A multi-energy management system, method, device, and computer storage medium

Abstract

This invention discloses a multi-energy management system, method, device, and computer storage medium. The system includes multiple energy subsystems, an environmental monitoring module, a BIM system, and a central control platform. The central control platform comprises an edge computing layer and a cloud analytics layer. The edge computing layer is responsible for real-time acquisition, parsing, and rapid control of local data. The cloud analytics layer deeply integrates multi-source energy data, environmental parameters, and BIM information, uses machine learning models such as LSTM for prediction, and generates cross-system collaborative scheduling strategies through multi-objective optimization algorithms before distributing them to the edge computing layer for execution. Simultaneously, the system achieves predictive early warning and fault location by constructing equipment health benchmark models and calculating health deviation indices. This invention solves the problems of response delay, difficulties in multi-energy coordination, and passive operation and maintenance inherent in traditional centralized platforms, improving the system's economy, reliability, and intelligence.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, and in particular to a multi-energy management system, method, device, and computer storage medium. Background Technology

[0002] Traditional energy management platforms generally employ a centralized processing architecture, making it difficult to achieve real-time coordination and closed-loop control of various heterogeneous energy devices such as power, photovoltaics, energy storage, HVAC, and charging piles. These systems have significant limitations in terms of real-time data processing, control response speed, and dynamic strategy optimization capabilities: Firstly, because all data must be uploaded to the cloud for centralized calculation and decision-making, network transmission and processing delays make it difficult for the system to respond quickly to localized emergencies. Secondly, each energy subsystem often operates independently, lacking optimization strategies based on real-time coupling of global status and multi-dimensional information, failing to achieve "source-grid-load-storage" coordination, and easily leading to energy waste and low operational economics. For example, during peak photovoltaic power generation periods, if energy storage charging and load adjustment strategies are not synchronized, surplus electricity may be fed into the grid or curtailed.

[0003] Furthermore, existing systems generally rely on fixed rules or post-event alarm mechanisms, resulting in rigid strategies and passive maintenance. Their shortcomings are mainly reflected in the following aspects: First, the control logic struggles to adapt to multi-dimensional dynamic changes such as weather, load, and equipment status, lacking sufficient intelligence; second, historical operating data is mostly used for report statistics, failing to drive strategy self-optimization; third, environmental parameters (such as temperature, humidity, and irradiance) are analyzed separately from energy consumption data, unable to support precise adjustments for systems such as HVAC; fourth, there is a lack of predictive maintenance capabilities based on data correlation, failing to identify potential faults in advance through the fusion analysis of equipment operating data and environmental parameters, still relying on periodic inspections and post-event maintenance, posing challenges to system reliability.

[0004] With the increasing penetration of renewable energy and the diversification of energy system forms, the traditional architecture mentioned above can no longer meet the higher requirements for real-time performance, collaboration, and intelligence. Although some technical solutions have attempted to integrate multi-energy data, there are still shortcomings in key aspects such as the depth of data fusion, the dynamic generation of real-time collaboration strategies, and the closed-loop execution of "cloud-edge" collaboration. In particular, there is a lack of a complete solution that deeply integrates equipment operation data, environmental parameters, and building information modeling (BIM) to achieve system-level energy efficiency optimization and accurate predictive maintenance. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a multi-energy management system, method, device, and computer storage medium capable of multi-energy data fusion analysis, dynamic strategy generation, edge-cloud collaborative execution, and predictive maintenance. This system integrates various energy devices such as photovoltaics, energy storage, charging piles, and HVAC systems, and utilizes machine learning and building information models to achieve global optimization scheduling and predictive intelligent maintenance.

[0006] One objective of this invention is to provide a multi-energy management system, comprising: Multiple energy subsystem modules are used to provide or consume energy and collect their own operational data; The environmental monitoring module is used to collect environmental parameters; BIM systems are used to provide virtual 3D models that include equipment space and attribute information; The central control platform includes an edge computing layer and a cloud analytics layer; And a distributed soft bus, used to enable data interconnection between modules; The edge computing layer is deployed on-site and includes a data access and parsing module and a local decision execution module, which are used to collect and parse data in real time and execute rapid control of local devices. The cloud-based analytics layer is deployed on a remote server and includes a data integration and analysis module, a machine learning analysis module, an optimization scheduling module, and an intelligent fault diagnosis module. The data integration and analysis module aggregates and stores data from the edge computing layer and integrates it with BIM static information to construct a unified multi-dimensional feature representation. The machine learning analysis module is used for modeling and training based on the multi-dimensional feature representation to predict key power parameters of the system. The optimization scheduling module, based on the prediction results of the machine learning analysis module and the coupled model, calculates the globally optimal energy scheduling scheme with energy efficiency and economy as comprehensive objectives and generates a collaborative optimization strategy. The intelligent fault diagnosis module is used to identify potentially faulty equipment.

