Multi-garden energy consumption prediction and scheduling method and control system based on digital twinning

Abstract

The application provides a multi-garden energy consumption prediction and scheduling method and control system based on digital twinning. By constructing a technical chain from data perception to closed-loop optimization, the pain points of multi-garden energy consumption management are systematically solved. First, by constructing a global unified digital twinning model, the standardization integration of scattered and heterogeneous garden assets and data is realized, and the information island problem is overcome. Second, federated learning is used for prediction. Under the premise of ensuring the privacy and security of the data of each garden, cross-garden knowledge sharing and joint modeling are realized, and the prediction accuracy of a single garden under limited data conditions is significantly improved. Finally, through the closed-loop process of "prediction-decision-execution-update", traditional passive and static energy consumption management is transformed into active and dynamic prediction and scheduling, which can proactively suppress energy consumption peaks and optimize energy distribution, effectively reducing the overall energy consumption cost and carbon emissions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-garden management, in particular to a multi-garden energy consumption prediction and scheduling method and control system based on digital twinning. BACKGROUND

[0002] Traditional garden energy consumption management mostly relies on deploying various sensors and intelligent meters to build isolated energy consumption monitoring platforms, and realizes data visualization and report statistics. However, such systems can only provide historical and real-time data presentation, which belongs to the "posteriori" passive management, cannot make prospective prediction on future energy consumption trend, lacks the ability of active intervention and regulation based on prediction results, and is difficult to realize the whole-cycle closed-loop optimization of "prior planning, in-process control and post-analysis" of energy use.

[0003] To improve energy efficiency, some advanced schemes have appeared in the prior art. Firstly, some systems try to introduce a single prediction algorithm, such as time series analysis based on historical data, to make rough prediction on overall power consumption of the garden. However, such method generally has the problem of insufficient accuracy, especially cannot cope with strong volatility caused by renewable energy access such as photovoltaic, and has rough prediction granularity, which is difficult to support fine management at the device level. Secondly, some single-garden systems begin to try simple automatic control, such as starting and stopping lighting and adjusting air conditioning temperature according to preset schedule. However, these control strategies are static and rigid, cannot be linked with real-time changing energy supply and demand (such as photovoltaic power generation), dynamic energy demand and cross-system coordination, and have limited optimization effect. Moreover, most existing schemes are designed for a single garden, and when the management object is expanded to multiple geographically dispersed gardens, the inherent defects are dramatically magnified: the data of each garden is isolated, and due to data privacy and security considerations, it is difficult to centrally process and jointly analyze the original data, resulting in inability to tap energy saving potential from the cluster level and realize cross-garden energy coordination and mutual aid.

[0004] In recent years, digital twinning and federated learning technologies have provided new ideas for solving the above problems. Digital twinning technology is expected to create a high-fidelity virtual mapping for physical gardens, but its application in current garden management mostly stays at the level of geometric model visualization and device state monitoring, and fails to deeply integrate with business flow and energy consumption prediction and optimization control logic, resulting in "good-looking but not useful" models. Although federated learning technology can realize joint modeling without domain data, existing researches mostly focus on theory or single function verification, and there is no mature case of successfully integrating it with digital twinning technology and applying it to the complex scenario of multi-garden energy consumption prediction and scheduling. There is still a lack of effective linkage between systems (such as energy consumption management, operation and maintenance work order, and micro-grid monitoring), forming "data islands" and "functional chimneys", for example, it is impossible to automatically convert abnormal results of energy consumption prediction into operation and maintenance actions, or dynamically adjust energy consumption strategies according to real-time state of micro-grid.

[0005] Therefore, there is an urgent need in the art for an innovative system that can break through the above barriers and achieve multi-park energy efficiency management. SUMMARY

[0006] To solve the above technical problems, the present application provides a multi-park energy consumption prediction and scheduling method based on digital twinning and a control system.

[0007] The first aspect of the embodiments of the present application provides a multi-park energy consumption prediction and scheduling method based on digital twinning, comprising the following steps:

[0008] S100: data acquisition step, collecting real-time operation data, energy consumption data and environmental parameters of each energy-using system and micro-grid in multiple distributed parks in parallel;

[0009] S200: digital twin construction step, based on the collected data, constructing a digital twin containing a four-level topological relationship of project-building-space-equipment for each park, and forming a global unified model;

[0010] S300: federated learning prediction step, taking the global unified model as a benchmark, driving the federated learning process, and generating energy consumption prediction data of each park in the future period by fusing the weight parameters of the local prediction model of each park;

[0011] S400: elastic decision step, comparing the energy consumption prediction data of each park with the elastic energy consumption budget, if the difference exceeds the preset value, generating a set of energy-saving control instructions containing at least one device of energy storage, heating, lighting and charging pile;

[0012] S500: instruction execution step, analyzing the device topological relationship of each park, issuing and executing the set of energy-saving control instructions, and collecting feedback data;

[0013] S600: closed-loop update step, updating the digital twin state and local prediction model weight parameters of each park respectively using the feedback data.

[0014] By constructing a technology chain from data perception to closed-loop optimization, the pain points of multi-park energy consumption management are systematically solved; first, by constructing a global unified digital twin model, the standardized integration of scattered and heterogeneous park assets and data is realized, and the problem of information island is overcome; second, federated learning is used for prediction, which realizes cross-park knowledge sharing and joint modeling under the premise of ensuring the privacy and security of each park data, and significantly improves the prediction accuracy of a single park in the case of limited data; finally, through the closed-loop process of "prediction-decision-execution-update", the traditional passive and static energy consumption management is transformed into active and dynamic prediction scheduling, which can proactively suppress energy consumption peak and optimize energy distribution, effectively reducing the comprehensive energy consumption cost and carbon emissions.

[0015] In one or more embodiments of the present application, the construction of the digital twin in step S200 includes the following specific processes: based on device access information and spatial layout, a project-building-space-device hierarchical path to which the device belongs is automatically generated to form a membership topology; at the same time, based on the membership data of smart meters, circuit breakers and distribution boxes, an electrical connection topology between devices is automatically generated; the digital twin is the fusion of the membership topology and the electrical connection topology.

[0016] In the above embodiment, by automatically generating and fusing the membership topology and the electrical connection topology, the digital twin not only reflects the management affiliation of the device, but also accurately depicts the actual energy flow path. This provides an irreplaceable data foundation for subsequent precise scheduling (such as preventing local overload), and avoids control command misallocation or failure caused by inaccurate topology relationship.

[0017] In one or more embodiments of the present application, when constructing the digital twin of each park in step S200, a digital energy consumption benchmark fingerprint is dynamically created and bound for each device, which is composed of device static attributes, dynamic operating parameters and historical statistical characteristics; wherein the static attributes at least include device type, rated power, factory date and energy efficiency level, the dynamic operating parameters are power consumption data collected by the device under normal working condition for at least 24 hours, and the historical statistical characteristics are steady-state power consumption Gaussian distribution parameters calculated from the power consumption data, including mean μ and variance σ².

[0018] In the above embodiment, the "energy consumption benchmark fingerprint" created establishes a dynamic energy consumption health record for each device. It not only provides a high-precision interpolation basis for data missing, ensuring the quality of model input, but also serves as a quantitative benchmark for device energy efficiency degradation and early fault diagnosis, realizing the functional extension from simple energy consumption monitoring to device health management.

