Building energy consumption collaborative optimization method and system based on mixed model and correlation analysis

Abstract

The application relates to the technical field of building energy management, and provides a building energy consumption collaborative optimization method and system based on a hybrid model and correlation analysis, which comprises the following steps: deploying a message middleware between a data acquisition module and a hybrid prediction model; calculating a joint fluctuation index and using the index to control the hybrid prediction model to switch between a simplified reasoning state and a complete reasoning state; constructing an energy consumption correlation directed graph containing a phase change energy storage unit; and determining an affected area and a replaceable energy supply path according to the directed graph; adopting a pulse type collaborative cycle, performing global correlation optimization in a long cycle, locally adjusting by each area controller in a short cycle, and triggering instant light coordination only when the deviation is out of limit or the phase change energy storage state reaches the boundary. The application solves the problems in the prior art, such as strong coupling between reasoning time and control cycle of a prediction model in building energy efficiency control, easy triggering of global degradation by local faults, and too frequent global collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of building energy management technology, and in particular to a method and system for collaborative optimization of building energy consumption based on hybrid models and correlation analysis. Background Technology

[0002] Buildings incorporating phase change energy storage units can shift heat load over time by absorbing and releasing latent heat during the phase transition process of phase change materials, providing flexible resources for HVAC systems and regional energy efficiency scheduling. Existing building energy efficiency control typically collects data on energy consumption, temperature, flow rate, and equipment status, uses predictive models to estimate future loads, and issues control commands to air conditioning terminals, valves, circulating pumps, or heat pumps through optimization algorithms.

[0003] In high-density building load scenarios, the prediction model often has a timing mismatch with the control cycle, and the state of charge of the phase change energy storage unit will change the available energy supply boundary. Therefore, it is necessary to integrate data buffering, prediction validity period, state switching boundary, fault isolation path and hierarchical collaborative control into the same technical link.

[0004] For phase change energy storage buildings, simply combining message middleware, hybrid neural network prediction, or phase change material temperature regulation in parallel is insufficient to fully demonstrate the clarity, completeness, and distinctiveness of the control link. At the same time, if the edge weights, state boundaries, and conservative backoff mechanisms are not clearly defined, the control link may exhibit excessively large spans.

[0005] Therefore, a technical solution is needed that connects asynchronous messages, prediction validity period, model state switching, phase change energy storage state estimation, directed graph fault isolation, and pulse-type cooperative control into an executable closed loop. Summary of the Invention

[0006] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a building energy consumption collaborative optimization method and system based on hybrid models and correlation analysis, in order to solve the problems in existing building energy efficiency control, such as strong coupling between prediction model inference time and control cycle, excessively high model complexity during the steady period, failure of phase change energy storage state to enter the model switching boundary, easy occurrence of global degradation due to local faults, and excessively frequent global collaborative optimization.

[0007] This invention provides a building energy consumption collaborative optimization method based on a hybrid model and correlation analysis, comprising: S1: Deploy a message middleware between the data acquisition module and the hybrid prediction model. The data acquisition module collects building energy consumption data in real time. The message middleware buffers the building energy consumption data and sorts it according to timestamp priority. The hybrid prediction model asynchronously obtains building energy consumption data from the message middleware in batch mode for prediction and pushes the prediction results to the local controllers of each region through the message middleware. The local controllers of each region generate control commands based on the prediction results. S2: Calculate the normalized volatility of building energy consumption data, and calculate the joint volatility index based on the normalized volatility of building energy consumption data, the prediction results of the hybrid prediction model, and the prediction residuals. S3: Set a low volatility threshold and a high volatility threshold. If the joint volatility index is lower than the low volatility threshold for multiple consecutive sampling periods and the prediction residual is lower than the residual threshold, switch the hybrid prediction model to simplified inference state. When the joint volatility index exceeds the high volatility threshold, or the prediction residual is abnormal, or the state of charge of the phase change energy storage unit enters the boundary range, or an equipment abnormality is detected, the hybrid prediction model switches to full inference state. S4: Based on the prediction results of the hybrid prediction model, a pulsed cooperative cycle is used for control. The pulsed cooperative cycle includes a long cycle and a short cycle. During the long cycle, global correlation optimization is performed to update the cross-regional cooperative parameters. During the short cycle between the long cycles, the local controller of each region autonomously fine-tunes the status of the equipment in its region according to the control command. When the deviation between the actual power consumption and the predicted power consumption of the region exceeds the deviation threshold or the state of charge of the phase change energy storage unit enters the boundary interval, instantaneous lightweight coordination is triggered. S5: Construct a directed graph of building energy consumption association containing phase change units. When a device fault is detected, perform a forward traversal of the directed graph of building energy consumption association starting from the faulty device to obtain the fault propagation path. The local controller issues energy-saving control commands to the devices in the affected area according to the fault propagation path and adjusts the collaborative control weights of devices in adjacent normal areas to compensate for the energy supply gap, thereby achieving local fault isolation and global collaborative optimization.