[0007] Furthermore, the data integration and analysis module adopts an attention-based fusion method to dynamically weight and fuse multi-source features from the energy subsystem module, environmental monitoring module, BIM system, and historical data. It also uses graph convolutional networks to extract deep features of the spatial topology of equipment in the BIM system in order to construct a unified multi-dimensional feature representation. The machine learning analysis module is based on multidimensional feature representation for modeling and training, and uses a long short-term memory network to predict key system power parameters such as photovoltaic output and system load. The optimization scheduling module, based on the prediction results of the machine learning analysis module and the coupled model, considers the total system operating cost. Minimize, energy efficiency Maximize and improve efficiency With maximizing the overall objective, calculate the globally optimal energy dispatch scheme and generate a collaborative optimization strategy; The intelligent fault diagnosis module constructs a device health benchmark model based on Gaussian process regression and calculates the device health deviation index to identify potentially faulty devices.

[0008] Furthermore, the energy subsystem module includes at least two of the following: a photovoltaic power generation module, an energy storage module, a charging pile module, and a heat pump module. Each module is equipped with an independent local controller to achieve its own intelligent management.

[0009] Furthermore, the optimization scheduling module uses a multi-objective optimization model to calculate the globally optimal energy scheduling scheme; Photovoltaic output predicted by a multi-objective optimization model and a machine learning analysis module and system load As input parameters, and based on a coupled model that includes power balance constraints, energy storage system operation constraints and equipment operation constraints; The objective function of the multi-objective optimization model is: ; in, The total operating cost of the system, For system energy efficiency, For system efficiency, γ, α, and β are the weighting coefficients of cost target, energy efficiency target, and efficiency target, respectively. Total system operating cost The calculation formula is: ; In the formula, , These represent the unit price and power of electricity purchased from the grid during time period t. , These represent the unit price and power of electricity sold to the grid during time period t. Let be the unit operation and maintenance cost coefficient for the i-th device. The variable is a 0-1 variable (1 for running, 0 for stopping) representing the operating status of the i-th device during time period t. System energy efficiency With system efficiency The calculation formulas are as follows: ; ; In the formula, The total load power of the system during time period t (including the equivalent electrical power of electrical, heating, and cooling loads). The charging power of the energy storage device during time period t. The photovoltaic power generation during time period t The total output of the system during time period t. The total input of the system during time period t.

[0010] Furthermore, the constraints of the objective function include at least system power balance constraints, energy storage system operation constraints, and equipment operation constraints.

[0011] Furthermore, the intelligent fault diagnosis module determines anomalies by calculating the equipment health deviation index. The calculation formula is as follows: , in, Let t be the actual observed vector of the equipment operating parameters. The mean vector of GPR prediction. This is the vector of parameter standard deviations under historical normal operating conditions. When the sampling period exceeds the preset threshold λ for N consecutive sampling periods, the system determines it as a potential fault, triggers an early warning, and highlights the potentially faulty equipment in the 3D model of the BIM system.

[0012] Furthermore, the machine learning analysis module employs a Long Short-Term Memory (LSTM) network for temporal prediction, and its cell state updates follow the formula: ; in, , These are the forget gate and the input gate, respectively. In cellular state, The candidate cell state is represented by ⊙, which indicates the Hadamard volume.

[0013] Furthermore, the multi-energy management system also includes a user interaction layer, providing web and / or mobile interfaces to display to users the real-time status of BIM-based energy equipment, energy consumption information, optimization strategy execution status, and fault warning information.

[0014] The second objective of this invention is to provide a multi-energy management method, applied to the aforementioned multi-energy management system, comprising the following steps: Step S1: Real-time collection of operational data and environmental parameters of each energy subsystem through the edge computing layer, followed by local parsing and preprocessing; Step S2: Upload the preprocessed data to the cloud analysis layer. The data integration and analysis module dynamically weights and fuses the multi-source features from the energy subsystem module, environmental monitoring module, BIM system and historical data to construct a unified multi-dimensional feature expression. Step S3: The cloud analysis layer uses a machine learning analysis module to predict key power parameters of the system based on a unified multidimensional feature representation, and establishes a multi-objective optimization model based on the prediction results, thereby generating a collaborative optimization strategy. Step S4: The collaborative optimization strategy is compiled into a control instruction set and sent to the edge computing layer. The local decision execution module drives the corresponding devices to execute the instruction, forming a closed-loop control between cloud decision-making and edge execution. Step S5: The intelligent fault diagnosis module compares the real-time equipment operating parameters with the predicted values ​​of the health benchmark model trained based on historical data, calculates the equipment health deviation index, triggers an early warning when the index is continuously abnormal, and locates the potentially faulty equipment in the 3D model of the BIM system.

