Building cluster energy consumption dynamic optimization method and system based on big data and AI prediction

Abstract

The invention discloses a building cluster energy consumption dynamic optimization method and system based on big data and AI prediction, and relates to the technical field of building energy management, and the method comprises the steps: taking each building as a building node, taking a building equipment physical relation as an edge, abstracting a building cluster into a building graph structure, and constructing a building energy consumption coupling graph; inputting the building energy consumption coupling graph and the building operation state data packet into a graph convolutional network model to generate a building energy consumption adjustment prediction packet; synchronously inputting the building energy consumption adjustment prediction packet and the building operation state data packet into a physical constraint neural network, calculating local sensitivity through automatic differentiation, constructing and solving a control dynamic optimization problem on the local sensitivity, and outputting a future control sequence; according to the invention, effective connection between energy consumption prediction and dynamic control is realized, and collaboration and implementability of building cluster energy consumption optimization are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building energy management, and particularly to a building cluster energy consumption dynamic optimization method and system based on big data and AI prediction. BACKGROUND

[0002] With the continuous advancement of urbanization and the continuous deepening of smart city construction, the proportion of building energy consumption in the overall energy consumption structure of the city has increased year by year, especially in large park, business district and city complex building cluster scenarios, building energy consumption presents the development trend of scale, coupling and dynamic. In order to improve energy utilization efficiency, reduce operating costs and achieve the goal of energy saving and emission reduction, the related technical field gradually evolves from traditional single building energy consumption management to building cluster energy consumption management based on informatization and digitization. With the development of big data and artificial intelligence technology, some technical solutions begin to introduce machine learning models to predict energy consumption data and develop energy consumption adjustment strategies based on the prediction results, thereby realizing intelligent management of building energy consumption to a certain extent.

[0003] In the existing building cluster energy consumption management and optimization technology, in view of the complex coupling relationship formed among buildings in terms of energy supply, equipment operation and environmental conditions, related technologies are mostly modeled and optimized in the form of independent buildings or simple aggregation, which is difficult to fully depict the structural association between building nodes and its influence on the evolution of overall energy consumption at the model level. When predicting energy consumption and making scheduling decisions, the existing technology often focuses on data-driven statistical correlation analysis, but lacks consideration of physical constraints, equipment operation mechanism and dynamic sensitivity between control variables involved in the energy consumption adjustment process, resulting in a certain deviation between the prediction results and the actual executable control strategy. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a building cluster energy consumption dynamic optimization method based on big data and AI prediction to solve the problems of difficult to depict building energy consumption coupling relationship and difficult to transform executable dynamic control strategy.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a building cluster energy consumption dynamic optimization method based on big data and AI prediction, which comprises collecting building energy consumption data, building equipment operation data and building environmental parameters, and obtaining a building operation state data packet. the building energy consumption coupling graph is constructed by regarding each building as a building node, regarding the physical relationship of the building equipment as an edge, and abstracting the building cluster into a building graph structure; The building energy consumption adjustment prediction package and the building operation state data package are input into a physical constraint neural network, local sensitivity is calculated through automatic differentiation, a control dynamic optimization problem is constructed and solved on the local sensitivity, and a future control sequence is output; The optimal control amount in the future control sequence is converted into a device control instruction, and is distributed to each building control terminal, and the execution state data and real-time measurement data of each building control terminal are collected to generate a state feedback data package.

[0007] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction, the building operation state data package is obtained by collecting building energy consumption data, building equipment operation data and building environment parameters, and the specific steps are as follows, The power consumption information, heating, ventilation and air conditioning consumption information, lighting consumption information and other energy consumption information in the building are collected to form building energy consumption data, and the on-off state, operation time, fault alarm and equipment performance index of the equipment in the building are collected to form building equipment operation data; Temperature, humidity, carbon dioxide concentration and light intensity are collected to form building environment parameters; The building energy consumption data, building equipment operation data and building environment parameters are packaged into a preliminary building data set through a formatting rule, and a time stamp is added to form a building operation state data package.

[0008] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction, the building energy consumption coupling graph is constructed by regarding each building as a building node, regarding the physical relationship of the building equipment as an edge, and abstracting the building cluster into a building graph structure, and the specific steps are as follows, A unique node identifier is created for each building, and building basic attribute data of each building is collected, and each building is defined as a building node; The device connection information between building equipment is identified to obtain the physical relationship of building equipment, the physical relationship of building equipment is defined as the edge between buildings, and the edge is weighted according to the energy transmission strength to form a building node edge; The building nodes are connected through the building node edge to form a building graph structure, the building equipment operation data is associated with the building node in the building graph structure, and the building environment parameters are taken as additional features of each building node, and the building energy consumption coupling graph is constructed.

