specially designed permanent circuits or logic devices (such as a dedicated processor such as an FPGA or ASIC) to perform specific operations.
  • Hardware modules may also include programmable logic devices or circuits (eg, including general-purpose processors or other programmable processors) temporarily configured by software to perform specific operations.
  • programmable logic devices or circuits eg, including general-purpose processors or other programmable processors

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Hardware Design (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Geometry (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

L'invention concerne un procédé d'entraînement de modèle pour un système de gestion d'énergie, un appareil et un support de stockage. Le procédé consiste à : établir un modèle de mécanisme d'un système de gestion d'énergie, le modèle de mécanisme du système de gestion d'énergie comprenant des modèles de mécanisme de sous-systèmes et une relation de connexion entre les modèles de mécanisme des sous-systèmes (étape 101); entrer des données de mesure dans le modèle de mécanisme du système de gestion d'énergie, les données de mesure étant mesurées au niveau d'un point de mesure du système de gestion d'énergie (étape 102); sur la base des données de mesure, effectuer une simulation de fonctionnement du modèle de mécanisme du système de gestion d'énergie pour obtenir des données de simulation (étape 103) ; sur la base des données de mesure et des données de simulation, construire des données d'apprentissage (étape 104) ; et, sur la base des données d'apprentissage, entraîner un modèle de réseau neuronal artificiel dans un modèle d'intelligence artificielle du système de gestion d'énergie (étape 105). Le procédé introduit une capacité d'intelligence artificielle dans le système de gestion d'énergie, combinant ainsi le mécanisme et l'intelligence artificielle, enrichissant les données d'apprentissage, et améliorant les performances du modèle d'intelligence artificielle.
PCT/CN2022/115274 2022-08-26 2022-08-26 Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage WO2024040608A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/115274 WO2024040608A1 (fr) 2022-08-26 2022-08-26 Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/115274 WO2024040608A1 (fr) 2022-08-26 2022-08-26 Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage

Publications (1)

Publication Number Publication Date
WO2024040608A1 true WO2024040608A1 (fr) 2024-02-29

Family

ID=90012234

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/115274 WO2024040608A1 (fr) 2022-08-26 2022-08-26 Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage

Country Status (1)

Country Link
WO (1) WO2024040608A1 (fr)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106447113A (zh) * 2016-10-08 2017-02-22 东南大学 一种基于运行优化模型的多区域综合能源系统运行方法
CN109858136A (zh) * 2019-01-26 2019-06-07 新奥数能科技有限公司 一种燃气锅炉效率的确定方法和装置
CN110275493A (zh) * 2018-03-15 2019-09-24 西门子股份公司 用于控制技术系统的方法和装置
KR102073323B1 (ko) * 2018-09-02 2020-02-03 주식회사 뉴로다임 인공지능 기반의 구조물 건전성 관리 시스템
CN113295722A (zh) * 2021-05-21 2021-08-24 厦门大学 一种基于深度学习算法的x射线光谱数据矫正方法及装置
CN114237057A (zh) * 2021-12-20 2022-03-25 东南大学 智慧楼宇用电分析的动态建模方法、系统、设备和介质
CN114371409A (zh) * 2022-03-07 2022-04-19 上海玫克生储能科技有限公司 电池状态预测模型的训练方法、电池状态预测方法及装置
CN114424217A (zh) * 2019-09-30 2022-04-29 西门子股份公司 综合能源系统的优化方法、装置和计算机可读存储介质
CN114450682A (zh) * 2019-09-30 2022-05-06 西门子股份公司 综合能源系统的控制方法、装置和计算机可读存储介质
CN114503120A (zh) * 2019-09-30 2022-05-13 西门子股份公司 综合能源系统的仿真方法、装置和计算机可读存储介质

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106447113A (zh) * 2016-10-08 2017-02-22 东南大学 一种基于运行优化模型的多区域综合能源系统运行方法
CN110275493A (zh) * 2018-03-15 2019-09-24 西门子股份公司 用于控制技术系统的方法和装置
KR102073323B1 (ko) * 2018-09-02 2020-02-03 주식회사 뉴로다임 인공지능 기반의 구조물 건전성 관리 시스템
CN109858136A (zh) * 2019-01-26 2019-06-07 新奥数能科技有限公司 一种燃气锅炉效率的确定方法和装置
CN114424217A (zh) * 2019-09-30 2022-04-29 西门子股份公司 综合能源系统的优化方法、装置和计算机可读存储介质
CN114450682A (zh) * 2019-09-30 2022-05-06 西门子股份公司 综合能源系统的控制方法、装置和计算机可读存储介质
CN114503120A (zh) * 2019-09-30 2022-05-13 西门子股份公司 综合能源系统的仿真方法、装置和计算机可读存储介质
CN113295722A (zh) * 2021-05-21 2021-08-24 厦门大学 一种基于深度学习算法的x射线光谱数据矫正方法及装置
CN114237057A (zh) * 2021-12-20 2022-03-25 东南大学 智慧楼宇用电分析的动态建模方法、系统、设备和介质
CN114371409A (zh) * 2022-03-07 2022-04-19 上海玫克生储能科技有限公司 电池状态预测模型的训练方法、电池状态预测方法及装置

Similar Documents

Publication Publication Date Title
WO2019200662A1 (fr) Procédé d'évaluation de stabilité et de commande statique de système à énergie totale électricité-chaleur-gaz
Pan et al. Building energy simulation and its application for building performance optimization: A review of methods, tools, and case studies
Zhang et al. Short-time multi-energy load forecasting method based on CNN-Seq2Seq model with attention mechanism
CN102411729B (zh) 基于自适应线性逻辑网络的风电功率预测方法
CN115238959A (zh) 一种面向用户侧能源综合利用的数字孪生系统和方法
Huang et al. A multi-rate dynamic energy flow analysis method for integrated electricity-gas-heat system with different time-scale
CN115758674A (zh) 一种电能表信息化评价校准模型测评验证系统及方法
Ahmad et al. Efficient energy planning with decomposition-based evolutionary neural networks
CN109412155B (zh) 一种基于图计算的配电网供电能力评估方法
CN102592004B (zh) 钢铁联合企业全工序能况在线分析诊断系统及方法
Mystakidis et al. One step ahead energy load forecasting: A multi-model approach utilizing machine and deep learning
CN113536591A (zh) 一种综合能源系统变步长动态仿真方法
WO2024040608A1 (fr) Procédé d'entraînement de modèle pour système de gestion d'énergie, appareil et support de stockage
CN112365082A (zh) 一种基于机器学习的公共能源消耗预测方法
Parfenenko et al. Information system for monitoring and forecast of building heat consumption
Testasecca et al. Towards Digital Twins of buildings and smart energy networks: Current and future trends
Dong et al. Hybrid possibilistic-probabilistic energy flow assessment for multi-energy carrier systems
Wang Application of deep learning model in building energy consumption prediction
Gu et al. A Digital Load Forecasting Method Based on Digital Twin and Improved GRU
WO2023004809A1 (fr) Procédé et appareil de modélisation pour système de gestion d'énergie intelligent, et support de stockage
CN109767044B (zh) 一种对地源热泵系统热能转换效率进行预测的方法
Wu et al. Neural network based feasible region approximation model for optimal operation of integrated electricity and heating system
Keyan et al. Anomaly detection method of distribution network line loss based on hybrid clustering and LSTM
CN115619334A (zh) 能源管理系统的动态监测装置、方法、电子设备及介质
CN112465214B (zh) 一种综合能源系统全生命周期的耦合度计算方法

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: 22956153

Country of ref document: EP

Kind code of ref document: A1