[0019] In one or more embodiments of the present application, the federated learning process in step S300 adopts a hybrid architecture of long and short time window cooperation; wherein the long time window model takes long-term data of 7 to 30 days in the past as the training set; the short time window model takes recent data of 1 to 24 hours in the past as the training set; in the cloud server, an adaptive weight fusion mechanism is designed, and its formula is represented as:

[0020] W = a W_long + (1-a) W_short;

[0021] Wherein, W represents the weight parameter of the fused prediction model;

[0022] W_long represents the weight of the long time window model, which contains the seasonal and periodic trend characteristics of the long term (7 to 30 days);

[0023] W_short represents the weight of the short time window model, which contains the instantaneous fluctuation and mutation characteristics of the short term (1 to 24 hours);

[0024] a is an adaptive weight factor, which is a dynamically changing coefficient, and its value range is [0, 1], which is determined by the variance of photovoltaic power fluctuation. If the variance is large, a is adjusted to be small, so that the model trusts the short time window model more; if the variance is small, a is adjusted to be large, so that the model trusts the long time window model more.

[0025] In the above embodiment, through the design of long and short time window models and adaptive weight factor, the prediction model can intelligently balance long-term regularity and short-term fluctuation. Especially, the weight factor is dynamically associated with the photovoltaic fluctuation variance, which greatly improves the response speed and prediction accuracy of sudden weather changes in the scene of high proportion of renewable energy access, and reduces the scheduling error caused by prediction lag.

[0026] In one or more embodiments of the present application, the elastic energy consumption budget in step S400 is a mechanism with upper and lower boundaries and width that can be dynamically adjusted, and its decision logic is as follows: the system monitors the running state of each park microgrid in real time, continuously calculates the surplus power of the current photovoltaic power generation and the adjustable capacity of the energy storage system; if the park energy consumption prediction data exceeds the upper limit of the budget, but the system determines that there is surplus photovoltaic power at present, the strip widening strategy is automatically triggered, the elastic energy consumption budget is temporarily relaxed to be the upper limit constraint value of the park budget, and the instruction to guide the consumption of local surplus photovoltaic power is generated preferentially; if the prediction data exceeds the upper limit of the budget and the park microgrid is in the state of insufficient photovoltaic output or relying on grid power supply, the tightening strategy is automatically triggered, the elastic energy consumption budget is temporarily tightened to be less than or equal to 70% of the upper limit constraint value of the park budget, and the instruction to reduce the running intensity of energy-consuming equipment in the park is generated.

[0027] In the above embodiment, the energy consumption budget is converted into a "flexible strip" that is linked to renewable energy in real time. This mechanism can dynamically adjust the intensity of energy-saving strategies. When there is excess photovoltaic power, green electricity is consumed first. When there is insufficient photovoltaic power, the load is suppressed decisively. In this way, while ensuring energy demand, the local consumption rate of renewable energy is maximized, achieving the unity of economy and greenness.

[0028] In one or more embodiments of the present application, in step S500, before the energy-saving control instruction set is issued to a certain park for execution, a centralized instruction conflict arbitration and verification step is introduced: this step traces all receiving devices of the instructions based on the digital twin of the park. When it is detected that the same device is assigned two or more instructions that have logical contradictions or operation exclusions within the same time window, the arbitration rule is triggered; the arbitration rule is based on a set of pre-set energy consumption sensitivity ranking knowledge base, which quantitatively ranks the sensitivity of all types of controlled devices and their control instructions to the global energy consumption. The arbitration engine preferentially selects the instruction that contributes most to the global energy-saving effect and has the highest sensitivity, while the conflicting secondary instructions are temporarily stored in the cache queue and are delayed to the next control period or are evaluated and issued after the premise changes.

[0029] In the above embodiment, by introducing a centralized arbitration mechanism based on energy consumption sensitivity, the problem of instruction conflict that may occur when multiple automation systems are controlled in parallel is effectively solved, which ensures the coordination and consistency of the control logic, avoids frequent action of the actuator or strategy failure, and greatly improves the reliability and stability of complex automation systems.

[0030] In one or more embodiments of the present application, in step S500, the energy-saving control instruction set supports cross-system scenario linkage based on the digital twin topology, which specifically embodies as follows: when the system predicts that the total load of charging piles in a certain park will exceed the safety threshold of the capacity of its upper transformer, the control logic does not simply cut off the charging power, but first accurately locates all intelligent lighting loops that are in the same transformer power supply loop as the charging piles by querying the digital twin; then, automatically generates instructions to uniformly lower the dimming values of these lighting loops to a pre-set level without affecting the safety lighting, thereby reducing the total load of the transformer; after the charging pile load peak, the lighting dimming value is automatically restored to 100%.

[0031] In the above embodiment, by using the accurate electrical topology of the digital twin, the intelligent lighting system is innovatively used as a flexible adjustment resource to cope with impact loads such as charging piles. This method realizes the "soft expansion" of the local power grid at a very low cost (compared to transformer expansion), ensures power supply safety, and significantly saves fixed asset investment.

[0032] In one or more embodiments of the present application, the updating of the digital twin state and the local prediction model weight parameters of each park in step S600 using the feedback data comprises: adopting a gradient truncation incremental learning strategy: when the system receives the device execution feedback data, if it is found that the error between it and the expected state of the model is greater than the preset value, the updating algorithm is updated; wherein the updating algorithm is not a global retraining of the entire federated learning model, but only a local fine-tuning of the fully connected weights of the last one or two layers of the model, and a very low learning rate is used for iteration; at the same time, the feature extraction weights of the intermediate layers of the model are completely frozen to prevent the destruction or covering of the learned knowledge with good generalization.

[0033] In the above embodiment, the gradient truncation incremental learning strategy is used for model updating, so that the system can adapt to changes in new data while effectively protecting the learned core knowledge from being destroyed (avoiding forgetting). This robust updating mechanism ensures the continuity and stability of the algorithm during long-term operation, reducing maintenance costs.

[0034] In a second aspect of the embodiments of the present application, a digital twin-based multi-park control system for implementing any of the above-described methods is provided, comprising: a cloud server, wherein the cloud server is configured with:

[0035] a digital twin module, configured to construct and maintain a four-level topological digital twin model containing membership and electrical connection relationships based on the project, building, space and device information of each park;

[0036] an energy consumption prediction module connected to the digital twin module, configured to access the energy consumption, environment and micro-grid operation data of each park, and generate energy consumption prediction data for future periods based on a federated learning algorithm;

[0037] a prediction and scheduling decision module connected to the energy consumption prediction module and the digital twin module, configured to compare the energy consumption prediction results with a preset energy consumption budget, and generate an energy-saving control instruction set for heating, lighting, energy storage and charging pile systems;

[0038] an operation linkage module connected to the energy consumption prediction module, configured to automatically generate and dispatch an inspection work order to an operation service module when an abnormal device energy consumption is predicted;

[0039] a plurality of edge collection control nodes respectively deployed in each park, configured to collect local data and upload it to the cloud server, and receive and execute the energy-saving control instruction set issued;

[0040] a micro-grid monitoring module integrated in the cloud server or the plurality of edge collection control nodes, configured to monitor the operation states of photovoltaic, energy storage, charging piles and power transmission and distribution systems in real time, and provide state data to the energy consumption prediction module and the prediction scheduling decision module;

[0041] The digital twin module, the energy consumption prediction module, the prediction scheduling decision module, the operation and maintenance linkage module, and the micro-grid monitoring module are cooperated with the plurality of edge collection control nodes to form a control system of data collection, prediction analysis, optimization decision and closed-loop execution across parks.