[0009] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction provided in the application, the building energy consumption coupling graph and the building operation state data packet are input into a graph convolution network model to generate a building energy consumption adjustment prediction packet, and the specific steps are as follows, The building operation state data packet is split according to a preset field order, and each field is filled with missing values and removed from abnormal values to obtain a standard operation state data packet, and the standard operation state data packet is spliced into a feature vector of a corresponding building node according to a fixed dimension to generate a building node feature matrix. According to the numbering order of the building nodes in the building energy consumption coupling graph, the energy consumption coupling weights of the building node pairs that have physical connections are filled into the corresponding matrix positions, and the building node pairs that do not have physical connections are filled with zeros to generate a building adjacency matrix. The building node feature matrix and the building adjacency matrix are input into the graph convolution network model, the neighborhood information of the building node feature matrix is aggregated and updated according to the building adjacency matrix, the energy consumption coupling influence between buildings is fused to generate energy consumption adjustment prediction values of each building, and the energy consumption adjustment prediction values are packaged into a building energy consumption adjustment prediction packet.

[0010] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction provided in the application, the energy consumption coupling influence between buildings is obtained by guiding the graph convolution network model to perform neighborhood aggregation on the energy consumption and operation characteristics of adjacent buildings based on the building adjacency matrix.

[0011] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction provided in the application, the local sensitivity is calculated by automatic differentiation, and the specific steps are as follows, The building energy consumption adjustment prediction packet and the building operation state data packet are matched one by one according to the node identifier, and the two types of building data are field expanded and sequentially rearranged in units of building nodes, and are spliced into a joint input vector through a fixed field order. The joint input vector is input into a physical constraint neural network, the joint input vector is received by the input layer, the nonlinear relationship between the control variables and the building operation state is represented by the feature mapping layer, the mapping result is input into the constraint fusion layer to apply the building device operation boundary and the change rate constraint, and the candidate control response result is generated through the output layer. The sensitivity relationship of the influence degree of the change of each control input variable on the candidate control response result is calculated by using the automatic differentiation mechanism, and the sensitivity relationship is used as the local sensitivity of the corresponding control variable.

[0012] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction provided in the application, the future control sequence is output, and the specific steps are as follows, Taking a building energy consumption regulation target as an optimization target, and combining a building equipment operation boundary and a safety boundary, a control dynamic optimization problem is constructed under local sensitivity guidance; The control dynamic optimization problem is solved by iterative optimization to update step by step, and the optimal control quantity is output, and the optimal control quantity is arranged in time sequence into a future control sequence.

[0013] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction, the optimal control quantity in the future control sequence is converted into a device control instruction, which means reading the optimal control quantity corresponding to the current control time in the future control sequence, and splitting the optimal control quantity into device target control values according to the node identifier and the device identifier, and converting each device target control value into a device executable control instruction field according to the device control mapping table.

[0014] As a preferred scheme of the building cluster energy consumption dynamic optimization method based on big data and AI prediction, the state feedback data packet is generated, and the specific steps are as follows, The device control instruction is sent to the corresponding building control terminal through the building control network, each building control terminal reads the execution result of the controlled device, collects the execution state data and real-time measurement data; The execution state data and real-time measurement data are gathered according to the node identifier, and are formatted and packaged according to the preset field order, and are attached with a unified timestamp and a data integrity mark to generate a state feedback data packet.

[0015] In a second aspect, the present application provides a building cluster energy consumption dynamic optimization system based on big data and AI prediction, comprising a data acquisition module for acquiring building energy consumption data, building equipment operation data and environmental parameters to obtain a building operation state data packet; The atlas prediction module regards each building as a building node, takes the physical relationship of building equipment as an edge, abstracts the building cluster into a building graph structure, and constructs a building energy consumption coupling atlas; the building energy consumption coupling atlas and the building operation state data packet are input into a graph convolution network model to generate a building energy consumption regulation prediction packet; The optimization control module inputs the building energy consumption regulation prediction packet and the building operation state data packet into a physical constraint neural network, calculates the local sensitivity by automatic differentiation, constructs and solves a control dynamic optimization problem on the local sensitivity, and outputs a future control sequence; The execution feedback module converts the optimal control quantity in the future control sequence into a device control instruction, and sends it to each building control terminal, while collecting the execution state data and real-time measurement data of each building control terminal to generate a state feedback data packet.

[0016] The beneficial effects of this invention are as follows: by constructing a building energy consumption coupling map and introducing a graph convolutional network, collaborative prediction of building cluster energy consumption is achieved. By combining a physical constraint neural network and automatic differential local sensitivity to guide dynamic optimization, an executable future control sequence is generated. Through closed-loop feedback, the control effect is continuously corrected, thus realizing the effective connection between energy consumption prediction and dynamic control, and improving the synergy and feasibility of building cluster energy consumption optimization. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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] Fig. 1 This is a flowchart of a method for dynamic optimization of building cluster energy consumption based on big data and AI prediction.

[0019] Fig. 2 This is a schematic diagram of a building cluster energy consumption dynamic optimization system based on big data and AI prediction.

[0020] Fig. 3 This is a flowchart of building energy consumption regulation prediction based on a graph convolutional network model.