[0008] Furthermore, the data acquisition module synchronously collects the operating data of each energy-consuming device, phase change energy storage unit, and regional environment in the building containing phase change energy storage, and performs timestamp alignment, range verification, and missing value marking on the operating data; the operating data includes equipment power, regional temperature, heat exchange medium flow rate, inlet and outlet temperatures of phase change energy storage unit, state of charge of phase change energy storage unit, and actuator status.

[0009] Furthermore, the hybrid prediction model includes an LSTM (Long Short-Term Memory) module, a GRU (Recurrent Neural Network) module, a Transformer module, and a Graph Neural Network (GNN) module. The LSTM module is used to extract short-term and medium-term features, the GRU module outputs the predicted value of the heat load change rate, the Transformer module is used for global dependency modeling, and the Graph Neural Network module is used to resolve the spatial correlation between equipment and regions.

[0010] Furthermore, step S2 includes: S21: Obtain the building energy consumption data sequence, calculate the standard deviation of the building energy consumption data sequence, normalize the standard deviation of the building energy consumption data sequence, and obtain the normalized fluctuation of the building energy consumption data. S22: Obtain the absolute value of the rate of change of heat load at the next moment from the output of the GRU module in the hybrid prediction model as the predicted value of the rate of change of heat load. S23: Calculate the prediction residual based on the predicted heat load change rate and the actual heat load change rate; S24: The normalized fluctuation of building energy consumption data, the predicted value of heat load change rate, and the prediction residual are weighted and summed to obtain the joint fluctuation index.

[0011] Furthermore, the simplified inference state retains only the LSTM and GRU modules performing online predictions, while keeping the parameters of the Transformer and Graph Neural Network modules in memory or GPU memory.

[0012] Furthermore, the directed graph of energy consumption correlation uses energy-consuming equipment, phase change energy storage units, regional environment and actuators as nodes, and physical energy transfer relationships, pipeline connection relationships, electrical connection relationships and energy consumption correlation relationships as directed edges, and updates edge weights according to physical topology, historical correlation and fault status; when equipment failure or phase change energy storage unit unavailable is detected, the affected area and alternative energy supply path are determined along the directed graph of energy consumption correlation.

[0013] Furthermore, the state of charge of the phase change energy storage unit is calculated based on the current available enthalpy, minimum available enthalpy, and maximum available enthalpy of the phase change energy storage unit. The current available enthalpy is estimated from the enthalpy-temperature curve of the phase change material, the inlet and outlet temperatures of the heat exchange medium, the flow rate, and the heat loss of the phase change energy storage unit.

[0014] Furthermore, the objective function of the global correlation optimization includes an operating cost term and a phase change energy storage unit state of charge equilibrium term. The operating cost term is obtained by multiplying and accumulating the electricity price and the total power of the building in the optimization time domain, and the state of charge equilibrium term is obtained by accumulating the degree to which the state of charge of each phase change energy storage unit deviates from the target state of charge.

[0015] Furthermore, instant lightweight coordination only occurs between controllers involved in the affected area, adjacent areas, and alternative power supply paths, and does not trigger a building-wide global optimization recalculation.

[0016] This invention also provides a building energy consumption collaborative optimization system based on a hybrid model and correlation analysis, for executing the aforementioned building energy consumption collaborative optimization method based on a hybrid model and correlation analysis, comprising: The data acquisition module synchronously collects the operating data of each energy-consuming device, phase change energy storage unit, and regional environment in the building containing phase change energy storage; A message middleware is deployed between the data acquisition module and the hybrid prediction model. It is used to buffer the building energy consumption data and sort it according to timestamp priority. The hybrid prediction model asynchronously obtains the building energy consumption data from the message middleware in batch mode for prediction and pushes the prediction results to the local controllers of each region through the message middleware. A joint volatility index calculation module is used to calculate the joint volatility index and switch the hybrid prediction model state between simplified inference state and full inference state based on the joint volatility index. A phase change energy storage state estimation module, which is used to output the state of charge of the phase change energy storage unit; An energy consumption correlation directed graph module is used to determine the fault propagation path, the affected area, and the alternative energy supply path; A pulsed collaborative control module is used to perform global correlation optimization over a long period and trigger instantaneous lightweight coordination over a short period. Local controller, which is used to generate control commands; An actuator drive module is used to drive the air conditioning terminal, valve, circulating pump, heat pump, or phase change heat circuit to execute the control command.