[0042] In the above embodiments, the system claim protects the hardware architecture and software module entity realizing the above method. Through the distributed architecture design of cloud cooperation, the powerful computing and cooperation ability of the cloud is ensured, and the real-time response and data privacy protection ability of the edge side are also given, which provides efficient, reliable and scalable physical support for the landing of the method.

[0043] In one or more embodiments of the present application, the operation and maintenance linkage module is configured to continuously monitor the device-level energy consumption anomaly prediction results and alarm events output by the prediction scheduling decision module; through an internal API interface, automatically call the operation and maintenance service work order creation function, convert the abnormal events into executable prediction inspection work orders, and automatically assign inspection priorities for inspection.

[0044] In the above embodiments, by realizing the automatic linkage of energy consumption prediction and operation and maintenance work order at the system architecture level, the data flow of "monitoring-prediction-operation and maintenance" is opened up, and a new mode of prediction and maintenance is created. This can convert energy efficiency anomalies into timely maintenance actions, eliminate device sub-health status from the source, and improve operation and maintenance efficiency and energy use efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A flowchart of a multi-park energy consumption prediction and scheduling method based on digital twinning provided by an embodiment of the present application;

[0046] Figure 2 A structural diagram of a multi-park control system based on digital twinning provided by an embodiment of the present application;

[0047] Figure 3 A schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0048] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods and apparatus are omitted so as not to obscure the description of the present application with unnecessary detail.

[0049] For the purpose of making the objectives, technical solutions and advantages of the present application clearer, the following will combine the accompanying drawings to make a detailed description. Figures 1-3 The present application is described by specific embodiments.

[0050] Reference is made to Figure 1 , Figure 1 A flowchart of a multi-park energy consumption prediction and scheduling method based on digital twinning provided by an embodiment of the present application includes the following steps:

[0051] S100: a data acquisition step, acquiring real-time operation data, energy consumption data and environmental parameters of each energy-using system and micro-grid in multiple distributed parks in parallel;

[0052] S200: a digital twinning construction step, constructing a digital twinning body containing a four-level topological relationship of project-building-space-equipment for each park based on the collected data, and summarizing to form a global unified model;

[0053] S300: a federated learning prediction step, taking the global unified model as a benchmark, driving a federated learning process, generating energy consumption prediction data of each park in a future period by fusing the weight parameters of the local prediction model of each park;

[0054] S400: an elastic decision step, comparing the energy consumption prediction data of each park with the elastic energy consumption budget, if the difference exceeds a preset value, generating a set of energy-saving control instructions containing at least one device of energy storage, heating, lighting and charging pile;

[0055] S500: an instruction execution step, analyzing the device topological relationship of each park, issuing and executing the set of energy-saving control instructions, and collecting feedback data;

[0056] S600: a closed-loop update step, updating the digital twinning body state and local prediction model weight parameters of each park respectively using the feedback data.

[0057] It can be understood that digital twin refers to a high-fidelity virtual model of a project, building, space and equipment created through digital means. The model not only reflects the geometric appearance, but more importantly, it maps its state (such as switches, energy consumption), behavior (such as operating mode) and rules (such as topological relationship) in real time, and can be simulated, analyzed and controlled. Multi-park refers to multiple park entities that are geographically dispersed and may be independently managed administratively. Federated learning is a distributed machine learning technique whose core idea is that data is retained locally (in each park) without exchanging raw data, only training the model locally, then uploading the model parameters (such as weights, gradients) to the cloud server for aggregation to generate a more powerful global model. Elastic energy budget is a budget indicator that is different from a fixed and rigid budget. It is a budget range or "strip" that can be dynamically adjusted up and down according to real-time conditions (especially renewable energy). Its essence is a dynamic resource allocation and decision boundary. Closed-loop update refers to the system's ability to re-input the results (feedback data) of the execution of control instructions into the system for effect evaluation and automatic adjustment of model parameters or control strategies, forming a "perception-decision-execution-learning" self-optimization cycle.

[0058] Principle of the present embodiment

[0059] (1) Data acquisition step

[0060] Sensing layer hardware: Deploy various sensors (smart meters, temperature / humidity / light sensors, flow meters), IoT gateways, smart device controllers (such as DDC controllers), and microgrid monitoring equipment (photovoltaic inverter monitoring, energy storage BMS communication interface, charging pile controller) in each park.

[0061] Communication protocol: Use industry standard protocols such as Modbus, BACnet, OPC UA, MQTT, etc. to aggregate data through IoT gateways to local edge computing nodes in the park or directly upload to the cloud server.

[0062] Specific data collected includes: energy consumption data: instantaneous and cumulative values of electricity, water, gas, etc. for each circuit; operating data: device start-stop status, operating frequency, power factor, set value (such as temperature set point); environmental parameters: indoor and outdoor temperature and humidity, illuminance, carbon dioxide concentration; microgrid status: real-time power of photovoltaic power generation, energy storage system SOC (state of charge), current load of charging pile, grid feeder power; key points: achieve multi-park "parallel" acquisition, which means a unified device access specification and communication framework are needed to ensure the timeliness and consistency of the data.

[0063] (2) Digital twin construction step

[0064] Modeling platform: Use a professional digital twin platform as the technical foundation.

[0065] Model definition: First, a standardized data model is established to define the entity types of "project", "building", "space", "device", as well as their respective attributes (such as name, ID, area), relationships and telemetry data points.

[0066] Instantiation: According to the collected asset information, create digital twin instances of each entity on the platform, such as "Park A Energy Saving Project", "Building 1", "Conference Room 101", "Air Conditioner-1", etc., and establish connections according to the actual affiliation.

[0067] Data binding: Bind the real-time data stream collected by S100 to the corresponding digital twin instance attributes through unique identifiers (such as device ID), so that the virtual model can reflect the state of the physical entity in real time.

[0068] Cloud server aggregation: The digital twins of each park are registered and associated on the cloud server, forming a unified digital twin model network that covers all parks and can be globally queried and analyzed.

[0069] (3) Federated learning prediction steps

[0070] Step 1: Cloud initialization: The cloud server initializes a global prediction model and distributes it to all edge nodes of participating parks;

[0071] Step 2: Local training: Each park's edge node uses local historical energy consumption, environmental data, etc. to train the received global model for several rounds to obtain local model updates;

[0072] Step 3: Parameter upload: Each edge node uploads the trained model weight parameters (not training data) to the cloud after encryption;

[0073] Step 4: Cloud aggregation: The cloud server collects all park model weights and uses methods such as weighted averaging to generate a new, improved global model;

[0074] Step 5: Model distribution: Distribute the new global model to each park;

[0075] Loop iteration: Repeat steps 2-5 to continuously optimize the global model.

