A method for coupling building photovoltaic power with building energy management system

Abstract

This invention provides a method for coupling building photovoltaic (PV) power with a building energy management system, belonging to the field of intelligent building energy management technology. It includes: deploying distributed edge computing nodes to establish bidirectional communication connections with a cloud-based collaborative control platform; configuring local autonomous operation modes and offline control strategies in case of communication failures; constructing a capability profile database for various controllable devices; establishing a three-tiered hierarchical control architecture and configuring corresponding control levels and priority rules for devices with different response levels; generating load forecast curves and PV output forecast curves for the next 24 hours; calculating optimized scheduling schemes for multiple future time windows and determining preliminary scheduling instructions for each controllable device; and automatically determining the final execution instruction based on the priority of each instruction and the current operating status of the device when multiple control levels simultaneously issue scheduling instructions to the same device.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent building energy management, and more particularly to a method for coupling building photovoltaic power with a building energy management system. Background Technology

[0002] With the rapid development of renewable energy, photovoltaic (PV) power generation has become one of the important forms of clean energy in building energy systems. Building-integrated photovoltaic (BIPV) systems can provide a considerable supply of electricity during the day, but their power generation fluctuates significantly due to variations in sunlight intensity, weather changes, and seasons. This leads to unstable PV power utilization and makes it difficult to accurately match the building's internal load demand. Traditional building energy management systems typically adjust the load based on fixed rules or single predictive models, lacking the dynamic response capability to changes in PV output, which can easily result in energy waste or decreased user comfort.

[0003] Current methods for coupling photovoltaic (PV) and building energy management systems largely rely on centralized control or simple rolling optimization strategies. These methods suffer from problems such as large computational delays, insufficient real-time performance, and inconsistent device responses. Conflicts or infeasible optimization solutions can easily arise, especially when multiple controllable devices participate in regulation simultaneously. Furthermore, in distributed building environments, the response speeds, regulation capabilities, and operational constraints of various controllable devices differ significantly, making it impossible for single-level control to simultaneously address the dual objectives of rapid fluctuations and medium- to long-term economic operation.

[0004] In recent years, edge computing and cloud-based collaborative control technologies have provided new solutions for building energy management. However, existing applications are mostly limited to single-node or local optimization, and a global, adaptive closed-loop control system for building photovoltaic power and multi-level load equipment has not yet been formed. In addition, issues such as offline equipment operation, communication interruptions, and prediction errors also restrict the stability and reliability of the system, resulting in a significant deviation between actual operating results and theoretical optimization.

[0005] In summary, how to achieve multi-level hierarchical control, real-time prediction, and adaptive optimization of building photovoltaic power and building energy management systems, and improve the self-consumption rate of photovoltaic power while ensuring user comfort and economy, has become an urgent technical problem to be solved. Summary of the Invention

[0006] In view of the technical problems mentioned in the background section, a method for coupling building photovoltaic power and building energy management system is provided.

[0007] The technical means employed in this invention are as follows:

[0008] A method for coupling building-integrated photovoltaic (BIPV) power with a building energy management system, wherein the building energy management system includes: a photovoltaic power generation monitoring module, a load forecasting module, distributed edge computing nodes, a cloud-based collaborative control platform, and a multi-layer equipment scheduling module; the coupling method includes the following steps:

[0009] Step 1: Deploy distributed edge computing nodes to establish a two-way communication connection with the cloud-based collaborative control platform, and configure local autonomous operation mode and offline control strategy in case of communication failure;

[0010] Step 2: Obtain the equipment type, rated power, response time, adjustment range, and operating constraints of all controllable equipment in the building, and construct a capability profile database for each type of controllable equipment.

[0011] Step 3: Based on the response time, adjustment range and operating constraints of each device in the capability profile database, all controllable devices are divided into second-level response devices, minute-level response devices and hour-level response devices. A three-level hierarchical control architecture is established, and corresponding control levels and priority rules are configured for devices with different response levels.

[0012] Step 4: Collect building load data, photovoltaic power generation data and environmental parameters in real time, and combine them with historical operating data to generate load forecast curves and photovoltaic output forecast curves for the next 24 hours by the load forecast module and photovoltaic power generation monitoring module, respectively.

[0013] Step 5: Based on the load forecast curve and the photovoltaic output forecast curve, calculate the optimized scheduling scheme for multiple future time windows through the multi-layer equipment scheduling module, and determine the preliminary scheduling instructions for each controllable device according to the actual operating status at the current moment.

[0014] Step 6: When multiple control levels issue scheduling instructions to the same device at the same time, the final instruction to be executed is automatically determined based on the priority of each instruction and the current operating status of the device.

[0015] Step 7: If there are conflicts between multiple control levels that make the optimization problem infeasible, then the priority weight coefficients of each control level are dynamically adjusted according to the current energy supply and demand tension, the degree of deviation of user comfort and the estimated economic loss.

[0016] Step 8: If no feasible solution is found after adjusting the weights, then, under the premise of always satisfying the hard constraints related to equipment safety, system stability and grid connection specifications, the non-critical constraints related to user comfort and economy are gradually relaxed within their preset allowable deviation range according to the preset relaxation priority; when the scheduling scheme satisfies all hard constraints and the deviation of each non-critical constraint is within the corresponding allowable deviation range, a feasible scheduling scheme is output.

[0017] Further, step 1 includes the following steps:

[0018] The installation location and number of each edge computing node are determined based on the actual spatial layout of the building and the density of equipment distribution, so that the communication latency within the coverage area of ​​each node does not exceed 200 milliseconds.

[0019] Configure an independent data cache space for each edge computing node;

[0020] Establish an encrypted two-way communication channel between each edge computing node and the cloud-based collaborative control platform;

[0021] Set offline control trigger conditions for each edge computing node; the offline control trigger conditions are: when the edge computing node fails to communicate with the cloud platform 5 times in a row or the single communication delay exceeds 10 seconds, it automatically switches to local autonomous operation mode and independently executes device scheduling and energy consumption control tasks based on local cached historical data and preset rule base.

[0022] Once communication between the offline edge computing node and the cloud platform is restored, the offline edge computing node proactively uploads its offline operation records and control decision data to the cloud platform.

[0023] Furthermore, step 2 includes the following steps:

[0024] Basic parameter acquisition: Collect basic information of each controllable device, and obtain real-time operating data of the device; the basic information includes: rated power marked on the nameplate, device type and operating mode;

[0025] Energy consumption characteristics statistical analysis; based on actual operating data collected continuously by edge computing nodes and stored in local or cloud collaborative control platforms for no less than 7 days, statistical analysis is performed on the power changes and energy consumption levels of each controllable device under different load rates;

[0026] Rated parameter correction: Based on the statistical analysis results of each controllable device, an energy consumption curve characterizing the actual operating characteristics is constructed, and the energy consumption curve is compared with the rated parameters marked on the nameplate. The rated power is corrected according to the comparison results to form correction parameters that reflect the actual energy consumption characteristics of the corresponding controllable device.