[0015] Furthermore, in step S2, an attention-based fusion method is used to dynamically weight and fuse multi-source features from the energy subsystem module, environmental monitoring module, BIM system, and historical data; for spatial topological relationships in BIM, a graph convolutional network is used for deep feature extraction. In step S3, the prediction of key system power parameters includes using a long short-term memory network to predict photovoltaic power output. System load Make predictions; use a multi-objective optimization model to calculate the total system operating cost. Minimize, energy efficiency Maximize and improve efficiency Maximizing is the overall objective; Total system operating cost The calculation formula is: ; In the formula, , These represent the unit price and power of electricity purchased from the grid during time period t. , These represent the unit price and power of electricity sold to the grid during time period t. Let be the unit operation and maintenance cost coefficient for the i-th device. The variable is a 0-1 variable (1 for running, 0 for stopping) representing the operating status of the i-th device during time period t. System energy efficiency With system efficiency The calculation formulas are as follows: ; ; In the formula, The total load power of the system during time period t (including the equivalent electrical power of electrical, heating, and cooling loads). The charging power of the energy storage device during time period t. The photovoltaic power generation during time period t The total output of the system during time period t. The total input of the system during time period t; In step S5, the health benchmark model is constructed using a Gaussian process regression model; the health deviation index... The calculation formula is: ; In the formula, Let t be the actual observed vector of the equipment operating parameters. The mean vector of GPR prediction. This is the standard deviation vector of parameters under historical normal operating conditions.

[0016] A third objective of this invention is to provide a computer device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the steps of the above-mentioned multi-energy management method.

[0017] A fourth objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned multi-energy management method.

[0018] The multi-energy management system, method, device, and computer storage medium based on edge-cloud collaboration and data coupling provided by this invention have the following advantages and beneficial effects: 1. This invention effectively solves the response latency problem of traditional centralized platforms by adopting a collaborative architecture of "real-time edge execution + global cloud decision-making." The edge side is responsible for rapid device-level control and command execution, ensuring the real-time performance and reliability of the system when faced with local disturbances; the cloud side focuses on the integration of massive amounts of data, training of machine learning models, and global optimization calculations, generating a better collaborative strategy. This design balances the speed of local control with the intelligence of global optimization.

[0019] 2. This invention deeply couples and analyzes real-time data from multiple energy sources, including electricity, photovoltaics, energy storage, HVAC, and charging piles. Combined with environmental parameters, BIM information, and historical patterns, it constructs a more comprehensive system analysis model. Based on this, machine learning methods are used for prediction, and a multi-objective optimization model integrating economic costs and energy efficiency indicators is employed for calculation. This enables the dynamic generation of refined collaborative strategies, such as "photovoltaic-storage-charging" linkage, thereby significantly improving the utilization rate of renewable energy and overall operational economy.

[0020] 3. By establishing an equipment health assessment model based on data correlation analysis and defining a quantifiable health deviation index, the system can identify early abnormal characteristics of equipment operation. This allows the system to issue early warnings before failures occur and quickly locate potentially problematic components using BIM models. This transforms traditional reactive maintenance or periodic inspection models into condition-based predictive maintenance, helping to reduce operating costs, minimize unexpected downtime, and improve system reliability.

[0021] 4. By deeply integrating Building Information Modeling (BIM), a 3D visualized digital platform encompassing spatial location and topological relationships is provided for energy management. Information such as equipment status, energy flow, and fault locations can be displayed and managed within a unified spatial context, significantly improving the intuitiveness and location efficiency of operation and maintenance. Simultaneously, BIM-based spatial information provides more accurate data support for load analysis, strategy simulation, and layout optimization. Attached Figure Description

[0022] Figure 1 A schematic diagram of the multi-energy management system of the present invention.

[0023] Figure 2 A schematic diagram of the specific architecture of the multi-energy management system in Example 1.

[0024] Figure 3 A schematic diagram of the central control platform in Example 1.

[0025] Figure 4 A schematic diagram of the user interaction layer interface in Example 1.

[0026] Figure 5 A schematic diagram of the multi-energy management method of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be further described below with reference to specific embodiments. The scope of protection of the present invention is not limited to the following embodiments; these examples are provided for illustrative purposes only and do not limit the present invention in any way.

[0028] Glossary

[0029] BIM: Building Information Modeling. It refers to the use of digital means to create three-dimensional models of the physical characteristics and functional properties of buildings, integrate information, and manage the entire process.

[0030] LSTM: Long Short-Term Memory. A special type of recurrent neural network (RNN) that effectively solves the problem of long-term dependencies by introducing a "gate" mechanism (forget gate, input gate, output gate).

[0031] GCN: Graph Convolutional Network. A deep learning model that can directly learn features from graph-structured data.

[0032] ABC: Artificial Bee Colony. A metaheuristic optimization algorithm that simulates the intelligent foraging behavior of bee colonies, comprising three roles: mercenary bees, observer bees, and scout bees, and finding the optimal solution through cooperative search.

