Load aggregation power model and modeling method thereof

Abstract

The application discloses a kind of load aggregation power models and its modeling method, the modeling method includes the following steps: S1: obtaining sample data;S2: respectively establish equipment and space internal air, space and outside heat exchange mathematical model, and obtain the differential equation of thermal power under aggregation form in this way;S3: by reducing time step, the differential equation is converted into difference equation, and based on black box-gray box learning method, the thermal characteristics in the difference equation are learned;S4: according to the thermal characteristics learned, obtain the internal heat power model of space, and the aggregation thermal power of interconnected equipment in space is characterized by internal heat power of space;S5: according to the internal heat power model, multi-scenario optimization modeling is carried out, and the internal optimization heat power model of space is obtained.The application can improve the precision of model, reduce the dependence on the size of sampling data, and realize the optimization heat power model in multiple scenes according to different targets.

Description

Technical Field

[0001] This invention relates to the field of load modeling technology, and in particular to a load aggregated power model and its modeling method. Background Technology

[0002] The Internet of Things (IoT) refers to a network of interconnected devices and the technologies that facilitate communication between devices and the cloud, as well as between devices themselves. Flexible load power management is crucial for coordinating loads within interconnected devices. To achieve this, an Energy Management System (EMS) requires a high-precision load aggregation power model applicable to various objectives to estimate different energy changes caused by EMS operation and to optimize load power scheduling.

[0003] Currently, the inherent physical model and data-driven approach are used to model the loads in the space separately, and a gray box is developed to learn the load aggregation power model. The main steps are: first, to establish a physical model of energy exchange in the region based on the system structure, to derive differential equations such as the distributed thermal energy flow in the study region, and to determine the equation parameters; then, to collect system input and output data through sensors, such as temperature data of different locations and interconnected devices in the region; and finally, to learn the model features by using optimization methods and different strategies.

[0004] However, the aforementioned modeling methods can only model multiple loads individually within a space. Furthermore, the model parameters typically include constant components and time-varying components that change with human behavior and environmental conditions, while gray-box modeling can only learn the constant characteristics of the model, ignoring its time-varying features; and the results are highly dependent on the selection of the test set. Therefore, a new method for modeling aggregated load power is urgently needed. Summary of the Invention

[0005] To address the aforementioned problems, this invention aims to provide a load aggregation power model and its modeling method.

[0006] The technical solution of the present invention is as follows: On the one hand, a method for modeling load aggregation power is provided, including the following steps: S1: Collect basic parameters of each device in the Internet of Things (IoT) at the environmental layer, and aggregate the basic parameters according to the IoT's perception layer and communication layer to obtain sample data; S2: Establish mathematical models for heat exchange between the equipment and the air in the space, and between the space and the outside world, and obtain the differential equation for heat power in the aggregation form. S3: The differential equation is transformed into a difference equation by reducing the time step, and the thermal features in the difference equation are learned based on the black-box-grey-box learning method. S4: Obtain the internal thermal power model of the space based on the learned thermal characteristics, and characterize the aggregated thermal power of interconnected devices in the space through the internal thermal power of the space. S5: Based on the proprietary thermal power model, perform optimization modeling for multiple scenarios to obtain an optimized thermal power model for the interior of the space.

[0007] Preferably, in step S2, the differential equation for the thermal power in the polymerization form is: (1) (2) (3) (4) In the formula: subscripts, Indicates the air inside the space. Indicates outside air. Indicates heating equipment. Indicates the local weather. This refers to the air flowing from the heating equipment into the space. It indicates the flow of air from inside the space to outside the space. This indicates the difference between external air quality and local weather. This indicates that the air inside the space is being disturbed. This indicates interference with the outside air; express Temperature at any moment; Indicates time; , These represent the constant characteristic matrices of the state variables and the inputs, respectively; This represents the thermal features that need to be learned, including constant features and time-varying features; Indicates thermal conductivity; Indicates specific heat capacity; Indicates density; Indicates volume; Indicates the rate of heat exchange; Indicates the time-varying component in the thermal characteristics; In step S3, the difference equation is: (5) In the formula: This represents a 2×2 identity matrix.