[0027] Response performance measurement; When implementing operational status control on controllable equipment, the cloud-based collaborative control platform records the time from the effective control to the stable operation of the equipment, in order to quantify the dynamic response performance of the equipment;

[0028] The adjustment characteristics are collected; for stepped adjustment equipment, the power value corresponding to each step is recorded; for continuous adjustment equipment, the minimum adjustment step size and adjustment rate are recorded to characterize the control accuracy and adjustment sensitivity of the equipment.

[0029] Operational constraint parameter recording: Collect the operational constraints of each controllable device; the operational constraints include: the minimum continuous operating time, the minimum downtime interval, and the recommended maximum number of daily start-ups and shutdowns as indicated in the device's factory technical parameters and operating specifications;

[0030] Structured modeling and storage: The acquired basic parameters, energy consumption characteristics, corrected rated power, response performance, regulation characteristics and operating constraints are structured and stored to build a database of controllable equipment capability profiles.

[0031] Furthermore, step 3 includes the following steps:

[0032] Devices with a response time of less than 30 seconds are classified as second-level response devices and assigned to the first control level, with a control cycle set to 10 seconds, to cope with rapid fluctuations in photovoltaic output and load;

[0033] Devices with response times between 30 seconds and 10 minutes are classified as minute-level response devices and assigned to the second control level, with a control cycle set to 5 minutes, to regulate short-term load trends.

[0034] Devices with a response time exceeding 10 minutes are classified as hourly response devices and assigned to the third control level, with a control cycle set to 15 minutes for medium- and long-term economic operation optimization.

[0035] Priority rules are configured for each control level: when the power deviation exceeds the threshold, the first control level intervenes and overrides the existing scheduling plans of the second and third control levels; when the first control level is not triggered or after intervention, the second control level adjusts the equipment according to the coordination results with the first control level; the third control level generates a plan in advance based on the prediction information and executes it without being overridden by the higher priority level. Each level achieves collaborative control through sharing status information.

[0036] Furthermore, generating the photovoltaic output prediction curve includes the following steps:

[0037] The environmental parameters of the area where the photovoltaic array is located are obtained based on numerical weather forecasts, and the irradiance of the tilted surface of the photovoltaic module is calculated to obtain the actual solar energy intensity data received by the photovoltaic module.

[0038] By combining the installed capacity, historical efficiency data and current operating status of the photovoltaic array, the expected power output of each photovoltaic module is calculated, and a preliminary photovoltaic power output prediction curve is generated.

[0039] When the average deviation between the actual irradiance and the predicted value in the last 30 minutes exceeds 30%, the latest cloud map data and short-term weather forecast are called to recalculate the irradiance of the tilted surface, and a corrected photovoltaic output prediction curve is generated based on the photovoltaic array parameters.

[0040] Furthermore, based on the load forecast curve and the photovoltaic output forecast curve, the multi-layer equipment scheduling module calculates an optimized scheduling scheme for multiple future time windows, including the following steps:

[0041] The next 24 hours are divided into multiple time windows, each lasting 15 minutes;

[0042] For each time window, an optimization problem is constructed based on load forecasting and photovoltaic output forecasting, covering economic objectives, energy utilization efficiency objectives, and comfort constraints. The objective function terms corresponding to each objective are combined according to preset weights and transformed into a single-objective optimization problem.

[0043] Multi-dimensional operation constraint rules are introduced into the optimization scheme for each time window. These multi-dimensional operation constraint rules impose restrictions from the aspects of power balance, equipment operating boundary, energy storage state of charge and grid connection point power.

[0044] Solving the single-objective optimization problem yields the power setpoints, start / stop states, and energy storage charging / discharging strategies for each controllable device within each time window. Specifically, day-ahead scheduling employs an accurate algorithm to generate a globally optimized scheduling strategy, while intraday rolling scheduling uses a fast heuristic algorithm to generate a real-time scheduling strategy that satisfies the constraint rules. These scheduling strategies are then passed to the multi-layer device scheduling module for execution.

[0045] Furthermore, step 6 includes the following steps:

[0046] Edge computing nodes receive scheduling instructions from various control levels and collect the urgency level identifier and instruction type of each instruction, wherein the urgency level includes: regular instructions, urgent instructions and emergency instructions;

[0047] The received dispatch instructions are initially prioritized: if there are emergency instructions, they are directly used as the final execution instructions; if all instructions are regular or urgent instructions, the initial priority of the instructions is first determined according to the control level, and the execution order is determined according to the time sequence within the same control level.

[0048] When a scheduling instruction exceeds the current controllable range of the device, the edge computing node corrects it to ensure that the instruction remains within the device's executable range.

[0049] Edge computing nodes select instructions that meet high-priority triggering conditions and are within a controllable range as the final execution instructions.

[0050] Furthermore, the coupling method further includes the following steps:

[0051] Step 9: The final execution instruction is sent to the corresponding controllable device through the edge computing node, and the instruction response time, execution deviation and operating status of each device are recorded in real time.

[0052] Step 10: During the equipment execution process, the edge computing node continuously collects actual energy consumption data, photovoltaic power generation data, load change data and equipment operation status data, calculates the error between the load prediction value and the actual value, the photovoltaic output prediction error and the equipment execution deviation, and feeds back the error and deviation data to the corresponding control level.

[0053] Step 11: Based on the error and deviation data, perform online correction on the prediction model parameters of the load prediction module and the photovoltaic power generation monitoring module, and dynamically adjust the weight coefficient adjustment strategy, priority rules and relaxation priority to generate updated control parameters.

[0054] Step 12: Determine whether the next rolling optimization time window has been reached based on the preset rolling optimization cycle. If not, continue to execute the current scheduling scheme, continuously monitor the equipment operating status, and make real-time fine adjustments for sudden load fluctuations or changes in photovoltaic output. If the window has been reached, re-execute steps 4 to 11 based on the updated control parameters, generate new scheduling instructions, and execute them cyclically to achieve adaptive intelligent closed-loop control of building photovoltaic power and building energy management system.

[0055] Furthermore, step 9 includes the following steps:

[0056] The edge computing node receives scheduling instructions from a cloud-based collaborative control platform or generated locally. The scheduling instructions include: target device address, control parameters, and execution time information.