[0033] GPR: Gaussian Process Regression. A nonparametric probabilistic model based on Bayesian theory and kernel methods, which can not only provide predicted values ​​but also estimates of the uncertainty of those predictions.

[0034] HDI: Health Deviation Index.

[0035] HVAC: Heating, Ventilation and Air Conditioning. Refers to the heating, ventilation, and air conditioning system within a building.

[0036] Unless otherwise specified, the terminology used in this invention has the meanings commonly understood by those skilled in the art.

[0037] The present invention will be further described in detail below through examples.

[0038] Example

[0039] Example 1

[0040] A schematic diagram of the multi-energy management system provided by this invention is shown below. Figure 1 As shown, the system comprises multiple energy subsystem modules, an environmental monitoring module, a BIM system, and a core central control platform, enabling intelligent interaction with users through a user interaction layer. The modules are interconnected and interact via a distributed soft bus. The central control platform, as the system's core, includes an edge computing layer and a cloud analytics layer. The edge computing layer includes a data access and parsing module and a local decision execution module; the cloud analytics layer includes a data integration and analysis module, a machine learning analysis module, an intelligent fault diagnosis module, and an optimized scheduling module.

[0041] The specific architecture diagram of the multi-energy management system provided in this embodiment is as follows: Figure 2 As shown, the system comprises multiple energy subsystems, an environmental monitoring module, a BIM system, and a core central control platform, enabling intelligent interaction with users through a user interface layer. All modules are interconnected and interact via a distributed soft bus.

[0042] In this embodiment, multiple energy subsystems include a photovoltaic power generation module, an energy storage module, a charging pile module, and a heat pump module. Each energy subsystem is equipped with an independent controller to achieve its own intelligent management.

[0043] The environmental monitoring module is used to collect indoor and outdoor environmental parameters such as temperature, humidity, and air velocity. In this embodiment, it includes a temperature sensor, a light sensor, and a humidity sensor.

[0044] BIM systems are used to construct virtual 3D models to simulate the operating status and heat transfer processes of various energy devices, which helps optimize the system's energy scheduling and equipment layout. In this embodiment, the BIM system provides data such as 3D building models, equipment spatial topology, and physical attribute information.

[0045] The central control platform, as the core of the system, includes an edge computing layer and a cloud analytics layer. A schematic diagram of the central control platform is shown below. Figure 3 As shown.

[0046] The edge computing layer is deployed on-site, acting as the nerve endings of the system. It is responsible for direct communication with system equipment such as photovoltaic inverters, energy storage converters, and charging piles to achieve data acquisition and command execution. The edge computing layer includes a data access and parsing module and a local decision execution module. The data access and parsing module is responsible for compatibility with multiple communication protocols and converting heterogeneous device data into a unified internal standard format. When the local decision execution module receives cloud scheduling instructions or detects anomalies such as emergency overload locally, it can immediately trigger preset control strategies to ensure the transient safety of the system.

[0047] ① The data integration and analysis module employs a combination of edge computing and cloud computing to achieve integrated analysis of multi-dimensional information such as energy consumption, environmental parameters, equipment status, building facility data, and historical trends. Specifically, the module integrates real-time data reported from the edge, BIM model information, and historical data. To achieve intelligent feature weighted fusion, this module uses a feature fusion function based on an attention mechanism to construct a unified multi-dimensional feature space F: .

[0048] Where E is the energy data vector, S is the environmental status vector, B is the feature vector derived from the BIM model, H is the historical trend feature vector, and Q is the query vector. , These are the key vector and value vector of each feature source, d k Let be the dimension of the vector.

[0049] For the spatial topological feature B derived from BIM, depth extraction is performed using a graph convolutional network (GCN), and the feature update formula for its (l+1)th layer is as follows: .

[0050] Where à is the device-space topological adjacency matrix with added self-loops. This is the corresponding degree matrix. Let σ be the trainable weight matrix of the l-th layer, and σ be the activation function.

[0051] ② The machine learning analysis module employs time-series forecasting algorithms such as Long Short-Term Memory (LSTM) networks to perform short-term forecasts of photovoltaic power output and system load. The core state update mechanism of the LSTM unit is as follows: .

[0052] in, , These are the forget gate and the input gate, respectively. In cellular state, The candidate cell state is represented by ⊙, which indicates the Hadamard volume.

[0053] ③ The scheduling module is optimized by establishing a multi-objective optimization function based on the system coupling model. The objective is to minimize the total system operating cost while maximizing the overall system energy efficiency. The objective function is formally expressed as: .

[0054] in, The total operating cost of the system, For system energy efficiency, Let γ represent the system efficiency, and let α and β be the weighting coefficients of the cost target, energy efficiency target, and efficiency target, respectively.

[0055] Total system operating cost The calculation formula is as follows: , , These represent the unit price and power of electricity purchased from the grid during time period t; , These represent the unit price and power of electricity sold to the grid during time period t; Let be the unit operation and maintenance cost coefficient for the i-th device; Let 0-1 be a variable representing the operating status of the i-th device during time period t (1 for running, 0 for stopping).