[0017] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: This invention deploys a message middleware between the data acquisition module and the hybrid prediction model and asynchronously pulls data in batch mode. Combined with a timeout rollback strategy within a preset time window, it decouples the control response delay from the model inference time, ensuring that the local controller can always obtain effective prediction results within a defined time interval.

[0018] This invention significantly reduces computational resource consumption by using dynamic model scaling driven by a joint volatility index to discard computationally expensive Transformer and GNN modules during periods of stable energy consumption, while retaining only GRU and LSTM modules.

[0019] By predicting fault propagation paths using energy consumption correlation directed graphs and proactively issuing energy-saving control commands, the scope of fault impact can be limited, and emergency degradation of the entire system can be avoided.

[0020] Pulse-based long and short cycle coordinated control performs global optimization in the long cycle and autonomous fine-tuning in each region in the short cycle, taking into account both global optimization and local response speed.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a building energy consumption collaborative optimization method based on a hybrid model and correlation analysis provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0025] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0026] The following is combined with Figure 1This invention describes a building energy consumption collaborative optimization method and system based on a hybrid model and correlation analysis.

[0027] The technical chain of this invention consists of an input layer, a message decoupling layer, a prediction state layer, a phase change energy storage constraint layer, a fault propagation layer, a collaborative control layer, and an execution feedback layer. The input layer obtains equipment power, zone temperature, heat exchange medium flow rate, inlet and outlet temperatures of the phase change energy storage unit, estimated state of charge of the phase change energy storage unit, electricity price, actuator status, and equipment fault identifiers, and performs timestamp alignment, range verification, missing item marking, and anomaly marking on the input.

[0028] The message decoupling layer takes verified operational data as input and outputs data batches sorted by device identifier and timestamp, as well as a queue of prediction results. The message decoupling layer does not directly generate control commands; its role is to isolate bursts in data acquisition and fluctuations in model inference time. Before using prediction results, the local controller must check the validity period of the prediction results, data quality identifiers, and message backlog status. When prediction results expire or the message middleware is unavailable, the local controller switches to a conservative control strategy. This conservative control strategy includes maintaining comfort boundaries, limiting the charging and discharging power of the phase change energy storage unit, freezing online model updates, and outputting maintenance prompts.

[0029] The prediction state layer receives data batches and outputs the predicted load, predicted heat load change rate, predicted residuals, and model state. The joint volatility index is used to select the full inference state or the simplified inference state. When the state of charge of the phase change energy storage unit enters the boundary range, equipment malfunctions, message backlog occurs, or the prediction residual exceeds the limit, the prediction state layer forces the restoration to the full inference state and does not allow further simplification simply because the energy consumption sequence is temporarily stable.

[0030] The phase change energy storage constraint layer estimates the state of charge (SOC) of the phase change energy storage unit based on the enthalpy-temperature curve of the phase change material (the curve showing the relationship between the enthalpy of the phase change material and temperature), the inlet and outlet temperatures of the heat exchange medium, the flow rate, and the heat loss. It then outputs permissible charge / discharge, boundary interval, or unavailable states to the prediction state layer and the collaborative control layer. These states are used to define model switching conditions, global optimization constraints, and short-cycle control actions.

[0031] The fault propagation layer is based on a directed graph of energy consumption correlation. It takes fault nodes, edge weights, and phase change energy storage status as inputs and outputs the affected area and alternative energy supply paths. The edge weights are not simply statistical correlation coefficients, but are jointly determined by physical topology indicators, historical correlations, and fault reachability states. When topology information is missing or data quality is insufficient, the system does not use the edge weights to generate fault isolation actions.

[0032] The collaborative control layer outputs cross-regional collaborative parameters over long periods and local control adjustments over short periods. When the deviation does not exceed the threshold and the phase change energy storage state does not reach the boundary, only local fine-tuning is performed in short periods; when the deviation exceeds the limit or the phase change energy storage state enters the boundary, lightweight coordination within the affected area is triggered. The execution feedback layer collects data from the actuators, equipment status, and regional environment, and sends it back to the input layer to form a closed loop.

[0033] like Figure 1 As shown, a building energy consumption collaborative optimization method based on a hybrid model and correlation analysis includes: S1: Deploy a message middleware between the data acquisition module and the hybrid prediction model. The data acquisition module collects building energy consumption data in real time. The message middleware buffers the building energy consumption data and sorts it according to timestamp priority. The hybrid prediction model asynchronously obtains building energy consumption data from the message middleware in batch mode for prediction and pushes the prediction results to the local controller through the message middleware. The local controller generates control commands based on the prediction results. The data acquisition module synchronously collects the operating data of each energy-consuming device, phase change energy storage unit, and regional environment in the building containing phase change energy storage, and performs timestamp alignment, range verification, and missing data marking on the operating data. The operating data includes equipment power, regional temperature, heat exchange medium flow rate, inlet and outlet temperatures of phase change energy storage unit, state of charge of phase change energy storage unit, and actuator status.