[0057] Edge computing nodes perform validity checks on the received scheduling instructions. The checks include: whether the target device address exists in the local device list, whether the control parameters are within the allowed operating range of the device, and whether the instruction timestamp meets the current timing requirements. Only when the instruction passes all checks can it enter the instruction conversion and issuance process.

[0058] Edge computing nodes convert standardized scheduling instructions into device-specific control messages based on the communication protocol type of the controlled device. Specifically, for devices using the Modbus communication protocol, corresponding function codes and register addresses are generated; for devices using the BACnet communication protocol, corresponding service request structures are constructed.

[0059] The edge computing node sends the converted control message to the target device, and at the same time starts response monitoring and sets a timeout threshold. If no response is received from the device or an error response is received within the set time limit, it is determined that the communication has failed and the exception handling procedure is triggered.

[0060] For controllable devices that successfully receive instructions, their operating status is monitored in real time through edge computing nodes, and the actual output is compared with the target set value. When the deviation between the actual output of the device and the target set value is detected to exceed the preset allowable range, it is determined that there is an execution abnormality. The allowable range is a threshold set based on the device performance specifications.

[0061] Upon detecting a communication failure or execution anomaly, a fault handling procedure is immediately initiated, which includes: retransmitting control commands, recording fault information, and notifying the upper-level control platform.

[0062] Furthermore, the error between the predicted and actual load values, the photovoltaic output prediction error, and the equipment execution deviation are calculated, and the error and deviation data are fed back to each control level, including the following steps:

[0063] The predicted load value within the current time window is compared point by point with the actual load value collected in real time. The mean square error and the mean absolute percentage error are used as the main indicators to calculate the load prediction error and record its time change trend.

[0064] By comparing the photovoltaic power output prediction curve with the actual power generation data, the instantaneous power deviation and cumulative energy deviation are calculated respectively. When the deviation exceeds the preset threshold within three consecutive sampling periods, the short-term correction mechanism of the photovoltaic prediction module is triggered.

[0065] Compare the actual output power of the equipment with the set value of the dispatch command, and calculate the percentage deviation and response lag time;

[0066] After normalizing and weighting the load forecasting error, photovoltaic output forecasting error and equipment execution deviation, a comprehensive deviation index is formed.

[0067] The comprehensive deviation index and individual deviation data are fed back to the corresponding control level in real time, and all feedback data are uploaded to the cloud-based collaborative control platform.

[0068] Compared with the prior art, the present invention has the following advantages:

[0069] This invention achieves adaptive intelligent closed-loop control of building photovoltaic power and energy management system by coupling distributed edge computing nodes with cloud-based collaborative control platform, combining multi-level equipment hierarchical control, real-time load and photovoltaic output prediction and dynamic optimization scheduling. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some embodiments of this invention, but not all embodiments.

[0073] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0075] like Figure 1 As shown, a method for coupling building photovoltaic (PV) power with a building energy management system is disclosed. The building energy management system includes: a PV power generation monitoring module, a load forecasting module, distributed edge computing nodes, a cloud-based collaborative control platform, and a multi-layer equipment scheduling module. The coupling method includes the following steps:

[0076] Step 1: Deploy distributed edge computing nodes to establish a two-way communication connection with the cloud-based collaborative control platform, and configure local autonomous operation mode and offline control strategy in case of communication failure;

[0077] Step 2: Obtain the equipment type, rated power, response time, adjustment range, and operating constraints of all controllable equipment in the building, and construct a capability profile database for each type of controllable equipment.

[0078] Step 3: Based on the response time, adjustment range and operating constraints of each device in the capability profile database, all controllable devices are divided into second-level response devices, minute-level response devices and hour-level response devices. A three-level hierarchical control architecture is established, and corresponding control levels and priority rules are configured for devices with different response levels.

[0079] Step 4: Collect building load data, photovoltaic power generation data and environmental parameters in real time, and combine them with historical operating data to generate load forecast curves and photovoltaic output forecast curves for the next 24 hours by the load forecast module and photovoltaic power generation monitoring module, respectively.

[0080] Step 5: Based on the load forecast curve and the photovoltaic output forecast curve, calculate the optimized scheduling scheme for multiple future time windows through the multi-layer equipment scheduling module, and determine the preliminary scheduling instructions for each controllable device according to the actual operating status at the current moment.

[0081] Step 6: When multiple control levels issue scheduling instructions to the same device at the same time, the final instruction to be executed is automatically determined based on the priority of each instruction and the current operating status of the device.

[0082] Step 7: If there are conflicts between multiple control levels that make the optimization problem infeasible, then the priority weight coefficients of each control level are dynamically adjusted according to the current energy supply and demand tension, the degree of deviation of user comfort and the estimated economic loss.

[0083] Step 8: If no feasible solution is found after adjusting the weights, then, under the premise of always satisfying the hard constraints related to equipment safety, system stability and grid connection specifications, the non-critical constraints related to user comfort and economy are gradually relaxed within their preset allowable deviation range according to the preset relaxation priority; when the scheduling scheme satisfies all hard constraints and the deviation of each non-critical constraint is within the corresponding allowable deviation range, a feasible scheduling scheme is output.

[0084] Step 9: The final execution instruction is sent to the corresponding controllable device through the edge computing node, and the instruction response time, execution deviation and operating status of each device are recorded in real time.

[0085] Step 10: During the equipment execution process, the edge computing node continuously collects actual energy consumption data, photovoltaic power generation data, load change data and equipment operation status data, calculates the error between the load prediction value and the actual value, the photovoltaic output prediction error and the equipment execution deviation, and feeds back the error and deviation data to the corresponding control level.

[0086] Step 11: Based on the error and deviation data, perform online correction on the prediction model parameters of the load prediction module and the photovoltaic power generation monitoring module, and dynamically adjust the weight coefficient adjustment strategy, priority rules and relaxation priority to generate updated control parameters.

[0087] Step 12: Determine whether the next rolling optimization time window has been reached based on the preset rolling optimization cycle. If not, continue to execute the current scheduling scheme, continuously monitor the equipment operating status, and make real-time fine adjustments for sudden load fluctuations or changes in photovoltaic output. If the window has been reached, re-execute steps 4 to 11 based on the updated control parameters, generate new scheduling instructions, and execute them cyclically to achieve adaptive intelligent closed-loop control of building photovoltaic power and building energy management system.