[0021] Fig. 4 This is a flowchart of the local sensitivity calculation based on a physically constrained neural network. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figs. 1-4This is one embodiment of the present invention, which provides a method for dynamic optimization of building cluster energy consumption based on big data and AI prediction, including the following steps: S1. Collect building energy consumption data, building equipment operation data and building environmental parameters to obtain building operation status data packets.

[0026] S1.1 Collect information on electricity consumption, HVAC consumption, lighting consumption, and other energy consumption in the building to form building energy consumption data. At the same time, collect information on the on / off status, running time, fault alarms, and equipment performance indicators of the equipment in the building to form building equipment operation data.

[0027] Furthermore, within the building, energy consumption information is collected from electricity metering points, HVAC energy consumption collection points, lighting energy consumption collection points, and other energy consumption collection points. This data undergoes unified dimensional conversion and field standardization to form building energy consumption data. Simultaneously, equipment on / off status, operating time, fault alarm information, and equipment performance indicators are obtained from the building equipment operation interface according to equipment identifiers. This data is then time-aligned and field-standardized to form structured building equipment operation data.

[0028] S1.2. Collect temperature, humidity, carbon dioxide concentration and light intensity to form building environment parameters.

[0029] Furthermore, temperature, humidity, carbon dioxide concentration, and light intensity data are collected from corresponding environmental parameter collection points within the building. The environmental data is then time-aligned and field-standardized, and the data is aggregated to form building environmental parameters after unifying the data format.

[0030] S1.3. Pack the building energy consumption data, building equipment operation data and building environmental parameters into a preliminary building dataset according to the formatting rules, and add a timestamp to form a building operation status data packet.

[0031] Furthermore, the building energy consumption data, building equipment operation data, and building environmental parameters are standardized in terms of field naming, data type, and unit of measurement. The common key fields (node ​​identifier, equipment identifier, and collection time) of the three types of building data are also validated for consistency. The building energy consumption data field, building equipment operation data field, and building environmental parameter field are concatenated in a fixed field order according to the formatting rules to generate a preliminary building dataset. A unified timestamp field is written to each record in the preliminary building dataset, and a data source mark and field integrity mark are generated. The preliminary building dataset with timestamp and mark fields is then structured and encapsulated to form a building operation status data package.

[0032] It should be noted that formatting rules are used to ensure that data from different sources have a consistent data structure and semantic meaning before being spliced ​​together. Formatting rules are set by sorting out the original data structure of building energy consumption data, building equipment operation data and building environmental parameters, and combining them with the target data structure requirements.

[0033] The fixed field order is defined based on the order of the field list in the target building's operating status data package, which includes building energy consumption data, building equipment operation data, and building environmental parameters.

[0034] S2. Treat each building as a building node and the physical relationships of building equipment as edges. Abstract the building cluster into a building graph structure and construct a building energy consumption coupling graph. Input the building energy consumption coupling graph and the building operation status data packet into the graph convolutional network model to generate a building energy consumption adjustment prediction packet.

[0035] S2.1 Create a unique node identifier for each building and collect the basic building attribute data for each building, defining each building as a building node.

[0036] Furthermore, the building name, building geographic location code, and building number of each building are combined and coded. After the combination result is uniquely verified, a unique node identifier is generated. The node identifier is then bound and registered with the basic building attribute data such as building name, building geographic location, building area, and building function type. Using the node identifier as an index, each building is mapped to a building node and written into the building node set.

[0037] S2.2 Identify the equipment connection information between building equipment to obtain the physical relationship between building equipment, define the physical relationship between building equipment as the edge between buildings, and assign weights to the edge according to the energy transmission intensity to form building node edge.

[0038] Furthermore, the building equipment list is traversed item by item, and the equipment connection information is read based on the equipment identifier. The connection pairs of "upstream equipment identifier - downstream equipment identifier" are extracted to determine the physical relationship of the building equipment. The physical relationship of the building equipment is mapped to "node identifier pairs" according to the node identifier, and the node identifier pairs are defined as edges between buildings to generate a set of building node edges. The edge weights are calculated based on the energy transmission intensity corresponding to the connection pairs, and the edge weights are written into the set of building node edges to obtain weighted building node edges.

[0039] S2.3 Connect the building nodes through building node edges to form a building graph structure. In the building graph structure, associate the building equipment operation data with the building nodes and use the building environment parameters as additional features of each building node to construct a building energy consumption coupling graph.

[0040] Furthermore, based on a weighted set of building node edges, the building node set is connected according to node identifiers. For each building node edge, the starting node identifier, ending node identifier, and edge weight are read. The starting and ending building nodes are written into the same connection relationship record, and the edge weight field is retained, thus forming a building graph structure. In the building graph structure, building equipment operation data is aggregated and processed according to node identifiers. The node identifier corresponding to the equipment identifier in the building equipment operation data is used as an index. The building equipment operation data is written into the node attribute field of the corresponding building node, and the field name is consistent and the time is aligned, realizing a one-to-one binding between building equipment operation data and building nodes. Building environmental parameters are matched according to node identifiers. The building environmental parameters are written into the additional feature field of each building node, and the field order is fixed and the time is aligned. After the unified encapsulation of the building graph structure, building equipment operation data node attributes, and building environmental parameter additional features, the building energy consumption coupling map is output.