[0056] System energy efficiency System efficiency The calculation formula is as follows: , .

[0057] in, The total load power of the system during time period t (including the equivalent electrical power of electrical, heating, and cooling loads). The charging power of the energy storage device during time period t. The photovoltaic power generation during time period t The total output of the system during time period t. The total input of the system during time period t.

[0058] The main constraints include: Power balance constraints: The real-time power balance constraints that the system must satisfy are as follows: .

[0059] in, The photovoltaic power generation during time period t , These represent the energy storage discharge and charging power during time period t, respectively. Let t be the total system load power during time period t.

[0060] Energy storage system operation constraints: Energy storage devices must meet the following constraints during operation: , , , , .

[0061] in, The state of charge of the energy storage device during time period t; , These refer to the charging efficiency and discharging efficiency of energy storage devices, respectively. The duration of a single scheduling period; This refers to the rated capacity of the energy storage device. , These are the lower and upper limits allowed for the state of charge; , Maximum permissible charging and discharging power; , Let t be a 0-1 variable representing whether the energy storage is in a charging or discharging state during time period t.

[0062] Equipment operation constraints: The operating power of all controllable equipment must be within its capacity. .

[0063] Let be the output power of the i-th device during time period t; , Minimum and maximum technical output for the i-th device.

[0064] This optimized scheduling module employs an improved Artificial Bee Colony (ABC) algorithm for efficient solution. In the ABC algorithm, a new solution is generated during the hired bee phase. The search formula is: .

[0065] in, Let j be the j-th dimension of the current nectar source (solution). Let the j-th dimension be a randomly selected neighboring honey source. The value is a random number in the range [-1, 1]. The algorithm converges through multiple iterations, generating a scheduling instruction set containing the optimal power setpoints for each device.

[0066] ④ The intelligent fault diagnosis module, through correlation analysis of equipment operating data and environmental parameters, can predict and identify potential equipment failure risks and generate corresponding maintenance suggestions, thereby achieving data-driven predictive maintenance. This module constructs a health benchmark model based on Gaussian process regression for key equipment. The predicted distribution of the normal operating parameter vector of the equipment at time t is as follows: .

[0067] in, The input feature vector (environmental parameters, runtime setpoint).

[0068] The formula for calculating the real-time health deviation index is as follows: .

[0069] in, Let t be the actual observed vector of the equipment operating parameters. The mean vector of GPR prediction. This is the vector of parameter standard deviations under historical normal operating conditions. When the sampling period exceeds the preset threshold λ for N consecutive sampling periods, the system determines it as a potential fault, triggers an early warning, and highlights the suspected faulty component in the BIM 3D model.

[0070] The user interaction layer provides web and mobile interfaces, allowing users to visualize and view the real-time status of equipment, multi-dimensional energy consumption curves, optimization strategy execution, and fault warning information within a 3D BIM scene. For example... Figure 4 As shown, in the multi-energy management system provided in this embodiment, the web interface provided by the user interaction layer can display key system indicators such as system energy efficiency, real-time carbon emissions, and renewable energy ratio, as well as system data such as equipment status and energy consumption curves. It can also display information such as optimization strategy execution dashboards, early warning information, and pending work orders. For example, in... Figure 4 The user interface displays that the photovoltaic array and energy storage system are operating normally, the charging pile group displays warnings, and the heat pump unit shows a fault reminder for maintenance.

[0071] The system provided in this embodiment constructs a complete technical closed loop from data perception and intelligent analysis to optimized execution and proactive operation and maintenance through an edge-cloud collaborative physical architecture, a deep fusion model of multi-source data and BIM information, a multi-objective optimization algorithm that balances economy and energy efficiency, and a fault prediction mechanism based on health quantification assessment. This system effectively solves problems such as slow response in traditional centralized architectures, difficulties in coordinating multiple energy systems, and passive operation and maintenance modes, providing a practical solution for improving the operational efficiency, economy, and reliability of regional integrated energy systems.

[0072] Example 2

[0073] This embodiment provides a multi-energy management method based on edge-cloud collaboration and data coupling, as illustrated in the diagram below. Figure 5 As shown. This method is applied to a system as described in Example 1, and its core process embodies a closed loop of "edge perception - cloud decision-making - edge execution - proactive maintenance," specifically including the following steps: (1) Real-time acquisition and edge preprocessing of multidimensional data Through an edge computing layer deployed on-site, real-time operational data from energy subsystems such as photovoltaic inverters, energy storage converters, charging piles, and heat pump units, as well as environmental parameters from sensors such as temperature, light intensity, and humidity, are collected. The data access and parsing module of the edge computing layer is responsible for compatibility with multiple industrial protocols such as Modbus, CAN, and MQTT, uniformly parsing the collected heterogeneous data and converting it into the system's internal standard format, completing the initial data cleaning and time alignment.