[0088] This method of coupling building photovoltaic (PV) power with a building energy management system enables deep collaboration and dynamic matching between the PV power generation side, the load side, and the control side without altering the existing building's energy consumption structure and equipment physical form. It systematically improves operational stability, energy utilization efficiency, scheduling precision, and long-term adaptive capabilities. By introducing distributed edge computing nodes with local autonomy on the building side, the method allows the building energy management system to fully utilize global collaborative optimization capabilities when cloud communication conditions are good. Even under communication limitations or anomalies, it can maintain continuous and controllable local operation, effectively reducing the risk of dependence on a single communication link and centralized control platform, and significantly improving the reliability and fault tolerance of the building energy system in complex operating environments.

[0089] By modeling the operational characteristics of various controllable devices within a building and constructing a capability profile database that includes response time, adjustment range, and operational constraints, the physical characteristics and control potential of different devices can be accurately characterized. Based on this, the devices are divided into multiple control levels according to their response time scale, enabling fast-response devices, medium-speed adjustment devices, and slow-speed adjustment devices to participate in scheduling within their respective appropriate time scales. This avoids the control lag, frequent adjustments, or execution deviations caused by dynamic differences in devices, which are common in traditional unified scheduling methods. Consequently, it significantly improves the consistency and predictability between control commands and actual device behavior.

[0090] By introducing a multi-time-window optimized scheduling mechanism based on load forecasting and photovoltaic (PV) output forecasting, future operating trends and real-time status changes can be comprehensively considered during the scheduling decision-making stage, thereby improving the foresight and overall coordination of the scheduling scheme. This approach effectively reduces the passive adjustment needs caused by fluctuations in PV output or sudden load changes, increases the proportion of PV power absorbed within buildings, reduces dependence on the external power grid and the probability of curtailment, and enables building energy systems to maintain a more stable and efficient operating state while meeting their own energy needs.

[0091] When multiple layers of control operate in parallel on the same device, priority adjudication and conflict resolution mechanisms can automatically complete instruction screening and decision convergence, avoiding operational uncertainties caused by the superposition or mutual cancellation of control instructions. When there are conflicts between scheduling objectives that render the optimization problem infeasible, a dynamic weight adjustment mechanism based on the degree of energy supply and demand tension, the degree of deviation in user comfort, and the prediction of economic losses can be introduced. This allows for adaptive reconstruction of the control focus according to the current operating situation, thereby achieving a dynamic balance between safety, comfort, and economy in different operating scenarios.

[0092] Building upon the above, by tiered relaxation of non-critical constraints, the scheduling problem can still output a feasible solution while strictly meeting hard constraints such as equipment safety, system stability, and grid connection standards. This mechanism avoids scheduling failures or system shutdowns caused by overly rigid constraints, improves the operational continuity and disturbance resistance of building energy management systems under extreme load fluctuations or abnormal photovoltaic output, and thus maintains a stable and controllable operating state under complex conditions.

[0093] By continuously collecting and feeding back data on response time, execution deviation, and operating status during equipment execution, online correction of prediction models and scheduling parameters can be performed, gradually reducing the cumulative impact of model errors and environmental changes on control effectiveness. Combined with rolling optimization and real-time fine-tuning mechanisms, while maintaining the overall stability of the scheduling strategy, it possesses rapid response capabilities to sudden load changes and photovoltaic output fluctuations, thereby avoiding adverse effects on equipment lifespan and user experience caused by significant control adjustments.

[0094] In summary, this coupling method, by forming a closed-loop linkage between system architecture, control strategy, and optimization mechanism, enables the stochasticity of building photovoltaic power and the time-varying nature of building load to be coordinated and handled under a unified control framework. It realizes the transformation of building energy management system from static control to dynamic adaptive control, and has formed significant and synergistic technical effects in improving energy utilization efficiency, enhancing system operation stability, reducing long-term operating costs, and supporting the goal of low-carbon building operation.

[0095] Furthermore, step 1 includes the following steps:

[0096] The installation location and number of each edge computing node are determined based on the actual spatial layout of the building and the density of equipment distribution, so that the communication latency within the coverage area of ​​each node does not exceed 200 milliseconds.

[0097] Configure an independent data cache space for each edge computing node;

[0098] Establish an encrypted two-way communication channel between each edge computing node and the cloud-based collaborative control platform;

[0099] Set offline control trigger conditions for each edge computing node; the offline control trigger conditions are: when the edge computing node fails to communicate with the cloud platform 5 times in a row or the single communication delay exceeds 10 seconds, it automatically switches to local autonomous operation mode and independently executes device scheduling and energy consumption control tasks based on local cached historical data and preset rule base.

[0100] Once communication between the offline edge computing node and the cloud platform is restored, the offline edge computing node proactively uploads its offline operation records and control decision data to the cloud platform.

[0101] As can be seen from the above, the deployment of edge computing nodes and the establishment of offline control mechanisms enable the building energy management system to maintain continuous and stable control capabilities even under fluctuating communication conditions or cloud unavailability, significantly reducing the risk of control failure caused by network latency or interruption. Each node independently undertakes local data processing and control decision-making tasks within a limited communication latency range, effectively shortening control link response time and improving the real-time performance and accuracy of equipment scheduling. Through local data caching and data backhaul mechanisms after communication recovery, the integrity and traceability of operational data and control decisions are ensured. While guaranteeing safe and stable operation, this approach balances centralized optimization and distributed autonomy, comprehensively enhancing the reliability, robustness, and operational continuity of the building energy system.

[0102] Furthermore, step 2 includes the following steps:

[0103] Basic parameter acquisition: Collect basic information of each controllable device, and obtain real-time operating data of the device; the basic information includes: rated power marked on the nameplate, device type and operating mode;

[0104] Energy consumption characteristics statistical analysis; based on actual operating data collected continuously by edge computing nodes and stored in local or cloud collaborative control platforms for no less than 7 days, statistical analysis is performed on the power changes and energy consumption levels of each controllable device under different load rates;

[0105] Rated parameter correction: Based on the statistical analysis results of each controllable device, an energy consumption curve characterizing the actual operating characteristics is constructed, and the energy consumption curve is compared with the rated parameters marked on the nameplate. The rated power is corrected according to the comparison results to form correction parameters that reflect the actual energy consumption characteristics of the corresponding controllable device.

[0106] Response performance measurement; When implementing operational status control on controllable equipment, the cloud-based collaborative control platform records the time from the effective control to the stable operation of the equipment, in order to quantify the dynamic response performance of the equipment;

[0107] The adjustment characteristics are collected; for stepped adjustment equipment, the power value corresponding to each step is recorded; for continuous adjustment equipment, the minimum adjustment step size and adjustment rate are recorded to characterize the control accuracy and adjustment sensitivity of the equipment.

