A method for forecasting heating load of community buildings based on system dynamics multi-agent modeling
The method addresses the challenge of static load forecasting by employing multi-agent modeling to simulate user interactions and energy consumption, resulting in precise heating load predictions for community energy planning.
Patent Information
- Application Number
- CN202210405526.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-18
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2042-04-18
AI Technical Summary
Current load forecasting methods in community energy planning are static and fail to account for the dynamic interactions between community users and buildings, leading to significant overestimation of heating loads, especially with diverse building and user types, which affects regional energy system design and integration of renewable energy measures.
A method using system dynamics multi-agent modeling to simulate user behavior and energy consumption patterns, incorporating building function clustering, travel activity chains, and equipment usage schedules to generate high-precision heating load forecasts.
Provides accurate, high-precision heating load data for community planning, enabling better design and optimization of regional energy systems by considering user diversity and behavior variability.
Smart Images
Figure CN115018116B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of load forecasting, and in particular to a method for forecasting heating load of community buildings based on system dynamics multi-agent modeling. Background Art
[0002] Regional integrated energy system planning is a basic link in the zoning planning and deployment of communities with the goal of achieving building development and construction within the region, and is an important basis for providing construction direction and overall layout for urban planning and development. However, the load forecasting method in the current community energy planning stage is mainly based on the load index method. The regional load predicted by this method is usually a static value, which is difficult to meet the time dynamic characteristics of load forecasting in energy planning. In the design stage, the building cooling and heating load is usually overestimated by more than 30%, and the error caused by this estimation method will be further amplified at the community level.
[0003] Accurately predicting community heating loads during the community energy planning stage can help guide the design of regional energy system capacity and optimize the integration of regional renewable energy measures. However, traditional bottom-up heating load forecasting methods usually use a fixed schedule as an input condition, ignoring the interaction between community users and multiple buildings in the community. When the diversity of building types and user types in the community increases, the uncertainty of user energy consumption behavior will increase, resulting in a significant overestimation of the community's peak load.
[0004] The energy consumption behavior of users at the community scale is a macro-complexity problem involving multiple micro-problems (buildings, users, equipment), and the spatial and temporal diversity that needs to be characterized is different from the building scale. At the same time, user behavior is essentially random, and even users of the same type will behave differently over time. Agent-based modeling is a simulation method in complexity science. By introducing agent-based models into heating load forecasting, the energy consumption decision-making behavior of various types of users in the community is simulated, and a high-precision building energy consumption schedule is obtained to provide support for building heating load forecasting in the community planning stage. Summary of the invention
[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a community building heating load prediction method based on system dynamics multi-agent modeling, which fully considers the influence of the region, the actual building scenario of the community and the diversity of residents' behavior, and can provide more accurate load data for the community planning stage that lacks basic heating load data.
[0006] In order to achieve the above object, the technical solution adopted by the present invention is:
[0007] A method for predicting heating load of community buildings based on system dynamics multi-agent modeling includes the following steps:
[0008] Step 1: Input the building outline data and building function POI data in the actual community to form a data set, where the data set includes community scale, community building area, number of floors of community buildings and community building functions;
[0009] Step 2: Using building functions as labels, K-means clustering is used to divide the community building outline and function dataset to construct a typical building scenario database;
[0010] Step 3: Input the travel time period function and energy consumption behavior decision probability of the community user population, including the travel behavior time on weekdays and the behavior probability of equipment regulation;
[0011] Step 4: Generate travel activity chains for various users based on the travel time period function of community users;
[0012] Step 5: Obtain the equipment usage behavior decision model of the user subject based on the user's travel activity chain and energy consumption behavior decision probability;
[0013] Step 6: Combine the user travel activity chain and the user energy consumption decision model to generate the indoor and equipment energy consumption schedule of community building users;
[0014] Step 7: Correlate the case community buildings with typical building scenarios based on the principle of similar building area;
[0015] Step 8: Input meteorological data, indoor and equipment energy consumption schedules and enclosure parameters to interrelated typical buildings for heat load simulation and calculation;
[0016] Step 9: Generate a standardized load curve based on the actual community building area and the heat load of typical buildings to obtain the heat load forecast data for each building at the community level.