[0076] Prediction generation: Finally, each park uses the optimized global model or combines the locally fine-tuned model to input the latest real-time data to generate energy consumption prediction curves for the park, building, and key equipment for hours, days, months, or even years.

[0077] (4) Elastic decision-making steps

[0078] Budget band setting: The management end sets a basic energy consumption budget target value (such as monthly total electricity bill or total electricity quantity), and allows to define a dynamic floating range.

[0079] Decision engine: Develop a rule engine or optimization algorithm module. The input of this module is the prediction curve generated in step S300 and the real-time microgrid state (photovoltaic output, energy storage SOC).

[0080] Decision logic: IF (predicted energy consumption > budget upper limit) AND (photovoltaic surplus power > threshold value) THEN: Trigger the "widening" strategy, generate instructions such as "increase energy storage charging power", "guide electric vehicle charging" and other green electricity consumption instructions.

[0081] IF (predicted energy consumption > budget upper limit) AND (photovoltaic output is insufficient) THEN: Trigger the "tightening" strategy, generate instructions such as "moderately increase air conditioning temperature set point", "reduce non-critical area lighting brightness" and other load reduction instructions.

[0082] Instruction set generation: The output of the decision engine is a structured instruction list that clearly indicates instruction content, target device, execution time window, etc.

[0083] (5) Execution and closed-loop update

[0084] The instruction set is reliably delivered to the edge controller of the specified park through the message queue or industrial protocol. After the controller executes, it will feedback the confirmation reply and new sensor readings back to the cloud.

[0085] Digital twin state update: Update the values of the corresponding attributes in the digital twin with the feedback real data (such as actual energy consumption).

[0086] Model incremental update: Use the new data (predicted value + actual value) as training samples to fine-tune the local parameters of the federated learning model in an online learning or small batch learning manner, and correct the prediction bias.

[0087] In the embodiments of the present application, the above-mentioned method realizes the transformation of multi-park energy consumption management from scattered and isolated, passive response, experience-driven to unified and coordinated, active intervention, and data intelligent driving. It effectively solves the contradiction between data privacy and collaborative optimization, greatly improves the prediction accuracy, realizes the fine and forward-looking scheduling of energy, and can continuously evolve through closed-loop learning, and finally achieves the goals of significant energy saving, cost reduction, and management efficiency improvement.

[0088] In one or more embodiments of the present application, the constructing of the digital twin in step S200 includes the following specific process: based on the device access information and the space layout diagram, a project-building-space-device hierarchical path to which the device belongs is automatically generated to form a membership topology; at the same time, based on the membership relationship data of the smart meter, the circuit breaker and the distribution box, an electrical connection topology between devices is automatically generated; the digital twin is a fusion of the membership topology and the electrical connection topology.

[0089] The membership topology refers to the hierarchical relationship from the perspective of organization management, and embodies the logic of “who manages who” and “who belongs to where”. For example, a device belongs to a certain space, which is located in a certain building, and the building belongs to a certain project. This is a tree structure. The electrical connection topology describes the connection relationship of the physical flow path of energy, and embodies the physical fact of “who supplies power for whom” and “how the current flows”. For example, a distribution box supplies power to multiple lighting circuits, which form a star connection with the distribution box; a complex network structure is formed among transformers, busbars and switch cabinets.

[0090] It can be understood that for a system aiming to conduct in-depth energy consumption analysis and optimization, only the management topology is not enough. It is necessary to deeply integrate the management dimension with the physical energy flow dimension. Only by accurately mastering the actual path of the current, can precise scheduling, fault impact range analysis and energy flow tracing based on real constraints (such as line capacity and transformer load) be realized. The association of the two topologies in the digital twin forms a complete “energy information model”.

[0091] The data of the membership topology mainly comes from the asset management system, the space database or manual input. It contains information such as the location code of the device and the department to which it belongs. The data of the electrical connection topology comes from electrical design drawings (such as single-line diagrams), point table configurations of intelligent power distribution systems, or is obtained through circuit identification technology.

[0092] The principle of the present embodiment is described as follows:

[0093] (1) Data import and processing: the above two kinds of topology data are imported into the digital twin platform. Data cleaning and conversion may be needed to ensure that the naming is consistent.

[0094] (2) Entity alignment: through the unique device identifier (such as asset number), location information or circuit number, “device A” in the administrative topology and “load controlled by circuit breaker B” in the electrical topology are associated as the same entity.

[0095] (3) Relationship establishment: In the digital twin model, one device instance will have two relationship attributes at the same time. For example, "Air Conditioner-01" has one relationship pointing to "Conference Room" (membership topology) and one relationship pointing to "Lighting Circuit" (electrical connection topology).

[0096] (4) Visualization and query: The platform should be able to display and query both topologies at the same time. For example, clicking on a power distribution box can display all the spaces and devices it manages (membership topology view) and all the electrical circuits and loads under it (electrical connection topology view).

[0097] In this implementation, the digital twin is upgraded from a "location map" to an "energy map" through the fusion of dual topologies, enabling the system to perform advanced analysis based on real physical connections, such as accurate load forecasting, line loss analysis, rapid fault location and isolation, and avoiding control inefficiency or safety accidents caused by inaccurate models.

[0098] In one or more embodiments of the present application, when constructing the digital twin of each park in step S200, a digital energy consumption benchmark fingerprint is dynamically created and bound for each device, which is composed of device static attributes, dynamic operating parameters, and historical statistical characteristics. The static attributes at least include device type, rated power, factory date, and energy efficiency level, the dynamic operating parameters are power consumption data collected by the device under normal operating conditions for at least 24 hours, and the historical statistical characteristics are steady-state power consumption Gaussian distribution parameters calculated from the power consumption data, including mean μ and variance σ².

[0099] It can be understood that the energy consumption benchmark fingerprint is a quantitative, multi-dimensional energy consumption characteristic established for a specific device. It is like the "energy consumption DNA" of the device, which is the unique identification of the energy consumption pattern under normal operating conditions. Steady-state power consumption refers to the power consumption of the device when it is running stably near the rated load after excluding transient conditions such as startup and shutdown.

[0100] Gaussian distribution parameters: mean (μ) and variance (σ²). The mean represents the typical level of power consumption, and the variance represents the normal fluctuation range of power consumption. Using it to describe power consumption is based on the statistical assumption that there is normal random fluctuation in device operation.

[0101] In this implementation, the rule is learned and quantified through historical data to establish a reliable normal benchmark. Any behavior that significantly and continuously deviates from this benchmark can be considered abnormal.

[0102] The principles of the present embodiment are described as follows:

[0103] (1) Data preparation: select the high-frequency power consumption data (e.g. one point per minute) of the equipment under the known healthy state, continuously running for at least 24 hours (preferably covering different working days / weekend modes).

[0104] (2) Feature calculation: including static attribute and dynamic parameter calculation: static attribute includes direct extraction from equipment account and storage in database field. Dynamic parameter calculation includes eliminating obvious abnormal points, identifying the period when the equipment is in stable operation through algorithm, and calculating the average value (μ) and standard deviation (σ) of the power consumption data of the steady state section. The variance (σ²) is the square of the standard deviation.

[0105] (3) Fingerprint binding and storage: store and associate μ and σ² calculated as two attributes of the digital twin instance of the equipment.