[0108] Operational constraint parameter recording: Collect the operational constraints of each controllable device; the operational constraints include: the minimum continuous operating time, the minimum downtime interval, and the recommended maximum number of daily start-ups and shutdowns as indicated in the device's factory technical parameters and operating specifications;

[0109] Structured modeling and storage: The acquired basic parameters, energy consumption characteristics, corrected rated power, response performance, regulation characteristics and operating constraints are structured and stored to build a database of controllable equipment capability profiles.

[0110] As can be seen from the above, by systematically collecting and analyzing the basic parameters, actual energy consumption behavior, and dynamic response characteristics of controllable equipment, the operating characteristics of the equipment are transformed from static nominal parameters into a quantitative description that closely reflects real operating conditions, significantly reducing the risk of scheduling misjudgments caused by parameter deviations. The corrected energy consumption parameters and response characteristics can accurately reflect the actual performance of the equipment under different loads and control conditions, making scheduling decisions more predictable and executable. Through unified modeling of adjustment accuracy and operating constraints, the safety and rationality of equipment participation in collaborative scheduling are enhanced, providing a reliable and precise parameter basis for subsequent hierarchical control and optimized scheduling.

[0111] Furthermore, step 3 includes the following steps:

[0112] Devices with a response time of less than 30 seconds are classified as second-level response devices and assigned to the first control level, with a control cycle set to 10 seconds, to cope with rapid fluctuations in photovoltaic output and load;

[0113] Devices with response times between 30 seconds and 10 minutes are classified as minute-level response devices and assigned to the second control level, with a control cycle set to 5 minutes, to regulate short-term load trends.

[0114] Devices with a response time exceeding 10 minutes are classified as hourly response devices and assigned to the third control level, with a control cycle set to 15 minutes for medium- and long-term economic operation optimization.

[0115] Priority rules are configured for each control level: when the power deviation exceeds the threshold, the first control level intervenes and overrides the existing scheduling plans of the second and third control levels; when the first control level is not triggered or after intervention, the second control level adjusts the equipment according to the coordination results with the first control level; the third control level generates a plan in advance based on the prediction information and executes it without being overridden by the higher priority level. Each level achieves collaborative control through sharing status information.

[0116] As can be seen from the above, by managing controllable equipment in layers according to response speed and matching differentiated control cycles, rapid adjustment capabilities can be prioritized to suppress instantaneous fluctuations in photovoltaic output and load, preventing the expansion of system power deviation. Simultaneously, medium- and slow-speed equipment can participate in adjustment within a more suitable time scale, reducing ineffective intervention and frequent start-stop cycles. The hierarchical priority and coverage mechanism ensures the determinism and convergence of control commands under sudden fluctuations, coordinating short-term stability control with medium- and long-term economic optimization, thereby improving overall scheduling efficiency, operational stability, and the rationality of resource utilization.

[0117] Furthermore, the load forecasting module includes:

[0118] The basic forecasting layer uses a long short-term memory neural network model to process the temporal characteristics and periodic patterns of the load. The input features include historical load data of the past 168 hours, date type identifiers, weather forecast information, and building usage plans.

[0119] The short-term correction layer uses an exponential smoothing strategy to make recent corrections to the basic forecast results and dynamically adjusts the forecast curve based on the actual load deviation in the most recent 24 hours.

[0120] As can be seen from the above, this load forecasting module improves the accuracy and adaptability of building load forecasting through a hierarchical structure. The basic forecasting layer utilizes a long short-term memory neural network model to perform deep learning on the time-series characteristics, periodic patterns, and external factors of the load. This fully captures the complex nonlinear relationship between building load and changes in weather and usage patterns, thereby generating basic forecast results with high fitting accuracy. The short-term correction layer, building upon this, introduces an exponential smoothing strategy to dynamically correct the forecast results. This allows the model to quickly self-adjust based on the actual load deviation of the latest 24 hours, significantly reducing forecast errors caused by sudden changes in usage or the environment.

[0121] Furthermore, generating the photovoltaic output prediction curve includes the following steps:

[0122] The environmental parameters of the area where the photovoltaic array is located are obtained based on numerical weather forecasts, and the irradiance of the tilted surface of the photovoltaic module is calculated to obtain the actual solar energy intensity data received by the photovoltaic module.

[0123] By combining the installed capacity, historical efficiency data and current operating status of the photovoltaic array, the expected power output of each photovoltaic module is calculated, and a preliminary photovoltaic power output prediction curve is generated.

[0124] When the average deviation between the actual irradiance and the predicted value in the last 30 minutes exceeds 30%, the latest cloud map data and short-term weather forecast are called to recalculate the irradiance of the tilted surface, and a corrected photovoltaic output prediction curve is generated based on the photovoltaic array parameters.

[0125] As can be seen from the above, by combining numerical weather prediction, irradiance calculation, and the operating characteristics of photovoltaic arrays, photovoltaic output prediction can more accurately reflect the actual light-receiving conditions and power generation status of the modules, reducing the systematic deviation between the prediction results and the actual output. When environmental changes cause a significant increase in prediction errors, the prediction results can be dynamically corrected by introducing cloud images and short-term weather information, enhancing the sensitivity and adaptability of the prediction curve to sudden weather changes, thereby improving the stability, reliability, and support capability of photovoltaic output prediction for dispatching decisions.

[0126] Furthermore, based on the load forecast curve and the photovoltaic output forecast curve, the multi-layer equipment scheduling module calculates an optimized scheduling scheme for multiple future time windows, including the following steps:

[0127] The next 24 hours are divided into multiple time windows, each lasting 15 minutes;

[0128] For each time window, an optimization problem is constructed based on load forecasting and photovoltaic output forecasting, covering economic objectives, energy utilization efficiency objectives, and comfort constraints. The objective function terms corresponding to each objective are combined according to preset weights and transformed into a single-objective optimization problem.

[0129] Multi-dimensional operation constraint rules are introduced into the optimization scheme for each time window. These multi-dimensional operation constraint rules impose restrictions from the aspects of power balance, equipment operating boundary, energy storage state of charge and grid connection point power.

[0130] Solving the single-objective optimization problem yields the power setpoints, start / stop states, and energy storage charging / discharging strategies for each controllable device within each time window. Specifically, day-ahead scheduling employs an accurate algorithm to generate a globally optimized scheduling strategy, while intraday rolling scheduling uses a fast heuristic algorithm to generate a real-time scheduling strategy that satisfies the constraint rules. These scheduling strategies are then passed to the multi-layer device scheduling module for execution.