[0008] Preferably, in step S3, when learning the thermal features in the difference equation based on the black-box-grey-box learning method, the gray-box method is first used to learn the constant features in the thermal features, and then the black-box method is used to learn the time-varying features in the thermal features.

[0009] Preferably, when using gray-box learning to learn the constant features, multiple different learning methods are used for learning, and then the learning results of each learning method are aggregated to obtain the constant features.

[0010] Preferably, the learning method includes fitting and statistical methods, clustering methods, and classification methods.

[0011] As a preferred approach, when using fitting and statistical methods for learning, different methods are first used each day. , and To learn the entire training set's timeframe, the formula is: (6) (7) In the formula: Represent the objective function; Indicates the number of days in the training set; Indicates a typical day; Indicates in Learning error over a period of time; , , All represent the learned constant features; Represents a 2×2 identity matrix; Then, using fitting and statistical methods, the constant features are obtained based on maximum likelihood estimation and kernel density estimation, as shown in the formula: (8) (9) In the formula: Represent the objective function; , , This represents the learned constant features.

[0012] As a preferred approach, when using clustering methods for learning, Euclidean weighting or Jaccard weighting is employed; when using classification methods for learning, the selected categories include human behavior, equipment heating status, and weather conditions.

[0013] Preferably, when learning the time-varying features in the thermal features using a black-box method, a deep GRU network based on a black-box method is used to learn the thermal model.

[0014] As a preferred option, in step S5, when performing optimization modeling, the thermal power is optimized from the scenario of minimizing user costs or the thermal load is optimized for the goal of green electricity consumption. When optimizing the scheduling of thermal power to minimize user costs, the objective function is to minimize user thermal load costs: (10) In the formula: This represents the objective function that aims to minimize the user's heat load cost. This indicates the total time period to be optimized; Indicates local Real-time electricity price for specific time periods; Describe the objective function After optimization Thermal power over a given period; Indicates the time step; Load fluctuation constraints are: (11) In the formula: and They represent the objective functions respectively. Optimize the maximum and minimum values ​​of the load curve; This represents the load fluctuation threshold set by the user. The power constraint is: (12) In the formula: Indicates the upper limit of the space load power; When scheduling heat load to achieve the goal of green electricity consumption, the objective function is to maximize the utilization rate of renewable energy. (13) In the formula: This represents the objective function aimed at maximizing the utilization rate of new energy sources. Describe the objective function After optimization Thermal power over a given period; Indicates new energy power generation equipment Power during a given time period; Load fluctuation constraints are: (14) In the formula: and They represent the objective functions respectively. Optimize the maximum and minimum values ​​of the load curve; This represents the load fluctuation threshold set by the user. The power constraint is: (15).

[0015] On the other hand, a load aggregation power model is also provided, which is established using any of the load aggregation power modeling methods described above.

[0016] The beneficial effects of this invention are: This invention employs multiple methods to accurately learn the constant and time-varying features of the thermal model, resulting in high model accuracy. It has a wide range of applications, allowing for the use of different learning methods to improve model feature accuracy based on varying test set sizes. Besides establishing regional thermal models, the black-box / grey-box model can also be used for aggregated modeling of distributed energy resources, enabling reasonable estimation of indoor temperature changes during EMS operation. Furthermore, the model exhibits high flexibility, allowing for model optimization for different scenarios based on its proprietary thermal power model, and enabling reasonable scheduling of load usage periods within the space in conjunction with EMS. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of the load aggregation power modeling method of the present invention; Figure 2 This is a schematic diagram of the data processing flow for the load aggregation power modeling method of the present invention; Figure 3 This is a schematic diagram of the black-box-grey-box learning method for modeling load aggregation power in this invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and technical features described in this application can be combined with each other. It should also be pointed out that, unless otherwise indicated, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terms "comprising" or "including" and similar words used in this invention refer to elements or objects preceding the word that encompass the elements or objects listed following the word and their equivalents, without excluding other elements or objects.