[0106] (4) Missing data interpolation: when data is missing at a certain time, a random number can be generated using N(μ, σ²) distribution or directly using μ value for filling. It is more consistent with the actual situation than using fixed value or previous and next mean value interpolation.

[0107] (5) Abnormality detection: real-time calculation of the deviation of the current power consumption from μ. If the deviation continues to exceed K times σ (such as K=3, i.e. 3σ principle), an abnormal alarm is triggered. This is more suitable for normal fluctuations of the equipment under different seasons and different loads than a fixed threshold.

[0108] In this implementation, the equipment management is improved from "data existence" to "data quality and data value" through the energy consumption benchmark fingerprint. The effects are as follows: significantly improving data quality to provide cleaner input for prediction models; realizing early and accurate equipment failure or energy efficiency degradation warning, supporting predictive maintenance, avoiding energy waste and equipment sudden failure; providing individualized and data-driven measurement standards for energy efficiency audit and optimization.

[0109] In one or more embodiments of the present application, the federated learning process in step S300 adopts a hybrid architecture of long and short time window cooperation; wherein the long time window model takes long-term data of 7 to 30 days in the past as the training set; the short time window model takes recent data of 1 to 24 hours in the past as the training set; in the cloud server, an adaptive weight fusion mechanism is designed, and its formula is represented as:

[0110] W = α W_long + (1-α) W_short;

[0111] Wherein, W represents the weight parameter of the fused prediction model;

[0112] W_long represents the weight of the long-time window model, which contains seasonal, periodic trend characteristics of a long period (7 to 30 days);

[0113] W_short represents the weight of the short-time window model, which contains instantaneous fluctuations, mutations of a short period (1 to 24 hours);

[0114] α is an adaptive weight factor, which is a dynamically changing coefficient, and its value range is [0, 1], and its size is determined by the variance of photovoltaic power fluctuation. If the variance is large, α is adjusted to be small, so that the model trusts the short-time window model more; if the variance is small, α is adjusted to be large, so that the model trusts the long-time window model more.

[0115] It can be understood that the long-time window model is a model trained using historical data of a longer time span (such as 7-30 days). Its advantage is that it learns the macro rules of energy consumption, such as diurnal cycle, weekly cycle, seasonal trend, and the model is robust but slow in response. The short-time window model is a model trained using recent (such as 1-24 hours) data. Its advantage is that it can quickly capture the latest changes, such as sudden weather events and temporary activities that cause load fluctuations, and the model is sensitive but may lack long-term vision.

[0116] The adaptive weight factor (α) is a dynamically changing fusion coefficient, and its value is determined by external environmental conditions (here, photovoltaic volatility), which is used to intelligently allocate the "speaking rights" of long and short-time window models in the final prediction.

[0117] The implementation principle is described as follows:

[0118] (1) In each local park, two prediction models with the same structure but independent initial weights are trained using long-time window data (such as the past 30 days) and short-time window data (such as the past 24 hours), respectively;

[0119] (2) The cloud server continuously receives the recent photovoltaic power sequence (such as every second data of the past 1 hour) reported by each park, and calculates the variance of the data in this time window. The larger the variance, the more intense the light changes, and the higher the uncertainty;

[0120] (3) A mapping function is pre-set. For example, a function can be designed such that when the photovoltaic variance is 0 (extremely stable), α = 1 (completely trust the long-time window model); when the photovoltaic variance exceeds a maximum threshold (extremely unstable), α = 0 (completely trust the short-time window model); and the intermediate state is linear or nonlinear interpolation.

[0121] (4) After receiving the long and short-time window model weights (W_long and W_short) of each park, the cloud server calculates the α value dedicated to this park according to the formula W = α W_long + (1-α) W_short are fused to obtain the weight parameter W of the customized prediction model for the park.

[0122] (5) The fused weight W is issued to the corresponding park for local prediction.

[0123] In this embodiment, the adaptive fusion mechanism greatly enhances the adaptability of the prediction model to the intermittency and volatility of renewable energy. The effect is that when the weather is stable, the prediction result is smooth and consistent with the long-term trend; when the weather changes suddenly, the prediction result can quickly track the instantaneous change of the load, significantly reducing the prediction error, especially avoiding the prediction lag problem of traditional models when photovoltaic output changes suddenly, providing a more reliable time window and decision basis for subsequent accurate scheduling.

[0124] In one or more embodiments of the present application, the said elastic energy consumption budget in step S400 is a mechanism with upper and lower boundaries and a dynamically adjustable width, and the decision logic is as follows: the system monitors the operation state of each park microgrid in real time, continuously calculates the surplus power of the current photovoltaic power generation and the adjustable capacity of the energy storage system; if the energy consumption prediction data of the park exceeds the upper limit of the budget, but the system determines that there is surplus photovoltaic power at present, the strip widening strategy is automatically triggered, the said elastic energy consumption budget is temporarily relaxed to be less than or equal to 70% of the upper limit of the budget of the park, and the instruction to guide the consumption of local surplus photovoltaic power is generated first; if the prediction data exceeds the upper limit of the budget and the park microgrid is in a state of insufficient photovoltaic output or relying on grid power supply, the tightening strategy is automatically triggered, the said elastic energy consumption budget is temporarily tightened to be less than or equal to 70% of the upper limit of the budget of the park, and the instruction to reduce the running intensity of the energy-using equipment in the park is generated.

[0125] It can be understood that the surplus photovoltaic power refers to the remaining power after the current photovoltaic power generation power is subtracted from the current total load in the park. The surplus power is positive, indicating that there is green power available for consumption; negative indicates that power needs to be purchased from the grid or energy storage is used. The strip widening / tightening strategy is a control strategy for dynamically adjusting the budget constraint strength according to the energy supply and demand situation. Widening means relaxing the restrictions and encouraging electricity consumption; tightening means strengthening the restrictions and requiring energy saving.

[0126] The principle of the present embodiment is described as follows:

[0127] (1) Mechanism setting: In the system management interface, the administrator sets a basic energy consumption budget target value (M) and a maximum allowed floating percentage (Δ%). Thus, an initial elastic strip [M (1-Δ%), M (1+Δ%)] is formed.

[0128] (2) Real-time monitoring and decision making: Calculate the surplus power: photovoltaic surplus power = photovoltaic power generation power - current total load.

[0129] (3) Decision logic:

[0130] Scenario 1 (Broaden): IF (Predicted Energy Consumption > Budget Upper Limit) AND (Photovoltaic Surplus Power > Threshold) THEN: Trigger the "Broaden" strategy, the system temporarily adjusts the budget upper limit to Budget Upper Limit + Budget Relaxation Increment ΔE (ΔE is an increment proportional to the surplus power), and generates instructions to guide the system to consume the "extra" budget to consume the surplus photovoltaic power, such as "Increase the energy storage charging power to the maximum value", "Increase the electric vehicle charging power upper limit by 20%".

[0131] Scenario 2 (Tighten): IF (Predicted Energy Consumption > Budget Upper Limit) AND (Photovoltaic Surplus Power < 0) THEN: Trigger the "Tighten" strategy, the system strictly implements or even strengthens the budget constraints, for example, tighten the effective budget upper limit to 70% of the original upper limit, and generate energy-saving instructions such as "Uniformly increase the office air conditioning set temperature by 1°C", "Turn off the landscape lighting of non-main corridors".