[0041] S2.4. Split the building operation status data packet according to the preset field order, fill in missing values ​​and remove outliers in each field to obtain the standard operation status data packet, and concatenate the standard operation status data packet into the feature vector of the corresponding building node according to a fixed dimension to generate the building node feature matrix.

[0042] Furthermore, the building operation status data packet is split into fields according to a preset field order, extracting the building energy consumption data field, building equipment operation data field, and building environmental parameter field one by one, and performing data type validation and unit consistency processing on each field; then, missing value completion processing and outlier removal processing are performed on each field, where outliers are identified by value range validation and time series mutation detection, and missing values ​​are completed again after removal to obtain the standard operation status data packet; the standard operation status data packet is collected according to node identifier, and the fields of the standard operation status data packet are concatenated into the feature vector of the corresponding building node according to a fixed dimension and fixed field order, and the feature vectors are stacked according to the node identifier order to generate the building node feature matrix.

[0043] It should be noted that the field order is defined by the arrangement of building energy consumption data, building equipment operation data and building environmental parameters in the feature vector in the building operation status data packet and then solidified during the formatting stage.

[0044] S2.5. According to the numbering order of building nodes in the building energy consumption coupling diagram, fill in the energy consumption coupling weight in the corresponding matrix position for building node pairs with physical equipment connections, and fill in zero for building node pairs without physical equipment connections, to generate a building adjacency matrix.

[0045] Furthermore, the number of building nodes is determined based on the set of building nodes in the building energy consumption coupling graph. Row index sequences and column index sequences are generated according to the numbering order of building nodes in the building energy consumption coupling graph. An initialization matrix with the same number of building nodes is created based on the row index sequences and column index sequences. The set of weighted building node edges is traversed. For each building node edge, the starting building node identifier, the ending building node identifier, and the energy consumption coupling weight are read. The starting building node number and the ending building node number are determined through the correspondence between the building node identifier and the building node number. The energy consumption coupling weight is written into the matrix position corresponding to the row of the starting building node number and the column of the ending building node number in the initialization matrix. For building node pairs that do not appear in the set of weighted building node edges, the zero value of the corresponding matrix position in the initialization matrix is ​​retained unchanged, thus obtaining the building adjacency matrix.

[0046] It should be noted that the energy consumption coupling weight is calculated by reading the energy transmission intensity index corresponding to the physical connection of the equipment for building node pairs with physical equipment connections, and then performing dimension unification and numerical normalization processing on the energy transmission intensity index.

[0047] S2.6 Input the building node feature matrix and building adjacency matrix into the graph convolutional network model. Based on the building adjacency matrix, perform neighborhood information aggregation and feature update on the building node feature matrix. Integrate the energy consumption coupling effects between buildings to generate the energy consumption regulation prediction value for each building and encapsulate it into a building energy consumption regulation prediction package.

[0048] Furthermore, the building node feature matrix and building adjacency matrix are aligned according to the building node number order and then input into the graph convolutional network model. Based on the energy consumption coupling weights corresponding to each row in the building adjacency matrix, the graph convolutional network model performs weighted summation and aggregation on the feature vectors of adjacent building nodes in the building node feature matrix to obtain the neighborhood aggregated feature vector of each building node. The graph convolutional network model performs combination operations on the neighborhood aggregated feature vector and the feature vector of the building node itself, and completes feature update through nonlinear mapping, outputting the updated feature representation of the building node. The updated feature representation of the building node is mapped to the energy consumption adjustment prediction value corresponding to each building, and the energy consumption adjustment prediction value is encapsulated according to the building node identifier and timestamp field to generate the building energy consumption adjustment prediction package.

[0049] It should be noted that the energy consumption coupling effect between buildings is obtained by neighborhood aggregation of energy consumption and operation characteristics of adjacent buildings based on the building adjacency matrix guided graph convolutional network model.

[0050] The training graph convolutional network model uses the building node feature matrix formed by imputing missing values ​​and removing outliers from historical building operation status data packets as input features, the building adjacency matrix aligned by building node identifiers within the same time window as structural input, and historical energy consumption adjustment results or historical actual energy consumption change results as supervision targets. The model outputs energy consumption adjustment prediction values ​​through forward computation, calculates the loss function between the energy consumption adjustment prediction values ​​and the supervision targets, and updates the graph convolutional network model parameters through backpropagation. This process is repeated iteratively until the loss function converges.

[0051] S3. Synchronously input the building energy consumption regulation prediction package and the building operation status data package into the physical constraint neural network, calculate the local sensitivity through automatic differentiation, construct and solve the control dynamic optimization problem on the local sensitivity, and output the future control sequence.