[0074] (2) Edge-cloud collaborative data transmission and feature fusion

[0075] The edge computing layer uploads the pre-processed key data to the cloud analysis layer via a secure network channel (a VPN leased line in this embodiment). In the cloud, the data integration and analysis module performs deep information fusion. ① Spatiotemporal data alignment: Spatiotemporally correlate and align time-series energy data (E) and environmental data (S) with static spatial topology and equipment attribute data (B) provided by the BIM system; ② Intelligent Feature Fusion: An attention-based fusion method is adopted to dynamically assign weights to features from different sources (energy, environment, BIM, history) and construct a unified, high-dimensional feature representation F to highlight the information most relevant to the current optimization objective. For complex spatial relationships in BIM, a graph convolutional network (GCN) is used for deep feature extraction to capture the spatial coupling and energy transfer characteristics between devices.

[0076] ③ Historical trend correlation: Combining real-time features with historical operating trend features (H) provides context for the prediction model.

[0077] (3) Dynamic policy generation based on machine learning and multi-objective optimization

[0078] The cloud-based analytics layer performs intelligent computation and decision-making based on the fused multidimensional features: ① Critical State Prediction: The machine learning analysis module uses a trained Long Short-Term Memory (LSTM) network model, inputting historical and current feature sequences, to predict photovoltaic power output in the short term (e.g., the next 15 minutes to several hours). and total system load Perform rolling forecasts to provide forward-looking input for optimizing scheduling.

[0079] ② Multi-objective optimization solution: The optimization scheduling module establishes a multi-objective optimization model with economy and energy efficiency as the core. This model aims to minimize the total system operating cost. Maximizing overall system energy efficiency (while also considering energy efficiency) With efficiency The objective is a composite problem. The model strictly incorporates power balance constraints, dynamic constraints of the energy storage system, and upper and lower bound constraints for the operation of all controllable devices. This mixed-integer nonlinear programming problem is efficiently solved using an improved artificial bee colony algorithm (ABC). The algorithm, through the cooperative search of hired bees, observer bees, and scout bees, ultimately outputs a sequence of optimal power setpoints for each device (photovoltaic, energy storage, charging piles, heat pumps, etc.) within a scheduling cycle, forming a specific collaborative control strategy.

[0080] (4) Strategy issuance and edge closed-loop execution

[0081] The optimized scheduling strategy generated in the cloud is compiled into an executable set of control instructions and sent to the corresponding edge computing layer via a secure channel. Upon receiving the instructions, the local decision execution module of the edge computing layer then: ① Command parsing and security verification: Parse commands and verify their compatibility and security with the real-time status of local devices.

[0082] ② Rapid Closed-Loop Control: Power setpoint and other commands are converted into control signals recognizable by the equipment, driving the actions of actuators such as photovoltaic inverters, energy storage converters, and charging pile controllers to achieve coordinated regulation of "photovoltaics-storage-charging-heating." Simultaneously, the edge computing layer continuously monitors the local status. If command execution anomalies or sudden local faults occur (such as overload), preset rapid protection logic can be immediately triggered to ensure system transient safety, forming a collaborative closed loop of "global optimization in the cloud and real-time protection at the edge."

[0083] (5) Predictive maintenance based on data association model

[0084] In parallel with the main control flow, the intelligent fault diagnosis module continuously runs the predictive maintenance process: ① Health benchmark modeling: For key equipment (such as centrifugal chillers and photovoltaic inverters), based on their historical normal operation data and corresponding environmental parameters, machine learning models such as Gaussian process regression are trained to establish equipment health benchmark models. These models can provide the probability distribution of normal operation parameters of the equipment under given operating conditions.

[0085] ② Real-time health assessment: Real-time collection of equipment operating parameters ( ) and environmental parameters, input into the health baseline model, and obtain the predicted mean value of the current parameters within the normal range ( Calculate the health deviation index: , in, This is the standard deviation vector of parameters under historical normal operating conditions.

[0086] ③ Fault early warning and location: When a certain device When the index continuously exceeds the preset threshold λ, the system determines that there is a potential fault risk and immediately generates an early warning. The early warning information not only includes the fault type and probability, but also highlights the suspected faulty component in the 3D visualization interface by associating it with the BIM model (such as "abnormal current in the 5th string of the roof photovoltaic array of Building 3"), and automatically generates a maintenance work order containing the location information and pushes it to the operation and maintenance personnel.

[0087] (6) Visual interaction and continuous optimization

[0088] Users access the system's 3D visualization interface through the web or mobile terminal of the user interaction layer. The interface is based on the BIM model and displays it dynamically. Overall situation: Key indicators such as total system energy efficiency, real-time carbon emissions, and the proportion of renewable energy.

[0089] Equipment status: Real-time operating parameters, energy consumption curves, and health status (normal / warning / fault) of each device.