[0132] Strategy execution: The generated instructions are sent to the S500 step for execution.

[0133] In this implementation, the energy consumption management strategy becomes "intelligent" and "flexible" through the elastic energy consumption budget mechanism. It avoids wasting green energy to strictly adhere to the budget in sunny weather, and ensures that energy-saving measures can be taken decisively when energy is tight. Ultimately, in the life cycle, the system adheres to the overall energy consumption control target while significantly improving the penetration rate of renewable energy, reducing carbon emissions, and achieving a win-win of economic and environmental benefits.

[0134] In one or more embodiments of the present application, before the set of energy-saving control instructions is issued to a certain park for execution in step S500, a centralized instruction conflict arbitration and verification step is introduced: this step traces all instructions receiving devices based on the digital twin of the park, when detecting that the same device is assigned two or more instructions with logical contradictions or operation exclusions within the same time window, the arbitration rule is triggered; the arbitration rule is based on a set of pre-set energy sensitivity sorting knowledge base, which quantitatively sorts the sensitivity of all types of controlled devices and their control instructions to the global energy consumption impact, the arbitration engine preferentially selects the instruction with the largest contribution to global energy saving effect and the highest sensitivity, while the conflicting secondary instructions are temporarily stored in the cache queue, delayed to the next control period or evaluated and issued after the premise changes.

[0135] It can be understood that the instruction conflict refers to that in an automation system, when two or more independent control logics or strategies based on different optimization targets (such as comfort energy saving, demand response) issue operation instructions (such as one instruction requiring "cooling" and the other requiring "heating") that are contradictory or mutually exclusive in content to the same actuator (such as an air conditioner) in the same time period, this phenomenon is called instruction conflict. The arbitration rule is a set of pre-defined decision logic for resolving disputes. It specifies the criteria for selecting an optimal instruction from multiple candidate instructions for execution when a conflict occurs. Energy sensitivity ranking refers to a quantitative indicator for evaluating the sensitivity and contribution of different devices or different control instructions to the global energy consumption. For example, the energy saving effect of turning off a large centrifugal cold water machine is several orders of magnitude higher than that of turning off an LED lamp, and the energy sensitivity of the former is much higher than that of the latter.

[0136] The principles of the present embodiment are described as follows:

[0137] (1) Knowledge base construction: It needs to be completed by domain experts and data analysis, and a "sensitivity score" is assigned to all controllable device types and their typical control instructions (such as air conditioner set temperature increase / decrease 1℃, lighting dim to 70%, charging pile power limit to 50%, etc.) in the system. This score can be calculated based on historical data statistics (how many kilowatt-hours of electricity can be saved on average per hour after the instruction is executed) or physical model calculation. The higher the score, the higher the energy saving priority of the instruction.

[0138] (2) Conflict detection: In the instruction scheduling layer, a "instruction-device" mapping table is maintained. Before the instruction is issued, the system will check whether the same device has been assigned other instructions in the upcoming control period. If so, the conflict detection algorithm is started to judge whether these instructions are logically exclusive (for example, one instruction requires "on" and the other requires "off").

[0139] (3) Arbitration execution: Once a conflict is detected, the arbitration engine queries the knowledge base to compare the sensitivity scores of all conflicting instructions. The instruction with the highest score is immediately issued for execution. For instructions that are not selected, they are not simply discarded, but are placed in a delay queue and their trigger conditions are recorded. The system will re-evaluate whether the execution conditions of these cached instructions still meet the conditions in the next control period, or after the prerequisite conditions change (such as the execution of high-priority instructions), the instructions will be issued again.

[0140] (4) Log and learning: All conflict events and arbitration results should be recorded for subsequent analysis, and even feedback to the knowledge base for optimizing the sensitivity score.

[0141] The embodiment greatly improves the reliability and intelligent level of a complex automation system through this mechanism, avoids frequent actions of an actuator due to receiving contradictory instructions, reduces equipment wear and tear, prolongs the service life of the equipment, ensures that the system can execute the most effective energy-saving action under any circumstances, prevents mutual offset among optimization strategies, and guarantees the realization of the overall energy efficiency target; and provides crucial coordination protection for the collaborative work of multiple systems and multiple strategies.

[0142] In one or more embodiments of the present application, in step S500, the energy-saving control instruction set supports cross-system scenario linkage based on a digital twin topology, specifically embodied as: when the system predicts that the total load of a certain park charging pile is about to exceed the safety threshold of the upper transformer capacity, the control logic does not simply cut off the charging power supply, but first accurately locates all intelligent lighting loops under the same transformer power supply loop as the charging pile through the query of the digital twin; then, automatically generates an instruction to uniformly down-regulate the dimming values of these lighting loops to a preset level without affecting the safety lighting, so as to reduce the total load of the transformer; and after the charging pile load peak, automatically restores the lighting dimming value to 100%.

[0143] It can be understood that the electrical topology refers to the electrical connection relationship between elements (such as transformers, distribution cabinets, circuit breakers, and loads) in a power supply system, which clearly describes the distribution path and upstream and downstream relationship of electrical energy. The transformer capacity refers to the maximum apparent power (unit: kVA) that the transformer can safely bear for a long time, which is a key bottleneck constraint of the power supply system. Flexible load refers to a load whose operating power can be adjusted within a certain range without causing substantial impact on its main function, or a load whose power variation is not sensitive to the user. Intelligent lighting is a typical flexible load, and dimming within a certain range does not affect the basic lighting function.

[0144] The principle of the embodiment is described as follows:

[0145] (1) Overload prediction: The system monitors the load rate of the transformer in real time, and combines the reservation information or historical data of the charging pile to predict whether the total load in the near future (such as 1 hour) will exceed the safety threshold (such as 95% of the transformer capacity).

[0146] (2) Topology query: Once the overload risk is predicted, the system immediately queries the digital twin. According to the electrical topology relationship, the risk transformer is traced upwards, and all power distribution loops supplied by the transformer are found downwards, and the “intelligent lighting loop” among them is further screened out as a flexible resource pool that can be controlled.

[0147] (3) Strategy formulation and execution:

[0148] (4) Safety constraint: Ensure the post-regulation lighting brightness is still above the minimum standard of safe lighting (e.g. 30%).

[0149] (5) Control instruction: The system automatically generates an instruction to uniformly set the dimming value of the selected lighting circuit to a pre-set, acceptable energy-saving level (e.g. from 100% to 70%). This reduction is calculated to be sufficient to offset the predicted overload.

[0150] (6) Instruction issuance: The dimming instruction is issued through the control system.

[0151] (7) Recovery mechanism: The system continuously monitors the transformer load. When the peak load of the charging pile subsides and the total load falls below the safety level, the system automatically generates a recovery instruction to gradually or instantaneously restore the lighting dimming value to the normal level (100%).

[0152] The effects of this embodiment are extremely significant: first, it solves the bottleneck problem of local power capacity shortage in a highly cost-effective way (nearly zero hardware cost), avoiding expensive transformer capacity expansion and saving a lot of expenditure. Second, it guarantees the continuity of critical business (electric vehicle charging), avoiding the decline in user experience and revenue loss caused by direct power limitation of charging piles. Finally, it fully taps and utilizes the adjustment potential of existing assets, and is a model application for building a flexible, efficient and resilient new power system.