[0052] It should be noted that, with a fully connected neural network as the main body, a multi-layer fully connected structure is used to establish a nonlinear mapping relationship between the control input and the building's operating state. In the network output stage, the operating boundary, rate of change and safety constraints of the building equipment are combined to perform constraint pruning or constraint mapping on the control results output by the fully connected network, so that the output of the fully connected neural network always meets the physical feasibility of the building, forming a neural network with physical constraint capabilities, namely, a physical constraint neural network. The hierarchical structure of the physical constraint neural network consists of an input layer, a feature mapping layer, a constraint fusion layer, and an output layer. The input layer receives the joint input vector formed by concatenating the building energy consumption regulation prediction packet and the building operation status packet. The feature mapping layer is composed of multiple fully connected networks and is used to learn the nonlinear relationship between the control variables and the building operation status. The constraint fusion layer applies physical constraints such as the building equipment operation boundary and change rate to the network output space. The output layer generates the control response result that satisfies the physical constraint conditions. The layers adopt a feedforward connection structure. The input layer passes the joint input vector to the feature mapping layer. The feature mapping layer maps the features to a high-dimensional representation layer by layer and then passes it to the constraint fusion layer. The constraint fusion layer performs constraint processing on the feature mapping result and passes the constrained result to the output layer, thus forming a continuous, differentiable and physically constrained network computing link from input to output. Using historical building operation status data and corresponding actual control results as training samples, control output is generated through forward calculation, and the energy consumption optimization objective and the degree of violation of physical constraints are considered in the loss function during the backpropagation process. This allows the physical constraint neural network parameters to automatically meet physical constraints such as equipment operation boundaries and change rates while minimizing prediction errors, thus obtaining a physical constraint neural network that conforms to physical laws and has control effects.

[0053] S3.1 Match the building energy consumption regulation prediction package and the building operation status data package one by one according to the node identifier, and expand and rearrange the fields of the two types of building data in units of building nodes, and concatenate them into a joint input vector by fixing the field order.

[0054] Furthermore, the building node identifier field and timestamp field are extracted from the building energy consumption regulation prediction package and the building operation status data package, respectively. The two types of building data are matched and verified one-to-one according to the building node identifier field to form building node-level records. The building node-level records are expanded with the building node identifier as the index, and the building energy consumption regulation prediction value field and the building energy consumption data field, building equipment operation data field and building environmental parameter field in the building operation status data package are expanded into the same level field sequence. Then, the same level field sequence is rearranged in a fixed field order and the consistency of field name and field type is verified. The field values ​​are concatenated in a fixed field order to form the joint input vector of the corresponding building node.

[0055] S3.2. Input the joint input vector into the physical constraint neural network. The input layer receives the joint input vector. After the feature mapping layer characterizes the nonlinear relationship between the control variables and the building operation state, the mapping result is input into the constraint fusion layer to apply constraints on the building equipment operation boundary and rate of change. The output layer generates candidate control response results.

[0056] Furthermore, after the joint input vector is input into the physical constraint neural network, the input layer performs dimension verification and field position consistency verification on the joint input vector, and writes the joint input vector into the physical constraint neural network computation link according to the input layer vector format; the feature mapping layer performs multi-layer fully connected linear transformation and nonlinear activation operation on the joint input vector, and obtains a high-dimensional representation result of the nonlinear relationship between the control variables and the building operation state through layer-by-layer mapping, and passes the high-dimensional representation result to the constraint fusion layer; the constraint fusion layer reads the building equipment operation boundary and change rate constraint, performs boundary pruning and change rate limiting processing on the control response component in the high-dimensional representation result, and generates a constrained representation result that satisfies the building equipment operation boundary and change rate constraint; the output layer performs output mapping operation on the constrained representation result and generates candidate control response results, which contain control output fields that correspond one-to-one with the control variables of the controlled equipment.

[0057] It should be noted that the operating boundaries of building equipment are obtained by reading the range of operating parameters registered by various building equipment during the configuration phase. The range of operating parameters includes the minimum and maximum operating values ​​allowed by the equipment, which are used to limit the variation of control output fields within the physically feasible range. The rate of change constraint is determined by statistically analyzing the changes in control output between adjacent control times in the equipment's historical control records and combining this with the equipment's allowable adjustment speed requirements. It is used to limit the maximum range of change in the control output field between adjacent control times.

[0058] S3.3 Calculate the sensitivity relationship of the influence of changes in each control input variable on the candidate control response using the automatic differentiation mechanism, and use the sensitivity relationship as the local sensitivity of the corresponding control variable.

[0059] Furthermore, since the differentiable computation path from the joint input vector to the candidate control response remains unchanged, an automatic differentiation mechanism is enabled to perform partial derivative calculations on the candidate control response with respect to each control input variable in the joint input vector. By calculating the gradient value of the candidate control response with respect to the change of each control input variable one by one, the influence of the change of control input variables on the candidate control response is quantified. The gradient values ​​are then organized and aggregated according to the name of the control input variable to form a local sensitivity that corresponds one-to-one with the control input variable.