[0090] Strategy Execution: Optimize the execution dashboard of scheduling strategies, showing a comparison between the plan and the actual operation.

[0091] Warning Work Order: Receive and process warnings and work orders issued by the predictive maintenance module.

[0092] In addition, new data generated during system operation will be continuously fed back to the cloud for continuous updating and training of the LSTM prediction model, health benchmark model and optimization algorithm parameters, so as to realize the continuous improvement of the system's self-evolution and optimization capabilities.

[0093] In summary, the method provided in this embodiment achieves deep coupling of multi-energy data, collaborative decision-making between edge and cloud intelligence, and proactive prevention of operation and maintenance through iterative steps, effectively improving the economy, reliability, and intelligence level of the energy system.

[0094] Example 3

[0095] This embodiment provides a computer device for implementing the multi-energy management method provided by the present invention. The device includes at least one processor and a memory coupled thereto; the memory stores a computer program, and the processor is capable of reading and executing the computer program to implement the various steps of the multi-energy management method provided by the present invention.

[0096] Specifically, the memory may include volatile memory (such as random access memory) and non-volatile memory (such as read-only memory, flash memory, or hard disk drive), used to store program code and required data. The processor and memory are interconnected and communicate via an internal bus; this bus may be an industry-standard architecture bus, a peripheral component interconnect standard bus, an extended industry-standard architecture bus, or an evolution thereof, responsible for transmitting address signals, data signals, and control signals between the processor, memory, and other possible peripherals. The computer program contains machine instructions that can be executed directly or indirectly by the processor; when the processor executes these instructions, the complete process of the multi-energy management method provided by this invention is implemented.

[0097] Example 4

[0098] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is loaded and executed by a processor, the processor implements the various steps of the multi-energy management method provided by this invention.

[0099] The computer-readable storage medium includes, but is not limited to, various volatile or non-volatile storage devices. For example, volatile memory may include static random access memory (SRAM), dynamic random access memory (DRAM), etc.; non-volatile memory may include mask read-only memory (MROM), programmable read-only memory (PROM), flash memory, magnetic storage (such as hard disks and magnetic tapes), optical storage (such as CD-ROMs, DVD-ROMs, and Blu-ray discs), and media based on emerging storage technologies (such as phase-change memory and resistive switching memory). This storage medium can be built into the device or be removable. The computer program stored on it is recorded in the form of instructions on the physical medium, which can be read and executed by a computing device to realize the various steps of the multi-energy management method provided by this invention.

[0100] Those skilled in the art should note that the embodiments described in this invention are merely exemplary, and various other substitutions, changes, and improvements can be made within the scope of this invention. Therefore, this invention is not limited to the above embodiments, but is defined only by the claims.

Claims

1. A multi-energy management system, characterized in that, include: Multiple energy subsystem modules are used to provide or consume energy and collect their own operational data; The environmental monitoring module is used to collect environmental parameters; BIM systems are used to provide virtual 3D models that include equipment space and attribute information; The central control platform includes an edge computing layer and a cloud analytics layer; And a distributed soft bus, used to enable data interconnection between modules; The edge computing layer is deployed on-site and includes a data access and parsing module and a local decision execution module, which are used to collect and parse data in real time and execute rapid control of local devices. The cloud-based analytics layer is deployed on a remote server and includes a data integration and analysis module, a machine learning analysis module, an optimization and scheduling module, and an intelligent fault diagnosis module. The data integration and analysis module is used to aggregate and store data from the edge computing layer and fuse it with BIM static information to construct a unified multi-dimensional feature representation; the machine learning analysis module is used to model and train based on the multi-dimensional feature representation to predict key power parameters of the system; the optimization scheduling module, based on the prediction results of the machine learning analysis module and the coupled model, calculates the globally optimal energy scheduling scheme with energy efficiency and economy as comprehensive objectives, and generates a collaborative optimization strategy; the intelligent fault diagnosis module is used to identify potentially faulty equipment.

2. The system as described in claim 1, characterized in that, The data integration and analysis module employs an attention-based fusion method to dynamically weight and fuse multi-source features from the energy subsystem module, the environmental monitoring module, the BIM system, and historical data. It also uses a graph convolutional network to extract deep features from the spatial topology of equipment in the BIM system to construct the unified multidimensional feature representation. The machine learning analysis module performs modeling and training based on the multidimensional feature representation, and uses a long short-term memory network to predict key system power parameters such as photovoltaic output and system load. The optimized scheduling module, based on the prediction results of the machine learning analysis module and the coupling model, calculates the total system operating cost. Minimize, energy efficiency Maximize and improve efficiency With maximizing the overall objective, the globally optimal energy scheduling scheme is calculated, and the collaborative optimization strategy is generated. The intelligent fault diagnosis module constructs a device health benchmark model based on Gaussian process regression and calculates a health deviation index to identify potentially faulty devices.