[0020] On the one hand, such as Figure 1-3 As shown, the present invention provides a method for modeling load aggregation power, comprising the following steps: S1: Collect basic parameters of each device in the Internet of Things (IoT) at the environmental layer, and aggregate the basic parameters according to the IoT's perception layer and communication layer to obtain sample data.

[0021] A space exchanges heat with the outside world and contains multiple interconnected loads via the Internet of Things (IoT). These loads can be categorized by function, such as heaters, air conditioners, and lighting equipment. The thermal power model is typically a state function composed of multiple parameters. Some parameters can be directly acquired, while others (such as heat capacity and conduction) are difficult to obtain directly and therefore require learning from measured temperature data.

[0022] The IoT architecture is divided into an environment layer, a perception layer, a communication layer, and an application layer. In a specific embodiment, temperature data at various locations within the space is collected by multiple DHT-11 sensors connected to a CC-2530 board. The sensor model can be selected based on the data to be measured. The CC-2530 is a System-on-Chip (SoC) solution compatible with the IEEE 802.15.4 protocol. One CC-2530 board is set as a central node, aggregating data from other CC-2530 boards based on the 802.15.4 protocol and ZigBee communication, and then sending it to the node. A Raspberry Pi microcomputer is used as a node, communicating with the central node via serial communication to achieve data integration.

[0023] S2: Establish mathematical models for heat exchange between the equipment and the air in the space, and between the space and the outside world, and obtain the differential equation for heat power in the aggregate form.

[0024] In a specific embodiment, the heat stored in the air within the space is represented by , satisfying: (16) In the formula: express The heat stored in the air within a given space at any given time; in the subscript, Indicates the air inside the space; Indicates specific heat capacity; Indicates density; Indicates volume; express Temperature at any moment; Indicates time; Room temperature changes over time, therefore the heat stored in the air within the space also changes over time. Let... The change is determined by the heat exchange between the air inside the space and the outside air and the interfering factors, then: (17) (18) (19) (20) In the formula: subscripts, This refers to the air flowing from the heating equipment into the space. It indicates the flow of air from inside the space to outside the space. This indicates that the air inside the space is being disturbed. Indicates outside air; express The rate of heat exchange at any given time; This represents the change in heat increment over time. Indicates thermal conductivity; This represents the hypothetical heat exchange function between the disturbance and the air within the space; The number of equations (18) can be freely expanded according to the number of devices in the space. The differential equation of thermal power in the polymerization form shown in the following equations (16)-(20) can be obtained: (1) (2) (3) (4) In the formula: subscripts, Indicates heating equipment. Indicates the local weather. This indicates the difference between external air quality and local weather. This indicates interference with the outside air; , These represent the constant characteristic matrices of the state variables and the inputs, respectively; This represents the thermal features that need to be learned, including constant features and time-varying features; Indicates the rate of heat exchange; This represents the time-varying component in the thermal characteristics.

[0025] S3: The differential equation is transformed into a difference equation by reducing the time step, and the thermal features in the difference equation are learned based on the black-box-grey-box learning method.

[0026] In one specific embodiment, the difference equation is: (5) In the formula: This represents a 2×2 identity matrix.

[0027] Comparing equations (1) and (5), it can be found that transforming the differential equation into a difference equation has little impact on the accuracy of the model, but can significantly reduce the computational complexity of the numerical solution. The thermal model parameters are limited by the number of sensors, and the matrix parameters can be shrunk or expanded according to the number of sensors to achieve flexible modeling based on different load distribution types.