[0060] The gradient value of the candidate control response with respect to the change of each control input variable is calculated one by one, and the expression is: ; In the formula, It is the candidate control response result with respect to the control input variable. The gradient value is used to characterize the control input variable. The degree of overall impact of changes on the candidate control response results; It is the first control output field in the candidate control response results. For control input variables The partial derivatives are used to characterize the control input variables. For control output fields The degree of sensitivity; It is the second control output field in the candidate control response results. For control input variables The partial derivatives; It is the first in the candidate control response results One control output field For control input variables The partial derivatives; It is the first control output field in the candidate control response result, corresponding to the control output value of a controlled device or control channel; It is the second control output field in the candidate control response result; It is the first in the candidate control response results One control output field; It is the first in the joint input vector Each control input variable corresponds to an input field that participates in control decisions; It is a vector transpose operation, used to arrange a sequence of partial derivatives into a column vector form; It is the index number of the control input variable in the joint input vector, used to distinguish the position of different control input variables in the joint input vector; It represents the total number of control output fields in the candidate control response results, and is used to represent the dimension of the candidate control response result vector; It should be noted that the automatic differentiation mechanism is a method that, while maintaining the accuracy of the calculation process, applies the chain rule layer by layer to the basic operations in the differentiable computation chain to accurately calculate the partial derivatives of the candidate control response with respect to the control input variable. This method is used to obtain gradient information on the degree of influence of changes in the control input variable on the candidate control response.

[0061] The automatic differentiation mechanism is obtained by representing the computation process from the joint input vector to the candidate control response as a computation graph composed of differentiable operators, and performing gradient backpropagation layer by layer on the computation graph based on the chain rule after the forward computation is completed.

[0062] S3.4. Taking the building energy consumption regulation target as the optimization target, and combining the building equipment operation boundary and safety boundary, construct a control dynamic optimization problem under the guidance of local sensitivity.

[0063] Furthermore, the optimization objective of the control dynamic optimization problem is determined by the building energy consumption regulation target, and the corresponding control output field is selected as the optimization variable from the candidate control response results; the gradient information of the control input variable with respect to the candidate control response results in the local sensitivity is read, and the gradient information is used to guide the update direction of the optimization variable; the constraints are extracted from the building equipment operating boundary, rate of change constraint and safety boundary and written into the optimization constraint set to construct the control dynamic optimization problem.

[0064] It should be noted that the building energy consumption regulation target is a quantitative target for the energy consumption operation status that the building is expected to achieve in the future control cycle, and is used to clarify the direction and magnitude of energy consumption adjustment during the control optimization process; The building energy consumption regulation target is calculated by combining the energy consumption regulation prediction value in the building energy consumption regulation prediction package with the current energy consumption level reflected in the building operation status data package, and is determined based on the building equipment operation boundary and safety boundary.

[0065] Safety boundaries are obtained by reading safety parameter configurations related to the operational safety of building equipment, and are used to prevent control outputs from causing the equipment's operating state to exceed the safe allowable range.

[0066] S3.5. The dynamic optimization problem of control is solved by iterative optimization to update the output step by step, and the optimal control quantity is then arranged into a future control sequence in chronological order.

[0067] Furthermore, using the candidate control response results as the initial optimization variables for the control dynamic optimization problem, the objective function value is calculated, and constraint satisfaction is verified for the building equipment operating boundary, rate of change constraint, and safety boundary. Based on the gradient information provided by local sensitivity, the update direction and update step size of the optimization variables are calculated, and an iterative update is performed on the optimization variables. Boundary pruning and rate of change restriction processing are then performed on the updated optimization variables to ensure constraint satisfaction. The objective function calculation, constraint satisfaction verification, and iterative update are repeated until the convergence condition is met or the iteration termination condition is reached, and the converged optimization variables are obtained as the optimal control variables. The optimal control variables are arranged time-by-time according to the control time index and encapsulated in the order of the timestamp field to form the future control sequence.

[0068] The objective function value is calculated as follows: ; In the formula, In the initial optimization variables The optimized objective function value calculated under the given conditions is used to measure the overall building energy consumption deviation under the influence of candidate control response results; In the initial optimization variables Under the action, the first Predicted energy consumption results for each building node; It is the first Building energy consumption regulation targets corresponding to each building node; It is the building node index number, used to distinguish different building nodes; It is the total number of building nodes participating in energy consumption regulation.

[0069] S4. Convert the optimal control quantity in the future control sequence into equipment control commands and send them to each building control terminal. At the same time, collect the execution status data and real-time measurement data of each building control terminal and generate a status feedback data packet.

[0070] S4.1 Read the optimal control quantity corresponding to the current control time in the future control sequence, and split the optimal control quantity into device target control values ​​according to the node identifier and device identifier. Based on the device control mapping table, convert each device target control value into a control instruction field that can be executed by the device, and generate device control instructions.