3. The system as described in claim 2, characterized in that, The optimization scheduling module uses a multi-objective optimization model to calculate the globally optimal energy scheduling scheme; The multi-objective optimization model uses the photovoltaic output predicted by the machine learning analysis module. and system load As input parameters, and constructed based on the coupled model that includes power balance constraints, energy storage system operation constraints and equipment operation constraints; The objective function of the multi-objective optimization model is: ； in, The total operating cost of the system, For system energy efficiency, For system efficiency, γ, α, and β are the weighting coefficients of cost target, energy efficiency target, and efficiency target, respectively. The total operating cost of the system The calculation formula is: ； in, , These represent the unit price and power of electricity purchased from the grid during time period t. , These represent the unit price and power of electricity sold to the grid during time period t. Let be the unit operation and maintenance cost coefficient for the i-th device. A 0-1 variable representing the operating status of the i-th device during time period t; The system energy efficiency With system efficiency The calculation formulas are as follows: ； ； in, Let t be the total load power of the system during time period t. The charging power of the energy storage device during time period t. The photovoltaic power generation during time period t The total output of the system during time period t. The total input of the system during time period t.

4. The system as described in claim 1, characterized in that, The intelligent fault diagnosis module determines abnormalities by calculating the equipment health deviation index. The calculation formula is as follows: ； in, Let t be the actual observed vector of the equipment operating parameters. The mean vector of GPR prediction. This is the vector of standard deviations of parameters under historical normal operating conditions; when When the sampling period exceeds the preset threshold λ for N consecutive sampling periods, the system determines it as a potential fault, triggers an early warning, and highlights the potentially faulty device in the 3D model of the BIM system.

5. The system as described in claim 1, characterized in that, The machine learning analysis module uses a Long Short-Term Memory (LSTM) network for temporal prediction, and its cell state update follows the formula: ； in, , These are the forget gate and the input gate, respectively. In cellular state, The candidate cell state is represented by ⊙, which indicates the Hadamard volume.

6. The system according to any one of claims 1-5, characterized in that, It also includes a user interaction layer, providing web and / or mobile interfaces to display to users the real-time status of BIM-based energy equipment, energy consumption information, optimization strategy execution status, and fault warning information.

7. A multi-energy management method, applied to the system according to any one of claims 1-6, characterized in that, Includes the following steps: Step S1: The edge computing layer collects the operating data and environmental parameters of each energy subsystem in real time and performs local parsing and preprocessing. Step S2: Upload the preprocessed data to the cloud analysis layer. The data integration and analysis module dynamically weights and fuses the multi-source features from the energy subsystem module, the environmental monitoring module, the BIM system, and historical data to construct the unified multi-dimensional feature expression. Step S3: Based on the unified multidimensional feature expression, the cloud analysis layer uses the machine learning analysis module to predict the key power parameters of the system, and establishes a multi-objective optimization model based on the prediction results, thereby generating the collaborative optimization strategy. Step S4: The collaborative optimization strategy is compiled into a control instruction set and sent to the edge computing layer, whereby the local decision execution module drives the corresponding device to execute it, forming a closed-loop control between cloud decision-making and edge execution; Step S5: The intelligent fault diagnosis module compares the real-time equipment operating parameters with the predicted values ​​of the health benchmark model trained based on historical data to calculate the equipment health deviation index. When the index is continuously abnormal, an early warning is triggered, and the potentially faulty equipment is located in the three-dimensional model of the BIM system.

8. The method as described in claim 7, characterized in that, In step S2, an attention-based fusion method is used to dynamically weight and fuse multi-source features from the energy subsystem module, the environmental monitoring module, the BIM system, and historical data; for spatial topological relationships in BIM, a graph convolutional network is used for deep feature extraction. In step S3, the prediction of key system power parameters includes using a long short-term memory network to predict photovoltaic output. System load Make predictions; the multi-objective optimization model takes the total system operating cost as the basis. Minimize, energy efficiency Maximize and improve efficiency Maximizing is the overall objective; The total operating cost of the system The calculation formula is: ； in, , These represent the unit price and power of electricity purchased from the grid during time period t. , These represent the unit price and power of electricity sold to the grid during time period t. Let be the unit operation and maintenance cost coefficient for the i-th device. A 0-1 variable representing the operating status of the i-th device during time period t; The system energy efficiency With system efficiency The calculation formulas are as follows: ； ； in, Let t be the total load power of the system during time period t. The charging power of the energy storage device during time period t. The photovoltaic power generation during time period t The total output of the system during time period t. The total input of the system during time period t; In step S5, the health benchmark model is constructed using a Gaussian process regression model; the health deviation index The calculation formula is: ； in, Let t be the actual observed vector of the equipment operating parameters. The mean vector of GPR prediction. This is the standard deviation vector of parameters under historical normal operating conditions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program that can run on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 7-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 7-8.

Images