[0131] As can be seen from the above, by refining the forecast period into multiple short-term windows and introducing multi-objective collaborative optimization, equipment scheduling can simultaneously consider economy, energy utilization efficiency, and operational constraints, achieving more refined and continuous power allocation in the time dimension. The unified introduction of multi-dimensional operational constraints ensures the executability of scheduling results in terms of power balance, equipment safety, energy storage status, and grid connection requirements. Combining day-ahead global optimization with intraday rapid rolling adjustments, the scheduling strategy possesses both overall optimality and timely response to operational deviations, thereby improving the operational stability, flexibility, and comprehensive optimization level of the building energy system.

[0132] Furthermore, step 6 includes the following steps:

[0133] Edge computing nodes receive scheduling instructions from various control levels and collect the urgency level identifier and instruction type of each instruction, wherein the urgency level includes: regular instructions, urgent instructions and emergency instructions;

[0134] The received dispatch instructions are initially prioritized: if there are emergency instructions, they are directly used as the final execution instructions; if all instructions are regular or urgent instructions, the initial priority of the instructions is first determined according to the control level, and the execution order is determined according to the time sequence within the same control level.

[0135] When a scheduling instruction exceeds the current controllable range of the device, the edge computing node corrects it to ensure that the instruction remains within the device's executable range.

[0136] Edge computing nodes select instructions that meet high-priority triggering conditions and are within a controllable range as the final execution instructions.

[0137] As can be seen from the above, by hierarchically identifying and locally adjudicating multi-source scheduling commands, the equipment can still form a unique, clear, and executable control result when receiving multiple layers of control commands simultaneously, avoiding command conflicts and execution uncertainties. The emergency priority and hierarchical sorting mechanism ensures rapid response capabilities under abnormal operating conditions, while the controllable range correction of commands prevents unreasonable control from impacting the equipment, thereby improving the safety, determinism, and real-time reliability of scheduling execution.

[0138] Furthermore, the relaxation priorities, from low to high, are: non-core area lighting illuminance constraints, non-critical period indoor temperature comfort constraints, large equipment economic operation range constraints, energy storage system cycle life protection constraints, and interruptible load response time constraints; after each level of constraint is relaxed, the optimization solution is re-executed, and the subsequent relaxation operation is terminated when a feasible solution is obtained.

[0139] As can be seen from the above, by setting relaxation priorities from low to high, an orderly constraint relaxation strategy is achieved in situations where scheduling is infeasible or conflicting. Priority is given to relaxing constraints with minimal impact on user comfort and non-core energy consumption, such as lighting in non-core areas and temperature during non-critical periods. Then, the economic operating range of equipment, energy storage cycle life, and interruptible load response time are considered sequentially to ensure that core safety and critical operating conditions are not affected. After each level of constraint relaxation, the optimization solution is re-executed to find a feasible scheduling scheme with minimal compromises, thereby guaranteeing the continuity, feasibility, and overall economic efficiency and stability of energy scheduling.

[0140] Furthermore, step 9 includes the following steps:

[0141] The edge computing node receives scheduling instructions from a cloud-based collaborative control platform or generated locally. The scheduling instructions include: target device address, control parameters, and execution time information.

[0142] Edge computing nodes perform validity checks on the received scheduling instructions. The checks include: whether the target device address exists in the local device list, whether the control parameters are within the allowed operating range of the device, and whether the instruction timestamp meets the current timing requirements. Only when the instruction passes all checks can it enter the instruction conversion and issuance process.

[0143] Edge computing nodes convert standardized scheduling instructions into device-specific control messages based on the communication protocol type of the controlled device. Specifically, for devices using the Modbus communication protocol, corresponding function codes and register addresses are generated; for devices using the BACnet communication protocol, corresponding service request structures are constructed.

[0144] The edge computing node sends the converted control message to the target device, and at the same time starts response monitoring and sets a timeout threshold. If no response is received from the device or an error response is received within the set time limit, it is determined that the communication has failed and the exception handling procedure is triggered.

[0145] For controllable devices that successfully receive instructions, their operating status is monitored in real time through edge computing nodes, and the actual output is compared with the target set value. When the deviation between the actual output of the device and the target set value is detected to exceed the preset allowable range, it is determined that there is an execution abnormality. The allowable range is a threshold set based on the device performance specifications.

[0146] Upon detecting a communication failure or execution anomaly, a fault handling procedure is immediately initiated, which includes: retransmitting control commands, recording fault information, and notifying the upper-level control platform.

[0147] As can be seen from the above, by performing complete verification, protocol adaptation, and execution monitoring of scheduling instructions, risks such as address errors, parameter out-of-bounds errors, and timing mismatches can be eliminated before control instructions are issued, significantly improving the accuracy and success rate of instruction execution. The local conversion mechanism based on protocol differences enhances compatibility with heterogeneous devices, while response monitoring and deviation judgment enable timely identification and handling of communication and execution anomalies, thereby ensuring the reliability, controllability, and overall stability of the equipment control process and the overall system operation.

[0148] Furthermore, based on actual energy consumption data, smart meters or power sensors are installed in key circuits of the building's power distribution system, with a sampling period set to 1 to 5 seconds, to obtain the electrical parameters of each circuit in real time.

[0149] For photovoltaic power generation data, DC side voltage and current, AC side voltage and current, real-time power and cumulative power generation parameters are read from the communication interface of the photovoltaic inverter. The sampling period is usually 5 to 10 seconds.

[0150] In addition to collecting the total electrical load of the building, the system also performs sub-metering of major energy-consuming systems, including air conditioning systems, lighting sockets, power equipment and special electrical equipment, in order to achieve real-time monitoring of the load structure.

[0151] For equipment operation status data, the device's start / stop status, operating mode, output parameters, and fault alarm information are collected through the device's built-in communication port.

[0152] As can be seen from the above, by implementing high-frequency, multi-source data acquisition in power distribution circuits, photovoltaic side, and key energy-consuming equipment, the energy flow, power generation status, load structure, and equipment operating conditions of the building energy system can be reflected synchronously and in fine detail, significantly improving the real-time and completeness of operational status perception. The combination of sub-metering and condition monitoring creates a clear mapping between energy consumption changes and specific equipment behavior, providing a reliable data foundation for identifying operational deviations, scheduling corrections, and subsequent model calibration, thereby enhancing the accuracy and traceability of energy management and control decisions.

[0153] Furthermore, the error between the predicted and actual load values, the photovoltaic output prediction error, and the equipment execution deviation are calculated, and the error and deviation data are fed back to each control level, including the following steps:

[0154] The predicted load value within the current time window is compared point by point with the actual load value collected in real time. The mean square error and the mean absolute percentage error are used as the main indicators to calculate the load prediction error and record its time change trend.

[0155] By comparing the photovoltaic power output prediction curve with the actual power generation data, the instantaneous power deviation and cumulative energy deviation are calculated respectively. When the deviation exceeds the preset threshold within three consecutive sampling periods, the short-term correction mechanism of the photovoltaic prediction module is triggered.