[0034] Buildings incorporating phase change energy storage include a building management master station, multiple area controllers, air conditioning terminals, circulating pumps, heat pumps, phase change heat transfer loops, smart meters, area temperature sensors, heat exchange medium flow meters, temperature sensors installed at the inlet and outlet of the phase change energy storage units, and actuator drive modules. The data acquisition module collects equipment power, area temperature, heat exchange medium flow rate, inlet and outlet temperatures of the phase change energy storage units, actuator status, and fault indicators according to a unified sampling period. It also performs timestamp alignment, range verification, abrupt change verification, and missing data flags on the data, outputting valid operating data and quality indicators.

[0035] A message middleware is deployed between the data acquisition module and the hybrid forecasting model. The data acquisition module writes verified operational data into the raw operational data queue. The hybrid forecasting model pulls data according to batch conditions and outputs the predicted load, predicted heat load change rate, predicted residual, and predicted effective time. When the local controller does not receive a new valid forecast result within a preset time window, it uses the previous forecast result that has not exceeded the effective time. If the previous forecast result exceeds the effective time or the data quality does not meet the conditions, it switches to a conservative control strategy. The predicted effective time is used to limit the maximum time range within which the local controller can use the previous forecast result. The calculation expression for the predicted effective time is: in, To predict the effective time, For the current control moment, When the prediction result is generated, Greater than the maximum effective time At that time, the prediction results must not continue to be used as normal control inputs. It is determined by the control cycle, network latency, and actuator response time.

[0036] When the effective prediction time is less than or equal to When the data corresponding to the prediction result is valid, the local controller can use the prediction result; when the prediction validity period is longer than [a certain time], the local controller can use the prediction result. When the message middleware is unavailable or the message backlog exceeds the threshold, the local controller enters a conservative control strategy, limiting the charging and discharging power of the phase change energy storage unit and maintaining zone control according to the comfort boundary.

[0037] The preset time window is determined based on the building control cycle, message middleware queue delay, model inference delay, and execution mechanism response time, and the effective prediction time is not greater than the preset time window.

[0038] The hybrid prediction model includes an LSTM module, a GRU module, a Transformer module, and a graph neural network module. The LSTM module is used to extract short-term and medium-term features, the GRU module outputs the predicted value of the heat load change rate, the Transformer module is used for global dependency modeling, and the graph neural network module is used to resolve the spatial relationship between equipment and regions.

[0039] S2: Calculate the normalized volatility of building energy consumption data, and calculate the joint volatility index based on the normalized volatility of building energy consumption data, the prediction results of the hybrid prediction model, and the prediction residuals. S21: Obtain the building energy consumption data sequence, calculate the standard deviation of the building energy consumption data sequence, normalize the standard deviation of the building energy consumption data sequence, and obtain the normalized fluctuation of the building energy consumption data. S22: Obtain the absolute value of the rate of change of heat load at the next moment from the output of the GRU module in the hybrid prediction model as the predicted value of the rate of change of heat load. S23: Calculate the prediction residual based on the predicted heat load change rate and the actual heat load change rate; S24: The normalized fluctuation of building energy consumption data, the predicted value of heat load change rate, and the prediction residual are weighted and summed to obtain the joint fluctuation index. The calculation expression of the joint fluctuation index is as follows: in, For joint volatility index, The normalized fluctuation weighting coefficient for building energy consumption data. This represents the normalized fluctuation of building energy consumption data. This is the weighting coefficient for the rate of change of heat load. This is the predicted rate of change of heat load at the next moment. To predict the weighting coefficients of the residual term, To predict residuals, .

[0040] Before being included in the weighted calculation, the normalized fluctuation of building energy consumption data, the predicted value of heat load change rate, and the predicted residual are normalized according to the statistical scale of the most recent normal window.

[0041] S3: Set a low volatility threshold and a high volatility threshold. If the joint volatility index is lower than the low volatility threshold for multiple consecutive sampling periods and the prediction residual is lower than the residual threshold, switch the hybrid prediction model to simplified inference state. When the joint volatility index exceeds the high volatility threshold, or the prediction residual is abnormal, or the state of charge of the phase change energy storage unit enters the boundary range, or an equipment abnormality is detected, the hybrid prediction model switches to full inference state. The predicted load and predicted heat load change rate output by the full inference state are entered into long-cycle global correlation optimization; the predicted results output by the simplified inference state are only used for short-cycle local fine-tuning within the prediction validity period and the allowed range of phase change state of charge. The simplified inference state only retains the LSTM module and GRU module to perform online prediction, disables the online forward computation of the Transformer module and graph neural network module, and keeps the parameters of the Transformer module and graph neural network module in memory or in video memory.