[0028] S4: Obtain the internal thermal power model of the space based on the learned thermal characteristics, and characterize the aggregated thermal power of interconnected devices in the space through the internal thermal power of the space.

[0029] In one specific embodiment, when learning the thermal features in the difference equation based on the black-box-grey-box learning method, the gray-box method is first used to learn the constant features in the thermal features, and then the black-box method is used to learn the time-varying features in the thermal features. Optionally, the gray-box method learns the constant features of the thermal model based on the physical model through a method pool; the black-box method relies on historical data and the physical model to learn the time-varying features through a deep GRU network.

[0030] In one specific embodiment, when using gray-box learning of the constant features, multiple different learning methods are employed for learning, and then the learning results of each learning method are aggregated using an aggregation method to obtain the constant features; the learning methods include fitting and statistical methods, clustering methods, and classification methods.

[0031] In one specific embodiment, when learning using fitting and statistical methods, different methods are first used each day. , and To learn the entire training set's timeframe, the formula is: (6) (7) In the formula: Represent the objective function; Indicates the number of days in the training set; Indicates a typical day; Indicates in Learning error over a period of time; , , All represent the learned constant features; Represents a 2×2 identity matrix; Then, using fitting and statistical methods, the constant features are obtained based on maximum likelihood estimation and kernel density estimation, as shown in the formula: (8) (9) In the formula: Represent the objective function; , , This represents the learned constant features.

[0032] In the above embodiments, the results of equations (6)-(7) are the best learned features for each specific date in the test set, but the accuracy is limited across the entire test set. More accurate constant features can be obtained through equations (8)-(9), and this method can be flexibly selected according to the size of the test set.

[0033] In one specific embodiment, when using a clustering method for learning, Euclidean weighting or Jaccard weighting is employed; when using a classification method for learning, the selected classifications include human behavior, equipment heating status, and weather conditions.

[0034] In the above embodiments, the Euclidean weighted method is a widely used clustering method used to calculate the minimum variance of the Euclidean distance between data points. The days in the training set are grouped into several groups based on the correlation of the temperature curves; while for the Jaccard weighted method, days with closer Jaccard correlation values ​​in the temperature curves are grouped together to train a new set. , , .

[0035] The objective function of the clustering method is: (twenty one) (twenty two) In the formula: This represents the objective function of the clustering method. express Number of days in the group; () indicates error; , , All represent the features of the thermal model learned through clustering methods; When the training set is small, clustering methods struggle to extract data features. In such cases, classification methods are used to divide the training set into several groups based on experience and physical knowledge. Three factors that may influence spatial thermal characteristics are considered: human behavior, heating system status, and weather conditions. Human behavior is taken into account; when people are in the room, the temperature is typically higher and more stable. The objective function in this case is... for: (twenty three) In the formula: , These represent the time spent indoors and outdoors, respectively. )and All of these are error functions.

[0036] For other classification methods, the objective function can also be expressed as: (twenty four) In the formula: Indicates the number of groups in a day; Indicates the first Error function for class state; Indicates the first The time of class state.

[0037] If the load within a space includes heating equipment such as heaters, its heating status will significantly affect the room's thermal characteristics. Based on the space's temperature status, the space is divided into several categories. In this case, the heat capacity of the air within the space may differ under different conditions. Considering weather conditions, two strategies are used to classify the weather: a) the average temperature throughout the day; b) the time periods of temperature variation within a day. The first strategy divides the days in the training set into several groups, and the objective function is shown below: (25) In the formula: Indicates the number of days in a certain group; express The group's error function.

[0038] The second strategy categorizes the temperature periods of each day, with the objective function being the same as (24).

[0039] In the above embodiments, through fitting and statistical methods, clustering methods, and classification methods, the learning results of multiple constant features are obtained in the gray box, which serve as a constant feature set. The obtained learning results include: 1) time information, such as date, hour, and minute; 2) the expected temperature of the radiator and local weather, for example... and 3) The learning results of gray box, such as The above results are used as input for time-varying feature learning in black boxes.