[0153] In one or more embodiments of the present application, the digital twin state and the local prediction model weight parameters of each park are updated respectively using the feedback data in step S600, including: adopting a gradient truncation incremental learning strategy: when the system receives the equipment execution feedback data, if it is found that the error between it and the model expected state is greater than the preset value, the update algorithm is updated; wherein the update algorithm is not a global retraining of the entire federated learning model, but only a local fine-tuning of the fully connected weights of the last one or two layers of the model, and a very low learning rate is used for iteration; at the same time, the feature extraction weights of the intermediate layers of the model are completely frozen to prevent damage or coverage to the past learned and well-generalized knowledge.

[0154] It can be understood that incremental learning refers to the ability of a machine learning model to continuously learn new knowledge from newly arrived data without forgetting existing knowledge. It is also called continuous learning or online learning. Gradient truncation does not refer to gradient clipping in optimization, but refers to truncating or limiting the path of gradient backpropagation when updating the model, i.e. only allowing the error gradient to update part of the network weights, while the other part of the weights is "frozen" and remains unchanged.

[0155] The principle of the embodiment is described as follows:

[0156] (1) Model structure: Assume the federated learning model used is a deep neural network, containing an input layer, multiple hidden layers (intermediate layers), and an output layer (the last one or several fully connected layers).

[0157] (2) Update trigger: The system receives feedback data from the device after execution, compares the model's previous prediction with the actual value, and calculates the error (such as mean square error). When the error exceeds a preset threshold (such as 10%), the incremental update process is triggered.

[0158] (3) Freeze operation: Before the backpropagation algorithm starts, set the weights of all layers in the network except the last one or two fully connected layers to "untrainable" (i.e., frozen state). This means that during the subsequent optimization process, the gradients of these weights will be calculated but not updated.

[0159] (4) Local fine-tuning: Only the last few layers of weights that are not frozen are trained using new feedback data (usually a small batch) and a very low learning rate (such as 0.001). Since the number of parameters to be adjusted is greatly reduced, and the learning rate is very low, this fine-tuning process is gentle and controlled.

[0160] (5) Model update: After fine-tuning, replace the original model weights with the updated weights. At this point, the model's feature extraction core has not changed much, but the final decision boundary has been slightly adjusted based on new data.

[0161] This strategy ensures the stability and adaptability of the model during long-term deployment. It allows the model to keep pace with the times, learn from new operational data, and correct possible biases, thereby maintaining prediction accuracy. More importantly, it effectively prevents "catastrophic forgetting" and ensures that the model does not suddenly forget valuable knowledge such as seasonal patterns learned over the past few months after learning new usage patterns. This is crucial for industrial systems that require 7x24 hour stable operation, avoiding prediction failures and scheduling chaos caused by improper model updates.

[0162] See Figure 2In another embodiment of the present application, a multi-garden control system for implementing a multi-garden energy consumption prediction scheduling method based on digital twinning is provided, comprising: a cloud server 20, wherein a digital twinning module 21 is configured to construct and maintain a four-level topological digital twinning model containing membership and electrical connection relationship based on project, building, space and equipment information of each garden; an energy consumption prediction module 22 connected to the digital twinning module 21, configured to access energy consumption, environmental and micro-grid operation data of each garden, and generate energy consumption prediction data for future period based on federated learning algorithm; a prediction scheduling decision module 23 connected to the energy consumption prediction module 22 and the digital twinning module 21, configured to compare the energy consumption prediction result with a preset energy consumption budget, and generate energy-saving control instruction set for heating, lighting, energy storage and charging pile system; an operation linkage module 24 connected to the energy consumption prediction module 22, configured to automatically generate and dispatch inspection work orders to the operation service module when predicting device energy consumption anomaly; a plurality of edge collection control nodes 25 respectively deployed in each garden, configured to collect local data and upload to the cloud server 20, and receive and execute the energy-saving control instruction set issued; a micro-grid monitoring module 26 integrated in the cloud server 20 or the plurality of edge collection control nodes 25, configured to monitor the operation state of photovoltaic, energy storage, charging pile and power distribution system in real time, and provide state data to the energy consumption prediction module 22 and the prediction scheduling decision module 23; wherein the digital twinning module 21, energy consumption prediction module 22, prediction scheduling decision module 23, operation linkage module 24, plurality of edge collection control nodes 25 and the micro-grid monitoring module 26 work together to form a control system for cross-garden data collection, prediction analysis, optimization decision and closed-loop execution.

[0163] The multi-garden control system of the present embodiment provides a high-performance, high-reliability and scalable technical carrier, enabling all the above methods to be implemented. It optimizes the allocation of computing resources, ensures data security and business continuity, and due to the modular design, the system is easy to maintain, upgrade and expand functions, and can adapt to future business changes.

[0164] In one or more embodiments of the present application, the operation linkage module 24 is configured to continuously monitor the device-level energy consumption anomaly prediction result and alarm event output by the prediction scheduling decision module 23; through an internal API interface, automatically call the work order creation function of the operation service, convert the abnormal event into an executable prediction inspection work order, and automatically assign an inspection priority for inspection.

[0165] It can be understood that the operation linkage is an advanced operation strategy. It predicts the future failure probability or performance degradation trend of the equipment through real-time or near real-time monitoring and analysis of the equipment state, so as to actively arrange maintenance activities before the failure occurs. The API interface (Application Programming Interface) is a set of predefined functions and protocols that allow different software applications to communicate and exchange data with each other. Here it is used to connect the energy consumption prediction system and the work order management system.

[0166] The principle of the embodiment is described:

[0167] Event listening: This module acts as a background service, continuously subscribing or listening to the message bus from the "prediction scheduling decision module". The event type it is interested in is "device-level energy consumption abnormal warning", which contains at least device ID, prediction deviation value, timestamp and other information.

[0168] Work order generation logic: When receiving the warning event, the logic inside the module will make a judgment (such as whether the deviation has been continuously exceeding the threshold for a certain period of time). If it is determined that a work order needs to be generated, the work order data is prepared.

[0169] API call: The module calls the RESTful API interface provided by the operation service module (which may be a separate work order management system such as Maximo, ServiceNow or a self-developed system) to create a work order, and passes the required information in structured data (usually in JSON format). The data passed includes: work order title (such as "air conditioner AHU-1 energy consumption abnormal warning"), associated equipment, description information (including predicted abnormal data), priority (which can be automatically set to "high"), recommended measures (such as "check filter, calibrate sensor").

[0170] Work order dispatch: After the operation service module receives the API request, it automatically creates a work order and dispatches it to the corresponding operation team or engineer according to the preset rules.

[0171] Closed-loop feedback (optional enhancement): The module can also provide a reverse interface to receive the processing results of the work order (such as "filter replaced"). This result can be used to update the "energy consumption baseline fingerprint" of the equipment (see claim 3), forming a complete "monitoring-prediction-operation-verification" closed loop.

[0172] This embodiment achieves proactive and intelligent operation and maintenance activities through coordinated operation and maintenance. It significantly advances the time of equipment failure detection, shifting from "reactive repair" to "proactive maintenance," thereby avoiding unplanned downtime and higher maintenance costs caused by minor issues escalating into major failures. Simultaneously, it reduces continuous energy waste caused by suboptimal or inefficient equipment operation, achieving synergistic optimization of energy conservation and asset management, and significantly improving overall operational efficiency and economic benefits.