[0042] Before entering the simplified inference state, the system must simultaneously meet the following conditions: the joint volatility index is below the low volatility threshold for multiple consecutive sampling periods and the prediction residual is below the residual threshold, the state of charge of the phase change energy storage unit is within the allowable range, the message backlog does not exceed the threshold, and no equipment abnormality is detected.

[0043] When the joint volatility index exceeds the high volatility threshold, the prediction residual exceeds the limit, the state of charge of the phase change energy storage unit enters the boundary range, the message backlog exceeds the threshold, or a device anomaly is detected, the system restores to the complete inference state. After restoring the complete inference state, the system uses the complete model output as the standard and records the model state switching event, the triggering reason, the prediction difference before and after the switch, and the executed control result. The model state serves as the input to the local controller's arbitration control command and does not generate a final control command separately.

[0044] The state of charge (SOC) of a phase change energy storage unit entering the boundary range refers to the SOC being lower than the lower limit threshold or higher than the upper limit threshold. In this embodiment, the upper limit threshold is 80% and the lower limit threshold is 20%.

[0045] S4: Based on the prediction results of the hybrid prediction model, a pulsed cooperative cycle is used for control. The pulsed cooperative cycle includes a long cycle and a short cycle. During the long cycle, global correlation optimization is performed to update the cross-regional cooperative parameters. During the short cycle between the long cycles, the local controller of each region autonomously fine-tunes the status of the equipment in its region according to the control command. When the deviation between the actual power consumption and the predicted power consumption of the region exceeds the deviation threshold or the state of charge of the phase change energy storage unit enters the boundary interval, instantaneous lightweight coordination is triggered. The phase change energy storage state estimation module estimates the state of charge of the phase change energy storage unit based on the enthalpy-temperature curve of the phase change material, the inlet and outlet temperatures of the heat exchange medium, the flow rate of the heat exchange medium, and the heat loss. It then outputs an indication of whether charging / discharging is allowed, the boundary, or unavailable to the joint fluctuation index calculation module and the pulse-type collaborative control module. If the indication is boundary or unavailable, the system restores the complete inference state and restricts the phase change heat exchange loop from continuing to move towards the boundary. The enthalpy-temperature curve is used to describe the enthalpy change characteristics of the phase change material at different temperatures.

[0046] The formula for calculating the state of charge of a phase change energy storage unit is: in, The state of charge of the g-th phase change energy storage unit is... This is the currently available enthalpy value. The minimum usable enthalpy value, The maximum available enthalpy is estimated based on the enthalpy-temperature curve of the phase change material, the inlet and outlet temperatures of the heat exchange medium, the flow rate of the heat exchange medium, and the heat loss.

[0047] When the state of charge of the phase change energy storage unit is below the lower limit of energy release or above the upper limit of energy storage, the phase change energy storage state estimation module outputs a boundary state identifier to the joint fluctuation index calculation module and the pulsed collaborative control module. The boundary state identifier enables the hybrid prediction model to recover its complete inference state and limits the phase change heat transfer loop from continuing to move towards the boundary in short-cycle control.

[0048] The pulse-type coordinated control module divides the control cycle into long cycles and short cycles.

[0049] In some specific embodiments of the present invention, the long cycle is set to 15 minutes and the short cycle is set to 1 minute, that is, global correlation optimization is performed once every 15 minutes, and local fine-tuning is performed once every 1 minute in each region during this period.

[0050] A global correlation optimization is performed once over a long period. Inputs include predicted load, electricity price, state of charge (SOC) of phase change energy storage units (PCS units), comfort constraints, upper and lower limits of equipment power, and fault availability status. Outputs cross-regional collaborative parameters. The objective function of the global correlation optimization includes an operating cost term and a SOC equilibrium term. The operating cost term is obtained by multiplying and summing the electricity price and the total building power within the optimization time domain. The SOC equilibrium term is obtained by summing the deviations of the SOC of each PCS unit from the target SOC. The calculation expression for the global correlation optimization objective function is as follows: in, To optimize the objective function for global correlation, For time period Electricity price, To optimize the total number of time periods in the time domain, For time period Total building power For discrete time steps, For the balance weight of the state of charge, The state of charge of the g-th phase change energy storage unit is... For the g-th phase change energy storage unit, This represents the number of phase change energy storage units.

[0051] Within a short period, each area controller autonomously adjusts the opening of the air conditioning terminal valves, the frequency of the circulating pump, or the flow rate of the phase change heat transfer loop based on local short-term forecasts. The expression for calculating the area power consumption deviation rate is as follows: in, Let z be the deviation rate of region z at time k. Let z be the actual power consumption of region z at time k. The predicted power consumption of region z at time k. This represents the rated power consumption of region z.