[0040] In one specific embodiment, when the black box learns the time-varying features in the thermal features, a black box-based deep GRU network is used to learn the thermal model.

[0041] In the above embodiments, GRU refers to a gated recurrent unit. A deep GRU network consists of an input layer, several GRU layers, a fully connected layer, and an output layer. It has high learning efficiency for time-varying features, and its learning curve exhibits lag, often resulting in a smooth curve, making it suitable for hot model learning. The input is sent to multiple GRU layers, each containing many GRU units. It is the timing input of the GRU unit. Its output is calculated by equation (26). and These are the update and reset gates of the GRU unit, expressed by equations (27)-(28). and This represents the weight state of the two gates. The hidden state of the candidate data is represented by equation (29): (26) (27) (28) (29) After learning through a deep GRU network, a new set of learning results is obtained. and Based on the new output, time-varying characteristics It can be calculated using equation (30): (30) Learn the internal and external heat capacity of the space , and thermal conductivity , Substituting into equation (16) yields the internal thermal power model of the space, i.e., the air thermal power curve. The combined thermal power of interconnected devices within the space is characterized by room thermal power.

[0042] S5: Based on the proprietary thermal power model, perform optimization modeling for multiple scenarios to obtain an optimized thermal power model for the interior of the space.

[0043] In one specific embodiment, when performing optimization modeling, the thermal power is optimized to minimize user costs or the thermal load is optimized to achieve the goal of green electricity consumption. When optimizing the scheduling of thermal power to minimize user costs, the objective function is to minimize user thermal load costs: (10) In the formula: This represents the objective function that aims to minimize the user's heat load cost. This indicates the total time period to be optimized; Indicates local Real-time electricity price for specific time periods; Describe the objective function After optimization Thermal power over a given period; Indicates the time step; Load fluctuation constraints are: (11) In the formula: and They represent the objective functions respectively. Optimize the maximum and minimum values ​​of the load curve; This represents the load fluctuation threshold set by the user. The power constraint is: (12) In the formula: Indicates the upper limit of the space load power; When scheduling heat load to achieve the goal of green electricity consumption, the objective function is to maximize the utilization rate of renewable energy. (13) In the formula: This represents the objective function aimed at maximizing the utilization rate of new energy sources. Describe the objective function After optimization Thermal power over a given period; Indicates new energy power generation equipment Power during a given time period; Load fluctuation constraints are: (14) In the formula: and They represent the objective functions respectively. Optimize the maximum and minimum values ​​of the load curve; This represents the load fluctuation threshold set by the user. The power constraint is: (15).

[0044] On the other hand, the present invention also provides a load aggregation power model, which is established using any of the load aggregation power modeling methods described above.

[0045] In a specific embodiment, equations (10)-(12) represent the application direction scheduling model with the lowest user load cost. The optimized thermal power model is used for applications with the lowest user load cost. Equations (13)-(15) are the scheduling models for green electricity consumption applications. The optimized thermal power model is designed for green energy consumption applications.