[0017] The data set in step 1 is extracted from an actual community and includes building function POI data and building outline data, wherein the frequency density and type ratio are used to determine the function of the building, as shown in formulas (1) and (2).
[0018]
[0019]
[0020] Where: i represents the POI type; n i N represents the number of POIs of the i-th category in the building; i represents the total number of POIs of the i-th category in the POI dataset; FD irepresents the frequency density of the i-th type of POI in the POI dataset; CR i It represents the ratio of the frequency density of the i-th type of POI to the frequency density of all types of POIs in the building.
[0021] The clustering process in step 2 uses building area as the key modeling indicator. Different data standardization methods are used according to different building types. The Z score method is used for residential and office buildings, and the normalization method is used for office and school buildings. The silhouette coefficient is calculated for each type of building to determine the optimal number of clusters in each building area interval, as shown in formula (3).
[0022]
[0023] Where: a(i) represents the average value of the dissimilarity between vector i and other points in the same cluster, and b(i) represents the minimum value of the dissimilarity between vector i and other points in the same cluster;
[0024] In step 2, in order to speed up the simulation of high-rise buildings and the application scope of various building height scenarios in the community, the representative floor modeling method was adopted to model three representative floors: the top floor, the bottom floor, and the middle floor with multiple floors.
[0025] In step 3, a Gaussian function is used to fit the travel probability of each user in the morning, afternoon and evening periods, thereby generating a function expression for each period, as shown in Formula 4.
[0026]
[0027] Where: x is the number of minutes in the time period.
[0028] The generation of the travel activity chain in step 4 uses the JAVA compiler in AnyLogic to input the travel expressions of student users, commuting users, and home users in different time periods, thereby generating the transfer behavior probability distribution of each type of user.
[0029] The step 5 obtains the building equipment usage schedule based on the transfer behavior probability distribution and energy consumption behavior decision probability of each type of user, wherein the four main energy terminal uses in the building are subject modeled, including washing machines, air conditioners, computers and electric lights. The equipment subject usually belongs to the household subject and is controlled by the user subject.
[0030] In step 7, since the structures and architectural features of buildings of the same type are similar, but there are some differences in the actual building area and number of floors, when selecting typical buildings, they are selected from the extracted typical building database based on the similarity of the actual building area. The typical building is the average building extracted by K-means clustering.
[0031] Step 8 uses the dynamic heat balance equation to perform detailed simulation calculations on each thermal zone in the typical building. Since the air conditioning or heating area may not be completely matched when the typical building scenario library is matched with the case community building scenario, the energy load demand curve standardization in step 9 is a necessary process for generating the community load curve so that it is applicable to the case community building that is similar to the reference building in configuration but has an air conditioning or heating area different from the typical building. The standardized load calculation process is shown in equations (5), (6), and (7).
[0032]
[0033] Where: is the standardized load curve of building B under building category n in the case community; For scenario S k The load curve of a typical building B corresponding to building category n; is the air-conditioning or heating area corresponding to a typical building;
[0034]
[0035]
[0036] In step 9, from C n S k The standardized load demand curve is generated by randomly selecting the building load demand curve in the cluster The curve is composed of a specific typical building R B,n It means that the normalized load curve is multiplied by the actual building area A obtained by area association. Bx,cond To obtain the actual building energy demand curve
[0037] The method for generating standardized building load requirements in the community planning stage proposed in step 9 of the present invention has the advantage of being applicable to the community planning stage where there is a lack of detailed information on users living in the community.
[0038] Beneficial effects of the present invention:
[0039] The community heating load forecasting method based on agent modeling proposed in the present invention is a method for high-precision forecasting in the community planning stage. However, other current community load forecasting methods usually adopt a deterministic schedule. To address this problem, the present invention adopts an agent-based modeling method to randomize the transfer and energy consumption decision-making behaviors of community users with high uncertainty.