[0052] Instant lightweight coordination occurs only between controllers in the affected area, adjacent areas, and alternative power supply paths, and does not trigger a building-wide global optimization recalculation.

[0053] When the deviation rate does not exceed the deviation threshold and the state of charge of the phase change energy storage unit is within the allowable range, only local fine-tuning is performed in the short cycle. When the deviation rate exceeds the deviation threshold, the state of charge of the phase change energy storage unit enters the boundary range, or the fault propagation layer output is affected, immediate lightweight coordination is triggered. Immediate lightweight coordination only exchanges supply and demand gaps, adjustable power, and comfort boundaries among controllers involved in the affected area, adjacent areas, and alternative power supply paths, without triggering a building-wide global optimization recalculation. The lightweight coordination output is converted into actuator drive quantities only after arbitration by the local controller.

[0054] Within the short cycles between adjacent long cycles, each zone controller autonomously adjusts the control quantities of the air conditioning terminals, valves, circulating pumps, or phase change heat exchange circuits based on local short-term forecasts.

[0055] S5: Construct a directed graph of building energy consumption association containing phase change units. When a device fault is detected, perform a forward traversal of the directed graph of building energy consumption association starting from the faulty device to obtain the fault propagation path. The local controller issues energy-saving control commands to the devices in the affected area according to the fault propagation path and adjusts the collaborative control weights of devices in adjacent normal areas to compensate for the energy supply gap, thereby achieving local fault isolation and global collaborative optimization.

[0056] The affected equipment includes air conditioning terminals, valves, circulating pumps, heat pumps, or phase change heat transfer circuits.

[0057] The directed graph of energy consumption correlation uses energy-consuming equipment, phase change energy storage units, regional environment, and actuators as nodes, and physical energy transfer relationships, pipeline connection relationships, electrical connection relationships, and confirmed energy consumption correlation relationships as directed edges. The edge weights are updated according to physical topology, historical correlation, and fault status. When equipment failure or unavailability of phase change energy storage units is detected, the affected area and alternative energy supply paths are determined along the directed graph of energy consumption correlation. The energy consumption correlation directed graph module uses energy-consuming equipment, phase change energy storage units, regional environment, and actuators as nodes. After reading the physical topology, historical correlations, and fault status, it outputs the affected areas and alternative energy supply paths. The pulse-type collaborative control module generates cross-regional collaborative parameters based on predicted load, electricity price, phase change energy storage unit state of charge, and equipment operating constraints over a long period. In a short period, it outputs valve opening, circulating pump frequency, heat pump start / stop, or phase change heat circuit flow rate adjustment based on local predictions, deviation rate, and boundary conditions.

[0058] In some specific embodiments of the present invention, the directed graph of energy consumption correlation uses chiller units, circulating pumps, air conditioning terminals, phase change energy storage units, zone temperature nodes, and actuators as nodes. Directed edges are initially determined by pipeline connections, electrical connections, and heat exchange relationships; when the topological relationship is established and the data quality meets the conditions, the edge weights are then adjusted based on historical energy consumption correlations. The expression for calculating the edge weights is: in, for Time Node To node edge weights, For physical topology weighting coefficients, Historical correlation weighting coefficient, These are the weighting coefficients for reachable states in the faulty state. For physical topology indication, for Time Node and nodes Historical energy consumption correlation coefficient, Representation Nodes and nodes The strength of the energy consumption correlation between them is not distinguished between positive and negative correlations. This is a fault-reachable state. The value is either 0 or 1, indicating whether a fault is likely to originate from a node. propagation to nodes .

[0059] In this embodiment, the heat exchange relationship is a specific manifestation of the physical energy transfer relationship. In actual deployment, the energy consumption correlation still needs to be confirmed through data quality verification before it can be included in the edge weight calculation.

[0060] The edge weights of the energy consumption correlation directed graph are jointly determined by the physical topology indicator, the historical energy consumption correlation coefficient, and the fault reachability state. The historical energy consumption correlation coefficient is updated only when time alignment, data quality, and topology consistency are met.

[0061] When a circulating pump stops, a valve gets stuck, a phase change energy storage unit becomes unavailable, or a sensor malfunctions, the system determines the affected area and alternative power supply paths along the directed graph of energy consumption correlation. If the data quality of the directed graph edge weights is insufficient or the topological relationships are uncertain, automatic fault isolation is not performed; instead, a conservative control strategy is adopted, and a maintenance prompt is output. The affected area and alternative power supply paths serve as inputs to a lightweight coordination range, avoiding unarbitrated parallel outputs between fault isolation results and global optimization results.