[0046] In summary, this invention improves the accuracy of thermal models by employing multiple methods to learn the constant and time-varying characteristics of the models. Compared with existing technologies, this invention represents a significant advancement.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for modeling load aggregation power, characterized in that, Includes the following steps: S1: Collect basic parameters of each device in the Internet of Things (IoT) at the environmental layer, and aggregate the basic parameters according to the IoT's perception layer and communication layer to obtain sample data; S2: Establish mathematical models for heat exchange between the equipment and the air in the space, and between the space and the outside world, respectively, and obtain the differential equation for the heat power in the aggregation form; the differential equation for the heat power in the aggregation form is: （1） （2） （3） （4） In the formula: subscripts, Indicates the air inside the space. Indicates outside air. Indicates heating equipment. Indicates the local weather. This refers to the air flowing from the heating equipment into the space. It indicates the flow of air from inside the space to outside the space. This indicates the difference between external air quality and local weather. This indicates that the air inside the space is being disturbed. This indicates that the outside air is being disturbed; express Temperature at any moment; Indicates time; , These represent the constant characteristic matrices of the state variables and the inputs, respectively; This represents the thermal features that need to be learned, including constant features and time-varying features; Indicates thermal conductivity; Indicates specific heat capacity; Indicates density; Indicates volume; Indicates the rate of heat exchange; Indicates the time-varying component in the thermal characteristics; S3: The differential equation is transformed into a difference equation by reducing the time step, and the thermal features in the difference equation are learned based on the black-box-grey-box learning method; the difference equation is: （5） In the formula: Represents a 2×2 identity matrix; When learning the thermal features in the difference equation based on the black-box-grey-box learning method, the gray box is first used to learn the constant features in the thermal features, and then the black box is used to learn the time-varying features in the thermal features. When using gray-box learning to acquire the constant features, multiple different learning methods are employed, and then the learning results of each method are aggregated to obtain the constant features. The learning methods include fitting and statistical methods, clustering methods, and classification methods. When using fitting and statistical methods for learning, first use different methods each day. , and The formula for learning the entire training set period is: （6） （7） In the formula: Represent the objective function; Indicates the number of days in the training set; Indicates a typical day; Indicates in Learning error over a period of time; , , All represent the learned constant features; Represents a 2×2 identity matrix; Then, using fitting and statistical methods, the constant features are obtained based on maximum likelihood estimation and kernel density estimation, as shown in the formula: （8） （9） In the formula: Represent the objective function; , , This represents the learned constant features; S4: Obtain the internal thermal power model of the space based on the learned thermal characteristics, and characterize the aggregated thermal power of interconnected devices in the space through the internal thermal power of the space. S5: Based on the proprietary thermal power model, perform optimization modeling for multiple scenarios to obtain an optimized thermal power model for the interior of the space.

2. The load aggregation power modeling method according to claim 1, characterized in that, When using clustering methods for learning, Euclidean weighting or Jaccard weighting is employed; when using classification methods for learning, the selected categories include human behavior, equipment heating status, and weather conditions.

3. The load aggregation power modeling method according to claim 1, characterized in that, When learning the time-varying features in the thermal features using a black-box approach, a deep GRU network based on a black-box approach is used to learn the thermal model.

4. The load aggregation power modeling method according to any one of claims 1-3, characterized in that, In step S5, when performing optimization modeling, optimize the scheduling of thermal power from the scenario of minimizing user costs or perform thermal load scheduling with the goal of green electricity consumption. When optimizing the scheduling of thermal power to minimize user costs, the objective function is to minimize user thermal load costs: （10） In the formula: This represents the objective function that aims to minimize the user's heat load cost. This indicates the total time period to be optimized; Indicates local Real-time electricity price for specific time periods; Describe the objective function After optimization Thermal power over a given period; Indicates the time step; Load fluctuation constraints are: （11） In the formula: and They represent the objective functions respectively. Optimize the maximum and minimum values ​​of the load curve; This represents the load fluctuation threshold set by the user. The power constraint is: （12） In the formula: Indicates the upper limit of the space load power; When scheduling heat load to achieve the goal of green electricity consumption, the objective function is to maximize the utilization rate of renewable energy. （13） In the formula: This represents the objective function aimed at maximizing the utilization rate of new energy sources. Describe the objective function After optimization Thermal power over a given period; Indicates new energy power generation equipment Power during a given time period; Load fluctuation constraints are: （14） In the formula: and They represent the objective functions respectively. Optimize the maximum and minimum values ​​of the load curve; This represents the load fluctuation threshold set by the user. The power constraint is: （15）。 5. A load aggregation power model, characterized in that, It was established using the load aggregation power modeling method described in any one of claims 1-4.

Images