[0040] The community heating load forecasting method based on subject modeling proposed in the present invention can obtain high-precision, multi-scenario community load data, which can provide urban planners with more options in the community planning stage, and provide high-precision load forecasting results for the later community comprehensive energy system planning, selection calculation, and scheme design. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 The figure is a workflow diagram of the community heat load prediction method based on agent modeling of the present invention.
[0042] Figure 2 This is a schematic diagram of the subject hierarchical relationship in the community user travel and energy consumption behavior model proposed in the present invention.
[0043] Figure 3 This is a schematic diagram of the actual community building occupancy rate generated by the subject-based modeling method used in the present invention.
[0044] Figure 4 This is a schematic diagram of a typical daily heating load in a community generated by the community heat load prediction method based on agent modeling adopted by the present invention. DETAILED DESCRIPTION
[0045] The present invention will be further described in detail below in conjunction with the accompanying drawings.
[0046] The method of the present invention describes the energy consumption differences among users from the perspective of "differential individuals", obtains the user's energy consumption behavior timetable to provide input parameters for subsequent typical building load calculations, and combines the "bottom-up" community load forecasting workflow to achieve high-precision forecasting of community heating load data.
[0047] Figure 1 This is the workflow of the community heat load prediction method based on subject modeling proposed in the present invention. The community load prediction method based on subject modeling proposed in the present invention is in step 1, data entry and preprocessing, and the community building data set composed of community building geometry and non-geometry data is extracted through GIS. The community building geometry data is the shp format data in GIS covering the community building area and the number of building floors, and the community building non-geometry data is the POI tag data corresponding to each type of building in the community.
[0048] The building profile data in step 2 will be clustered into multiple homogeneous subsets according to building type and building area, where each building subset is represented by only one typical building with similar energy behavior, by pre-modeling multiple typical building physical models as a database for generating standardized energy demand curves.
[0049] In steps 3 and 4, the present invention divides the primary and secondary structural relationships of different building types, standard building units, user types and equipment interaction logic in the community, and uses AnyLogic to build a community building user travel activity chain and energy consumption decision model based on subject modeling. Among them, the indoor status of community users is the core of equipment energy consumption judgment. Therefore, the present invention obtains different user activity chains and specific event triggering probabilities based on actual survey data, and establishes a community user transfer model based on this.
[0050] The simulation model of community residents' travel and energy consumption decision-making behavior contains 10 types of subjects: ① community subject type; ② commuting user subject type; ③ student user subject type; ④ home user subject type; ⑤ community outside user subject type; ⑥ standard residential unit type; ⑦ office subject type; ⑧ shopping mall subject type; ⑨ classroom subject type; ⑩ equipment subject type. The hierarchical relationship between the subjects is as follows: Figure 2 shown.
[0051] Residential buildings in the community are the main energy-consuming units, so the present invention divides family members with significant differences in travel characteristics and energy-using habits in residential buildings into three categories: commuting users, student users, and home users. The main activities of community users on weekdays are sleeping, getting up, going to work, and going home, so the user's activity chain on weekdays is divided into four parts: morning, noon, afternoon, and evening.
[0052] The travel rules for each family user are as follows:
[0053] (1) Commuters: They travel during the morning rush hour on weekdays and return during the evening rush hour. Their lunch break will determine whether they take a break in the office or at home. The morning rush hour is approximately 7:00-8:00, and the evening rush hour is approximately 17:00-18:00. The overtime behavior of commuters is randomly characterized based on the extracted overtime behavior decision probability.
[0054] (2) Student users: They travel during the morning rush hour on weekdays and return during the evening rush hour. During the lunch break, they decide whether to rest in the classroom or at home. The morning rush hour is approximately 6:30-8:00, and the evening rush hour is approximately 17:30-18:00.
[0055] (3) Home users: The travel behavior judgment of users in each travel period is concentrated in the first hour of the period, such as the morning peak period is approximately 6:00-7:00, the afternoon period is concentrated between 13:00-14:00, and the evening period is concentrated between 18:00-19:00 after dinner.