[0062] When the message middleware is unavailable, the prediction results time out continuously, the state of charge estimation of the phase change energy storage unit is unavailable, or the fault propagation path is uncertain, the local controller enters a conservative control strategy.

[0063] The local controller arbitrates the validity of the message decoupling layer, the model state of the prediction state layer, the boundary markers of the phase change energy storage constraint layer, the affected area of ​​the fault propagation layer, and the control quantities of the collaborative control layer to generate the final control command. The control command is sent to the air conditioning terminal, valves, circulating pumps, heat pumps, or phase change heat transfer loops; the actuator feedback, regional environmental data, and equipment status are returned to the data acquisition module.

[0064] A building energy consumption collaborative optimization system based on a hybrid model and correlation analysis, used to execute the aforementioned building energy consumption collaborative optimization method based on a hybrid model and correlation analysis, includes: The data acquisition module synchronously collects the operating data of each energy-consuming device, phase change energy storage unit, and regional environment in the building containing phase change energy storage; A message middleware is deployed between the data acquisition module and the hybrid prediction model. It is used to buffer the building energy consumption data and sort it according to timestamp priority. The hybrid prediction model asynchronously obtains the building energy consumption data from the message middleware in batch mode for prediction and pushes the prediction results to the local controllers of each region through the message middleware. A joint volatility index calculation module is used to calculate the joint volatility index and switch the hybrid prediction model state between simplified inference state and full inference state based on the joint volatility index. A phase change energy storage state estimation module, which is used to output the state of charge of the phase change energy storage unit; An energy consumption correlation directed graph module is used to determine the fault propagation path, the affected area, and the alternative energy supply path; A pulsed collaborative control module is used to perform global correlation optimization over a long period and trigger instantaneous lightweight coordination over a short period. Local controller, which is used to generate control commands; An actuator drive module is used to drive the air conditioning terminal, valve, circulating pump, heat pump, or phase change heat circuit to execute the control command.

[0065] Through the collaborative work of the above modules, a message middleware is deployed between the data acquisition module and the hybrid prediction model to asynchronously pull data in batch mode. Combined with the timeout rollback strategy within the preset time window, the control response delay and the model inference time consumption are decoupled, and the local controller can always obtain effective prediction results within a certain time interval.

[0066] This invention significantly reduces computational resource consumption by using dynamic model scaling driven by a joint volatility index to discard computationally expensive Transformer and GNN modules during periods of stable energy consumption, while retaining only GRU and LSTM modules.

[0067] By predicting fault propagation paths using energy consumption correlation directed graphs and proactively issuing energy-saving control commands, the scope of fault impact can be limited, and emergency degradation of the entire system can be avoided.

[0068] Pulse-based long and short cycle coordinated control performs global optimization in the long cycle and autonomous fine-tuning in each region in the short cycle, taking into account both global optimization and local response speed.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A building energy consumption collaborative optimization method based on hybrid model and correlation analysis, characterized in that, include: S1: Deploy a message middleware between the data acquisition module and the hybrid prediction model. The data acquisition module collects building energy consumption data in real time. The message middleware buffers the building energy consumption data and sorts it according to timestamp priority. The hybrid prediction model asynchronously obtains building energy consumption data from the message middleware in batch mode for prediction and pushes the prediction results to the local controllers of each region through the message middleware. The local controllers of each region generate control commands based on the prediction results. S2: Calculate the normalized volatility of building energy consumption data, and calculate the joint volatility index based on the normalized volatility of building energy consumption data, the prediction results of the hybrid prediction model, and the prediction residuals. S21: Obtain the building energy consumption data sequence, calculate the standard deviation of the building energy consumption data sequence, normalize the standard deviation of the building energy consumption data sequence, and obtain the normalized fluctuation of the building energy consumption data. S22: Obtain the absolute value of the rate of change of heat load at the next moment from the output of the GRU module in the hybrid prediction model as the predicted value of the rate of change of heat load. S23: Calculate the prediction residual based on the predicted heat load change rate and the actual heat load change rate; S24: The normalized fluctuation of building energy consumption data, the predicted value of heat load change rate, and the prediction residual are weighted and summed to obtain the joint fluctuation index; S3: Set a low volatility threshold and a high volatility threshold. If the joint volatility index is lower than the low volatility threshold for multiple consecutive sampling periods and the prediction residual is lower than the residual threshold, switch the hybrid prediction model to simplified inference state. When the joint volatility index exceeds the high volatility threshold, or the prediction residual is abnormal, or the state of charge of the phase change energy storage unit enters the boundary range, or an equipment abnormality is detected, the hybrid prediction model switches to full inference state. The simplified inference state retains only the LSTM and GRU modules to perform online prediction, while keeping the parameters of the Transformer and Graph Neural Network modules in memory or GPU memory; the complete inference state is a hybrid prediction model, which includes the LSTM, GRU, Transformer, and Graph Neural Network modules. The state of charge of the phase change energy storage unit is calculated based on the current available enthalpy, minimum available enthalpy, and maximum available enthalpy of the phase change energy storage unit. The current available enthalpy is estimated from the enthalpy-temperature curve of the phase change material, the inlet and outlet temperatures and flow rate of the heat exchange medium, and the heat loss of the phase change energy storage unit. S4: Based on the prediction results of the hybrid prediction model, a pulsed cooperative cycle is used for control. The pulsed cooperative cycle includes a long cycle and a short cycle. During the long cycle, global correlation optimization is performed to update the cross-regional cooperative parameters. During the short cycle between the long cycles, the local controller of each region autonomously fine-tunes the status of the equipment in its region according to the control command. When the deviation between the actual power consumption and the predicted power consumption of the region exceeds the deviation threshold or the state of charge of the phase change energy storage unit enters the boundary interval, instantaneous lightweight coordination is triggered. The objective function of the global correlation optimization includes an operating cost term and a phase change energy storage unit state of charge equilibrium term. The operating cost term is obtained by multiplying and accumulating the electricity price and the total power of the building in the optimization time domain. The state of charge equilibrium term is obtained by accumulating the degree to which the state of charge of each phase change energy storage unit deviates from the target state of charge. Instant lightweight coordination occurs only between controllers in the affected area, adjacent areas, and alternative power supply paths, and does not trigger a building-wide global optimization recalculation. S5: Construct a directed graph of building energy consumption association containing phase change units. When a device fault is detected, perform a forward traversal of the directed graph of building energy consumption association starting from the faulty device to obtain the fault propagation path. The local controller issues energy-saving control commands to the devices in the affected area according to the fault propagation path and adjusts the collaborative control weights of devices in adjacent normal areas to compensate for the energy supply gap, thereby achieving local fault isolation and global collaborative optimization.