[0156] Compare the actual output power of the equipment with the set value of the dispatch command, and calculate the percentage deviation and response lag time;

[0157] After normalizing and weighting the load forecasting error, photovoltaic output forecasting error and equipment execution deviation, a comprehensive deviation index is formed.

[0158] The comprehensive deviation index and individual deviation data are fed back to the corresponding control level in real time, and all feedback data are uploaded to the cloud-based collaborative control platform.

[0159] As can be seen from the above, by quantitatively calculating and uniformly integrating the deviations in load, photovoltaic output, and equipment execution results, the sources and extent of prediction inaccuracies and execution deviations can be identified in a timely manner, enhancing the objectivity and continuity of operational status assessment. The deviation triggering and feedback mechanism enables the prediction model and control strategy to have rapid self-correction capabilities, avoiding the amplified impact of long-term error accumulation on scheduling effectiveness, thereby improving overall control accuracy, operational stability, and the system's adaptive capability under dynamic operating conditions.

[0160] Finally, 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 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 method for coupling a building photovoltaic electric energy and a building energy management system, characterized in that, The building energy management system comprises a photovoltaic power generation monitoring module, a load prediction module, a distributed edge computing node, a cloud collaborative control platform and a multi-layer device scheduling module; the coupling method comprises the following steps: Step 1, deploying a distributed edge computing node, establishing a two-way communication connection with the cloud collaborative control platform, configuring a local autonomous operation mode and an offline control strategy in the event of communication failure; Step 2, obtaining the device type, rated power, response time, adjustment capability range and operation constraint condition of all controllable devices in the building, and constructing a capability profile database of various controllable devices; Step 3, according to the response time, adjustment capability range and operation constraint condition of each device in the capability profile database, dividing all controllable devices into second-level response devices, minute-level response devices and hour-level response devices, establishing a three-layer hierarchical control architecture, and configuring corresponding control levels and priority rules for devices of different response levels; Step 4, real-time collection of building load data, photovoltaic power generation data and environmental parameters, combined with historical operation data, generating a 24-hour load prediction curve and a photovoltaic output prediction curve by the load prediction module and the photovoltaic power generation monitoring module respectively; Step 5, based on the load prediction curve and the photovoltaic output prediction curve, calculating the optimal scheduling scheme of multiple time windows in the future by the multi-layer device scheduling module, and determining the preliminary scheduling instruction of each controllable device according to the actual operation state at the current time; Step 6, when multiple control levels simultaneously issue scheduling instructions to the same device, the final execution instruction is automatically determined according to the priority of each instruction and the current operation state of the device; Step 7, if there is a conflict between multiple control levels resulting in no feasible solution to the optimization problem, then according to the current energy supply and demand tension, the user comfort deviation and the economic loss estimate, dynamically adjusting the priority weight coefficient of each control level; Step 8, if there is still no feasible solution after adjusting the weight, then under the premise of always meeting the hard constraint conditions related to device safety, system stability and grid connection specifications, relaxing the non-critical constraint conditions related to user comfort and economy in the preset allowable deviation range according to the preset relaxation priority; when the scheduling scheme meets all the hard constraint conditions and the deviation of each non-critical constraint condition is within the corresponding allowable deviation range, output the feasible scheduling scheme.

2. The method of claim 1, wherein the method further comprises: The step 1 comprises the following steps: According to the actual spatial layout and device distribution density of the building, determine the installation position and number of each edge computing node, so that the communication delay in the coverage range of each node does not exceed 200 milliseconds; Configure an independent data cache space for each edge computing node; Establish an encrypted two-way communication channel between each edge computing node and the cloud collaborative control platform; Set the trigger condition for offline control for each edge computing node; the trigger condition for offline control is: when it is detected that the edge computing node fails to communicate with the cloud platform for 5 times in a row or the single communication delay exceeds 10 seconds, automatically switch to the local autonomous operation mode, and independently execute device scheduling and energy consumption control tasks based on the locally cached historical data and the preset rule library; When the communication between the offline edge computing node and the cloud platform is restored, the offline edge computing node actively uploads the running records and control decision data during the offline period to the cloud platform.

3. The method of claim 1, wherein the method further comprises: Step 2 includes the following steps: Basic parameter acquisition: acquiring basic information of each controllable device, and obtaining real-time running data of the device; the basic information includes rated power, device type and running mode marked on the nameplate; Energy consumption characteristic statistical analysis; based on the actual running data continuously collected by the edge computing node and stored in the local or cloud collaborative control platform for not less than 7 days, the power variation and energy consumption level of each controllable device under different load rates are statistically analyzed; Rated parameter correction; based on the statistical analysis results of each controllable device, an energy consumption curve representing the actual running characteristics is constructed, and the energy consumption curve is compared with the rated parameters marked on the nameplate, and the rated power is corrected according to the comparison result, forming a correction parameter reflecting the true energy consumption characteristics of the corresponding controllable device; Response performance measurement; when the running state of the controllable device is regulated, the cloud collaborative control platform records the time from the regulation taking effect to the running state reaching stability, so as to quantify the dynamic response performance of the device; Adjustment characteristic acquisition; for stepwise adjustment devices, the power value corresponding to each gear is recorded; for continuous adjustment devices, the minimum adjustment step and adjustment rate are recorded to characterize the control accuracy and adjustment sensitivity of the device; Running constraint parameter recording: acquiring the running constraint conditions of each controllable device; the running constraint conditions include the minimum continuous running time, minimum shutdown interval time and the upper limit of the recommended daily start-stop times marked in the factory technical parameters and operation specifications of the device; Structured modeling and storage: the obtained basic parameters, energy consumption characteristics, corrected rated power, response performance, adjustment characteristics and running constraint conditions are structured and stored to construct a controllable device capability portrait database.

4. The method of claim 1, wherein the method further comprises: Step 3 includes the following steps: Devices with a response time less than 30 seconds are classified as second-level response devices and assigned to the first control level, with a control period set to 10 seconds, for responding to rapid fluctuations in photovoltaic output and load; Devices with a response time between 30 seconds and 10 minutes are classified as minute-level response devices and assigned to the second control level, with a control period set to 5 minutes, for adjusting short-term changes in load; Devices with a response time exceeding 10 minutes are classified as hour-level response devices and assigned to the third control level, with a control period set to 15 minutes, for medium and long-term economic operation optimization; Priority rules are configured for each control level: when the power deviation exceeds the threshold, the first control level intervenes and overrides the existing scheduling plan of the second and third control levels; after the first control level is not triggered or intervention is completed, the second control level adjusts the device according to the coordination results of the first control level; the third control level generates a plan in advance based on prediction information and executes it without being overridden by high-priority levels, and each level achieves collaborative control through state information sharing.