[0056] The process of user subject regulating the device subject follows the principle of "judgment before implementation". When the user enters a certain state or is in a transition path between different states, the code or function will be executed to pass a specific message to the target device to control the appliance.
[0057] Before matching community buildings with typical buildings in step 7, the complex scenarios of community user transfer and energy consumption behavior are characterized by agent-based modeling methods, and the corresponding indoor and equipment energy consumption schedule profiles of community building users are generated to associate the user behavior scenarios corresponding to each building in the actual community. Finally, the standardized energy demand curves of all typical buildings are summarized, and through the repeated association process of community buildings with typical buildings, a high-resolution energy demand curve is generated at the community scale.
[0058] Example Analysis
[0059] In order to make the present invention more obvious and easy to understand, a case community is selected to apply the proposed method, and a detailed description is given as follows with the accompanying drawings. A residential community in Xi'an is selected as community γ, and the method proposed in the present invention will generate the annual energy demand of community buildings with a time resolution of 15 minutes.
[0060] All buildings in the community are connected to the district heating network. Buildings of the same type have similar structures and architectural features, but there are slight differences in the actual area and number of floors. The community building group γ is represented by a set of typical buildings βi that have been extracted through clustering, as shown in formula (1). The total area of the case community is 144,500 m 2 , consisting of 46 residential buildings, 5 office buildings, 2 commercial buildings and 3 school buildings. The correlation between community buildings and typical buildings is shown in Table 1:
[0061]
[0062] In the formula: RB represents residential buildings, OB represents office buildings, CB represents commercial buildings, and SB represents school buildings.
[0063] Table 1 Correlation between actual buildings and typical buildings in the community
[0064]
[0065]
[0066] The floor plan of the community building complex is imported into AnyLogic to obtain the corresponding case community building outline. The size of the simulated community is consistent with the actual community. The simulation time scale is 1 minute. The simulation mode adopts the random simulation mode, that is, each simulation process is an independent random process.
[0067] The average value of 100 repeated random simulation results is extracted as the schedule input data for building load simulation. There are 19 buildings in the case community associated with the typical residential building RB4, of which RB4 uses the actual community building occupancy rate generated by the agent-based modeling method as shown in Figure 3 The model essentially represents the random process of user behavior through logic, so the reasons for the changes in the room rate at each event node are marked in the figure.
[0068] The hourly heat load on a typical day (January 21) during the actual community heating season is as follows: Figure 4 As shown. The typical daily peak load is used to characterize the highest load level of the heating load of each building during the heating season. The community peak heat load under different heating modes ranges from 3102.24kW to 3140.75kW. The community peak heat load occurs mainly in the morning period of 6:30-7:30, mainly due to the load transfer caused by user transfer activities in community residential and non-residential buildings (going to work, school, shopping, etc.). The heat load of residential buildings decreases due to the decrease in user occupancy rate, the heat load of office buildings increases due to the increase in user occupancy rate, and the heat load of commercial and school buildings increases due to the arrival of a fixed time period in the morning.