2. The building energy consumption collaborative optimization method based on hybrid model and correlation analysis according to claim 1, characterized in that, The data acquisition module synchronously collects the operating data of each energy-consuming device, phase change energy storage unit, and regional environment in the building containing phase change energy storage, and performs timestamp alignment, range verification, and missing value marking on the operating data; the operating data includes equipment power, regional temperature, heat exchange medium flow rate, inlet and outlet temperatures of phase change energy storage unit, state of charge of phase change energy storage unit, and actuator status.

3. The building energy consumption collaborative optimization method based on hybrid model and correlation analysis according to claim 1, characterized in that, The LSTM module is used to extract short- and medium-term features, the GRU module outputs predicted values ​​of heat load change rate, the Transformer module is used for global dependency modeling, and the graph neural network module is used to resolve the spatial relationships between equipment and regions.

4. The building energy consumption collaborative optimization method based on hybrid model and correlation analysis according to claim 1, characterized in that, The directed graph of energy consumption correlation uses energy-consuming equipment, phase change energy storage units, regional environment and actuators as nodes, and physical energy transfer relationships, pipeline connection relationships, electrical connection relationships and energy consumption correlation relationships as directed edges. The edge weights are updated according to physical topology, historical correlation and fault status. When equipment failure or phase change energy storage unit unavailable is detected, the affected area and alternative energy supply path are determined along the directed graph of energy consumption correlation.

5. A building energy consumption collaborative optimization system based on hybrid models and correlation analysis, characterized in that, To implement the building energy consumption collaborative optimization method based on hybrid model and correlation analysis as described in any one of claims 1 to 4, comprising: The data acquisition module synchronously collects the operating data of each energy-consuming device, phase change energy storage unit, and regional environment in the building containing phase change energy storage; A message middleware is deployed between the data acquisition module and the hybrid prediction model. It is used to buffer the building energy consumption data and sort it according to timestamp priority. The hybrid prediction model asynchronously obtains the building energy consumption data from the message middleware in batch mode for prediction and pushes the prediction results to the local controllers of each region through the message middleware. A joint volatility index calculation module is used to calculate the joint volatility index and switch the hybrid prediction model state between simplified inference state and full inference state based on the joint volatility index. A phase change energy storage state estimation module, which is used to output the state of charge of the phase change energy storage unit; An energy consumption correlation directed graph module is used to determine the fault propagation path, the affected area, and the alternative energy supply path; A pulsed collaborative control module is used to perform global correlation optimization over a long period and trigger instantaneous lightweight coordination over a short period. Local controller, which is used to generate control commands; An actuator drive module is used to drive the air conditioning terminal, valve, circulating pump, heat pump, or phase change heat circuit to execute the control command.

Images