5. The method of claim 1, wherein the method further comprises: Generating the photovoltaic output prediction curve includes the following steps: According to the numerical weather prediction, the environmental parameters of the region where the photovoltaic array is located are obtained, and the irradiance of the inclined surface of the photovoltaic module is calculated to obtain the solar intensity data actually received by the photovoltaic module; In combination with the installed capacity of the photovoltaic array, historical efficiency data and current operating state, the expected power generation output of each unit photovoltaic module is calculated, and a preliminary photovoltaic output prediction curve is generated by summarizing. When the average deviation between the actual irradiance intensity and the predicted value in the last 30 minutes is more than 30%, the latest cloud image data and short-term weather forecast are called to recalculate the irradiance of the inclined surface, and the corrected photovoltaic output prediction curve is generated based on the photovoltaic array parameters.

6. The method of claim 1, wherein the method further comprises: Based on the load prediction curve and the photovoltaic output prediction curve, the multi-layer equipment scheduling module is used to calculate the optimized scheduling scheme in multiple time windows in the future, including the following steps: Divide the next 24 hours into multiple time windows, each with a duration of 15 minutes; For each time window, based on the load prediction and photovoltaic output prediction, an optimization problem is constructed to cover economic targets, energy utilization efficiency targets and comfort constraint targets, and the target function items corresponding to the targets are combined according to the preset weights to form a single-target optimization problem; In the optimization scheme of each time window, multi-dimensional operation limit rules are introduced, which limit from the aspects of power balance, equipment operation boundary, energy storage state of charge and grid-connected point power respectively; Solve the single-target optimization problem to obtain the power set value, start-stop state and energy storage charging-discharging strategy of each controllable device in each time window; wherein, the day-ahead scheduling adopts an accurate algorithm to generate a globally optimized scheduling strategy, the intra-day rolling scheduling adopts a fast heuristic algorithm to generate a real-time scheduling strategy that meets the limit rules, and the scheduling strategy is transmitted to the multi-layer equipment scheduling module for execution.

7. The method of claim 1, wherein the method further comprises: The step 6 includes the following steps: The edge computing node receives scheduling instructions from each control level and collects the emergency level identifier and instruction type of each instruction, wherein the emergency level includes: regular instruction, urgent instruction and emergency instruction; Preliminary priority determination is performed on the received scheduling instructions: if there is an emergency instruction, it is directly used as the final execution instruction; if all instructions are regular instructions or urgent instructions, the preliminary priority of the instructions is determined according to the control level, and the execution order is determined according to the time sequence in the same control level; When the scheduling instruction exceeds the current controllable range of the equipment, the edge computing node modifies it to ensure that the instruction is within the executable range of the equipment; The edge computing node selects the instruction that meets the high-priority trigger condition and is within the controllable range as the final execution instruction.

8. The method of claim 1-7, wherein the method further comprises: The coupling method further includes the following steps: Step 9: The final execution instruction is sent to the corresponding controllable equipment through the edge computing node, and the instruction response time, execution deviation and running state of each device are recorded in real time. Step 10, during the device execution process, the edge computing node continuously collects actual energy consumption data, photovoltaic power generation data, load change data and device operation state data, calculates the error between the load prediction value and the actual value, the photovoltaic output prediction error and the device execution deviation, and feeds back the error and deviation data to the corresponding control level; Step 11, according to the error and deviation data, the prediction model parameters of the load prediction module and the photovoltaic power generation monitoring module are corrected online, and the weight coefficient adjustment strategy, priority rule and relaxation priority are dynamically adjusted to generate updated control parameters; Step 12, according to the preset rolling optimization period to judge whether the next rolling optimization time window has arrived: if not, continue to execute the current scheduling scheme, continuously monitor the device operation state, and make real-time fine adjustment to the sudden load fluctuation or photovoltaic output change; if it has arrived, re-execute steps 4 to 11 based on the updated control parameters to generate new scheduling instructions and execute in a loop, thereby realizing adaptive intelligent closed-loop control of building photovoltaic electric energy and building energy management system.

9. The method of claim 8, wherein the method further comprises: The step 9 comprises the following steps: The edge computing node receives the scheduling instructions from the cloud collaborative control platform or locally generated, which includes target device address, control parameters and execution time information; The edge computing node performs validity check on the received scheduling instructions, and the check content includes whether the target device address exists in the local device list, whether the control parameters are within the allowed operating range of the device, and whether the instruction timestamp meets the current timing requirements; only when the instruction passes all checks, the instruction conversion and issuing process is entered; The edge computing node converts the standardized scheduling instruction into a device-specific control message according to the communication protocol type of the controlled device; for devices using Modbus communication protocol, corresponding function code and register address are generated; for devices using BACnet communication protocol, corresponding service request structure is constructed; The edge computing node issues the converted control message to the target device, starts response monitoring and sets a timeout threshold, and when no device response is received or an error response is received within the set time limit, it is determined that the communication has failed, and an exception handling program is triggered; For controllable devices that successfully receive instructions, the edge computing node monitors their running state in real time and compares the actual output with the target set value. When the deviation between the actual output and the target set value of the device exceeds the preset allowed range, it is determined that there is an execution exception, wherein the allowed range is a threshold set based on the device performance specification; After detecting communication failure or execution exception, the fault handling program is immediately started, which includes resending control instructions, recording fault information and notifying the upper control platform.

10. The method of claim 8, wherein the method further comprises: Calculate the error between the load prediction value and the actual value, the photovoltaic output prediction error and the device execution deviation, and feed back the error and deviation data to each control level, including the following steps: The predicted load value in the current time window is compared with the actual load value collected in real time point by point, the mean square error and the average absolute percentage error are used as the main indicators, the load prediction error is calculated and its time variation trend is recorded; The photovoltaic output prediction curve is compared with the actual power generation data, the instantaneous power deviation and the cumulative energy deviation are calculated respectively, and when it is detected that the deviation exceeds the preset threshold in three consecutive sampling periods, the short-term correction mechanism of the photovoltaic prediction module is triggered; The difference between the actual output power of the equipment and the set value of the scheduling instruction is compared, and the deviation percentage and the response lag time are calculated; The load prediction error, the photovoltaic output prediction error and the equipment execution deviation are normalized and then weighted and fused to form a comprehensive deviation index; The comprehensive deviation index and each single deviation data are fed back to the corresponding control level in real time, and all the feedback data are uploaded to the cloud collaborative control platform.

Images