Claims
1. A method for predicting heating load of community buildings based on system dynamics multi-agent modeling, characterized in that: The steps include: Step 1: Input the case community building outline and function dataset, including community size, community building area and building function parameters; Step 2: Using building functions as labels, K-means clustering is used to divide the community building outline and function dataset to construct a typical building scenario database; Step 3: Input the travel time period function and energy consumption behavior decision probability of the community user population, including the travel behavior time on weekdays and the behavior probability of equipment regulation; Step 4: Generate travel activity chains for various users based on the travel time period function of community users; Step 5: Obtain the user energy consumption decision model of the user subject based on the user travel activity chain and energy consumption behavior decision probability; Step 6: Combine the user travel activity chain and the user energy consumption decision model to generate the indoor and equipment energy consumption schedule of community building users; Step 7: Correlate the case community buildings with typical building scenarios based on the principle of similar building area; Step 8: Input meteorological data, indoor and equipment energy consumption schedules and enclosure parameters to interrelated typical buildings for heat load simulation and calculation; Step 9: Generate a standardized load curve by combining the actual community building area and the heat load of typical buildings to obtain the heat load forecast data of each building in the community level; The step 8 uses the dynamic heat balance equation to perform detailed simulation calculations on each thermal zone in the typical building. Since the air conditioning or heating area may not be completely matched when the typical building scenario library matches the case community building scenario, the energy load demand curve in step 9 is standardized as a necessary process for generating the community load curve so that it is applicable to the case community building that is similar to the reference building in configuration but has an air conditioning or heating area different from the typical building. The standardized load calculation process is shown in equations (5), (6), and (7); Where: is the standardized load curve of building B under building category n in the case community; For scenario S k The load curve of a typical building corresponding to building category n; is the air-conditioning or heating area corresponding to a typical building; In step 9, from C n S k The standardized load demand curve is generated by randomly selecting the building load demand curve in the cluster The curve is composed of a specific typical building R B,n It means that the normalized load curve is multiplied by the actual building area A obtained by area association. Bx,cond To obtain the actual building energy demand curve 2. According to claim 1, a community building heating load prediction method based on system dynamics multi-agent modeling is characterized in that: The data set in step 1 is extracted from an actual community and includes building function POI data and building outline data, wherein the frequency density and type ratio are used to determine the function of the building, as shown in formulas (1) and (2); Where: i represents the POI type; n i N represents the number of POIs of the i-th category in the building; i represents the total number of POIs of the i-th category in the POI dataset; FD i represents the frequency density of the i-th type of POI in the POI dataset; CR i It represents the ratio of the frequency density of the i-th type of POI to the frequency density of all types of POIs in the building.
3. According to claim 1, a community building heating load prediction method based on system dynamics multi-agent modeling is characterized in that: The clustering process in step 2 uses building area as the key modeling indicator, and adopts different data standardization methods according to different building types. The Z score method is used for residential and office buildings, and the normalization method is used for office and school buildings. The silhouette coefficient is calculated for each type of building to determine the optimal number of clusters in each building area interval, as shown in formula (3); Where: a(i) represents the average value of the dissimilarity between vector i and other points in the same cluster, and b(i) represents the minimum value of the dissimilarity between vector i and other points in the same cluster; In step 2, in order to speed up the simulation of high-rise buildings and the application scope of various building height scenarios in the community, the representative floor modeling method was adopted to model three representative floors: the top floor, the bottom floor, and the middle floor with multiple floors.
4. The method for predicting heating load of community buildings based on system dynamics multi-agent modeling according to claim 1 is characterized in that: In step 3, a Gaussian function is used to fit the travel probability of each user in the morning, afternoon and evening periods, thereby generating a function expression for each period, as shown in Formula 4; Where: x is the number of minutes in the time period.
5. The method for predicting heating load of community buildings based on system dynamics multi-agent modeling according to claim 1 is characterized in that: The generation of the travel activity chain in step 4 uses the JAVA compiler in AnyLogic to input the travel expressions of student users, commuting users, and home users in different time periods, thereby generating the transfer behavior probability distribution of each type of user.
6. The method for predicting heating load of community buildings based on system dynamics multi-agent modeling according to claim 1 is characterized in that: The step 5 obtains the building equipment usage schedule based on the transfer behavior probability distribution and energy consumption behavior decision probability of each type of user, wherein the four main energy terminal uses in the building are subject modeled, including washing machines, air conditioners, computers and electric lights. The equipment subject usually belongs to the household subject and is controlled by the user subject.
7. The method for predicting heating load of community buildings based on system dynamics multi-agent modeling according to claim 1 is characterized in that: In step 7, since the structures and architectural features of buildings of the same type are similar, but there are some differences in the actual building area and number of floors, when selecting typical buildings, they are selected from the extracted typical building database based on the similarity of the actual building area. The typical building is the average building extracted by K-means clustering.