[0071] Furthermore, the optimal control quantity record corresponding to the current control time index is extracted from the future control sequence, and the optimal control quantity record is verified. The optimal control quantity record is grouped and split according to the node identifier and device identifier, and the control output field corresponding to each group of node identifier and device identifier is parsed into the device target control value. The mapping relationship of control field matching the device identifier and the instruction value rule are read from the device control mapping table, and the unit conversion and value range conversion are performed on the device target control value to obtain the control instruction field that the device can execute. The device identifier, node identifier, control instruction field and current control time timestamp are encapsulated in the field order to generate the device control instruction.

[0072] It should be noted that the device control mapping table is a collection of mapping information used to describe the correspondence between the target control values ​​of the device and the fields of the executable control instructions of the device; The equipment control mapping table is obtained by registering and organizing the control parameters, control command formats, and value rules supported by each type of building equipment during the building equipment access and configuration phase.

[0073] S4.2. The equipment control commands are sent to the corresponding building control terminals through the building control network. Each building control terminal reads the execution results of the controlled equipment and collects execution status data and real-time measurement data.

[0074] Furthermore, according to the node identifier and device identifier, the device control command is sent to the corresponding building control terminal through the building control network. After receiving the device control command, the building control terminal writes the control command into the control interface of the controlled device and triggers the device to execute. After the device completes the execution, the execution result of the controlled device is read, and execution status data including the control command execution status and fault alarm status is collected according to the collection cycle. Real-time measurement data corresponding to the control command is collected synchronously.

[0075] S4.3 Aggregate the execution status data and real-time measurement data according to node identifiers, and after formatting and packaging them according to the preset field order, attach a unified timestamp and data integrity mark to generate a status feedback data packet.

[0076] Furthermore, the execution status data and real-time measurement data are aggregated by node identifier to form node-level feedback records. The node-level feedback records are then expanded, ordered, and formatted according to a preset field order. During the formatting and encapsulation process, a uniform timestamp field is added to each node-level feedback record, and a data integrity marker field is generated to form a status feedback data package.

[0077] This embodiment also provides a building cluster energy consumption dynamic optimization system based on big data and AI prediction, including: a data acquisition module, used to collect building energy consumption data, building equipment operation data and environmental parameters, and obtain building operation status data packets; The graph prediction module treats each building as a building node and uses the physical relationships of building equipment as edges to abstract the building cluster into a building graph structure, thus constructing a building energy consumption coupling graph. The building energy consumption coupling graph and building operation status data packets are input into the graph convolutional network model to generate a building energy consumption adjustment prediction packet. The optimized control module is used to synchronously input the building energy consumption regulation prediction package and the building operation status data package into the physical constraint neural network, calculate the local sensitivity through automatic differentiation, construct and solve the control dynamic optimization problem on the local sensitivity, and output the future control sequence; The execution feedback module is used to convert the optimal control quantity in the future control sequence into equipment control commands and send them to each building control terminal. At the same time, it collects the execution status data and real-time measurement data of each building control terminal and generates status feedback data packets.

[0078] In summary, this invention achieves collaborative prediction of building cluster energy consumption by constructing a building energy consumption coupling map and introducing a graph convolutional network, combining a physical constraint neural network with automatic differential local sensitivity to guide dynamic optimization solutions, generating executable future control sequences, and continuously correcting the control effect through closed-loop feedback. This effectively connects energy consumption prediction and dynamic control, improving the synergy and feasibility of building cluster energy consumption optimization.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dynamic optimization of building cluster energy consumption based on big data and AI prediction, characterized by: include, Collect building energy consumption data, building equipment operation data, and building environmental parameters to obtain building operation status data packages; Each building is treated as a building node, and the physical relationships of building equipment are used as edges. The building cluster is abstracted into a building graph structure to construct a building energy consumption coupling graph. The building energy consumption coupling map and building operation status data packet are input into the graph convolutional network model to generate a building energy consumption regulation prediction packet; The building energy consumption regulation prediction package and the building operation status data package are synchronously input into the physical constraint neural network. The local sensitivity is calculated by automatic differentiation, and the control dynamic optimization problem is constructed and solved on the local sensitivity to output the future control sequence. The optimal control quantity in the future control sequence is converted into equipment control commands and sent to each building control terminal. At the same time, the execution status data and real-time measurement data of each building control terminal are collected to generate status feedback data packets.

2. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 1, characterized in that: The specific steps for collecting building energy consumption data, building equipment operation data, and building environmental parameters to obtain building operation status data packets are as follows. Collect information on electricity consumption, HVAC consumption, lighting consumption, and other energy consumption in buildings to form building energy consumption data. At the same time, collect information on the on / off status, operating time, fault alarms, and equipment performance indicators of equipment in buildings to form building equipment operation data. Temperature, humidity, carbon dioxide concentration, and light intensity are collected to form building environment parameters; Building energy consumption data, building equipment operation data, and building environmental parameters are packaged into a preliminary building dataset according to formatting rules, and a timestamp is added to form a building operation status data package.

3. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 1, characterized in that: The method involves treating each building as a building node and using the physical relationships between building equipment as edges to abstract the building cluster into a building graph structure, thereby constructing a building energy consumption coupling graph. The specific steps are as follows. Create a unique node identifier for each building and collect basic building attribute data for each building to define each building as a building node; Identify the equipment connection information between building equipment to obtain the physical relationship of building equipment, define the physical relationship of building equipment as the edge between buildings, and assign weights to the edge according to the energy transmission intensity to form building node edge; Building nodes are connected by building node edges to form a building graph structure. Building equipment operation data are associated with building nodes in the building graph structure, and building environmental parameters are used as additional features of each building node to construct a building energy consumption coupling graph.

4. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 3, characterized in that: The steps for inputting the building energy consumption coupling map and building operation status data into the graph convolutional network model to generate a building energy consumption adjustment prediction package are as follows: The building operation status data packet is split according to the preset field order, and missing values ​​are filled and outliers are removed from each field to obtain the standard operation status data packet. The standard operation status data packet is then concatenated into the feature vector of the corresponding building node according to a fixed dimension to generate the building node feature matrix. According to the numbering order of building nodes in the building energy consumption coupling diagram, for building node pairs with physical equipment connections, energy consumption coupling weights are filled in the corresponding matrix positions, and for building node pairs without physical equipment connections, zeros are filled in to generate a building adjacency matrix. The building node feature matrix and building adjacency matrix are input into the graph convolutional network model. Based on the building adjacency matrix, the neighborhood information of the building node feature matrix is ​​aggregated and the features are updated. The energy consumption coupling effect between buildings is integrated to generate the predicted energy consumption adjustment value of each building, and then packaged into a building energy consumption adjustment prediction package.

5. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 4, characterized in that: The energy consumption coupling effect between buildings is obtained by using a building adjacency matrix guided graph convolutional network model to aggregate the energy consumption and operational characteristics of adjacent buildings in a neighborhood.

6. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 1, characterized in that: The specific steps for calculating local sensitivity through automatic differentiation are as follows: The building energy consumption regulation prediction package and the building operation status data package are matched one by one according to the node identifier. The two types of building data are expanded and rearranged in order by building node. They are then concatenated into a joint input vector by fixing the field order. The joint input vector is input into the physical constraint neural network. The input layer receives the joint input vector, and after the feature mapping layer characterizes the nonlinear relationship between the control variables and the building operation state, the mapping result is input into the constraint fusion layer to apply constraints on the building equipment operation boundary and rate of change. The output layer generates candidate control response results. The sensitivity relationship of the influence of changes in each control input variable on the candidate control response is calculated using an automatic differentiation mechanism, and the sensitivity relationship is used as the local sensitivity of the corresponding control variable.

7. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 1, characterized in that: The specific steps for outputting the future control sequence are as follows. Taking the building energy consumption regulation target as the optimization objective, and combining the building equipment operation boundary and safety boundary, a control dynamic optimization problem is constructed under the guidance of local sensitivity. The dynamic optimization problem of control is solved by iterative optimization to update the output step by step, and the optimal control quantity is then arranged into a future control sequence in chronological order.

8. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 7, characterized in that: The process of converting the optimal control quantity in the future control sequence into device control instructions refers to reading the optimal control quantity corresponding to the current control moment in the future control sequence, splitting the optimal control quantity into device target control values ​​according to the node identifier and device identifier, and converting each device target control value into a control instruction field that can be executed by the device according to the device control mapping table.

9. The method for dynamic optimization of building cluster energy consumption based on big data and AI prediction as described in claim 1, characterized in that: The specific steps for generating the status feedback data packet are as follows: Equipment control commands are sent to the corresponding building control terminals through the building control network. Each building control terminal reads the execution results of the controlled equipment and collects execution status data and real-time measurement data. The execution status data and real-time measurement data are aggregated by node identifier, formatted and packaged according to the preset field order, and then a unified timestamp and data integrity mark are attached to generate a status feedback data packet.

10. A building cluster energy consumption dynamic optimization system based on big data and AI prediction, based on the building cluster energy consumption dynamic optimization method based on big data and AI prediction as described in any one of claims 1 to 9, characterized in that: include, The data acquisition module is used to collect building energy consumption data, building equipment operation data and environmental parameters, and obtain building operation status data packets; The graph prediction module is used to treat each building as a building node, use the physical relationships of building equipment as edges, abstract the building cluster into a building graph structure, and construct a building energy consumption coupling graph. The building energy consumption coupling map and building operation status data packet are input into the graph convolutional network model to generate a building energy consumption regulation prediction packet; The optimized control module is used to synchronously input the building energy consumption regulation prediction package and the building operation status data package into the physical constraint neural network, calculate the local sensitivity through automatic differentiation, construct and solve the control dynamic optimization problem on the local sensitivity, and output the future control sequence; The execution feedback module is used to convert the optimal control quantity in the future control sequence into equipment control commands and send them to each building control terminal. At the same time, it collects the execution status data and real-time measurement data of each building control terminal and generates status feedback data packets.

Images