[0173] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to execute the aforementioned multi-campus energy consumption prediction and scheduling method based on digital twins.

[0174] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0175] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0176] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0177] In particular implementations, the processor 301, the input device 302, and the output device 303 described in the embodiments of the present application can perform the implementation described in any embodiment of the multi-garden energy consumption prediction and scheduling method based on digital twinning provided by the embodiments of the present application, and can also perform the implementation of the electronic device described in the embodiments of the present application, which will not be described here.

[0178] In another embodiment of the present application, a computer-readable storage medium is provided, which stores a computer program. The computer program includes program instructions, which, when executed by a processor, implement all or part of the processes of the multi-garden energy consumption prediction and scheduling method based on digital twinning described in the above embodiments. The computer program can also be used to instruct related hardware to complete the implementation. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0179] The computer-readable storage medium can be an internal storage unit of the electronic device of any of the preceding embodiments, such as a hard disk or a memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer-readable storage medium can include both the internal storage unit and the external storage device of the electronic device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0180] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0181] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device and the unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0182] In several embodiments provided in the present application, it should be understood that the disclosed electronic device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, and the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces or units, and can also be electrical, mechanical or other forms of connection.

[0183] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0184] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software functional unit.

[0185] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-garden energy consumption prediction scheduling method based on digital twinning, characterized in that, Comprising the following steps: S100: data acquisition step, collecting real-time operation data, energy consumption data and environmental parameters of multiple distributed park energy systems and microgrids in parallel; S200: digital twin construction step, based on the collected data, a digital twin containing a four-level topological relationship of project-building-space-equipment is constructed for each park, and a global unified model is formed by summarizing; S300: federated learning prediction step, taking the global unified model as the benchmark, driving the federated learning process, generating energy consumption prediction data of each park in the future period by fusing the weight parameters of the local prediction model of each park; S400: elastic decision step, comparing the energy consumption prediction data of each park with the elastic energy consumption budget, if the difference exceeds the preset value, generating a set of energy-saving control instructions for at least one of the energy storage, heating, lighting and charging pile devices; wherein the elastic energy consumption budget in step S400 is a mechanism with upper and lower boundaries and a dynamically adjustable width, and its decision logic is as follows: the system monitors the operation state of the microgrid of each park in real time, continuously calculates the surplus power of the current photovoltaic power generation and the adjustable capacity of the energy storage system; if the energy consumption prediction data of the park exceeds the upper limit of the budget, but the system determines that there is surplus photovoltaic power at present, the strip widening strategy is automatically triggered, the elastic energy consumption budget is temporarily relaxed to be the upper limit constraint value of the park budget, and instructions are generated to guide the consumption of local surplus photovoltaic power; if the prediction data exceeds the upper limit of the budget and the microgrid of the park is in a state of insufficient photovoltaic output or relying on grid power supply, the tightening strategy is automatically triggered, the elastic energy consumption budget is temporarily tightened to be less than or equal to 70% of the upper limit constraint value of the park budget, and instructions are generated to reduce the running intensity of energy-consuming devices in the park; S500: instruction execution step, analyzing the device topological relationship of each park, issuing and executing the set of energy-saving control instructions, and collecting feedback data; wherein the set of energy-saving control instructions supports cross-system scenario linkage based on digital twin topology, which is specifically embodied as: when the system predicts that the total load of the charging pile in a park will exceed the safety threshold of the capacity of its upper transformer, the control logic does not simply cut off the charging power, but first locates all intelligent lighting loops that are in the same transformer power supply loop as the charging pile through querying the digital twin; then, automatically generate instructions to lower the dimming value of these lighting loops to a preset level without affecting the safety lighting, thereby reducing the total load of the transformer; after the charging pile load peak, automatically restore the lighting dimming value to 100%; S600: closed-loop update step, updating the digital twin state and local prediction model weight parameters of each park using feedback data.

2. The method of claim 1, wherein, The specific process of constructing a digital twin in step S200 includes: Based on the device access information and the spatial layout diagram, a project-building-space-device hierarchical path to which the device belongs is automatically generated to form a membership topology; at the same time, based on the membership relationship data of the smart meter, the circuit breaker and the distribution box, an electrical connection topology between the devices is automatically generated; the digital twin is a fusion of the membership topology and the electrical connection topology.

3. The method according to claim 1 or 2, characterized in that, In step S200, when constructing the digital twin of each park, a digital energy consumption benchmark fingerprint is dynamically created and bound for each device, which is composed of device static attributes, dynamic operating parameters and historical statistical characteristics; Wherein, the static attributes at least include device type, rated power, factory date and energy efficiency level, the dynamic operating parameters are power consumption data collected by the device under normal working condition for at least 24 hours, and the historical statistical characteristics are steady-state power consumption Gaussian distribution parameters calculated from the power consumption data, including mean μ and variance σ².

4. The method according to claim 1 or 2, characterized in that, In step S300, the federated learning process adopts a hybrid architecture of long and short time window cooperation; wherein the long time window model takes the long-term data of the past 7 to 30 days as the training set; the short time window model takes the recent data of the past 1 to 24 hours as the training set; in the cloud server, an adaptive weight fusion mechanism is designed, and its formula is represented as: W = α W_long + (1-α) W_short; Wherein, W represents the weight parameter of the fused prediction model; W_long represents the weight of the long time window model, which contains long-term seasonal and periodic trend characteristics; W_short represents the weight of the short time window model, which contains short-term instantaneous fluctuation and mutation characteristics; Alpha is an adaptive weight factor, which is a dynamic coefficient, and its value range is [0, 1], which is determined by the variance of photovoltaic power fluctuation. If the variance is large, alpha is adjusted to be small, so that the model trusts the short time window model more; if the variance is small, alpha is adjusted to be large, so that the model trusts the long time window model more.

5. The method according to claim 1 or 2, characterized in that, In step S500, before the energy-saving control instruction set is issued to a certain park for execution, a centralized instruction conflict arbitration and verification step is introduced: this step traces all receiving devices of the instructions based on the digital twin of the park, and when it is detected that the same device is assigned two or more instructions with logical contradictions or operation exclusions in the same time window, the arbitration rule is triggered; the arbitration rule is based on a set of pre-set energy consumption sensitivity sorting knowledge base, which quantitatively sorts the sensitivity of all types of controlled devices and their control instructions to global energy consumption, and the arbitration engine preferentially selects the instruction with the largest contribution to global energy saving effect and the highest sensitivity, while the conflicting secondary instructions are temporarily stored in the cache queue and delayed to the next control period or evaluated and issued after the premise changes.

6. The method of claim 1 or 2, wherein, The digital twin state and the local prediction model weight parameter of each park are updated respectively in step S600 using the feedback data, including: adopting a gradient truncation incremental learning strategy: when the system receives device execution feedback data, if it is found that the error between it and the expected state of the model is greater than the preset value, the update algorithm is updated; wherein the update algorithm is not a global retraining of the entire federated learning model, but only a local fine-tuning of the fully connected weights of the last one or two layers of the model, and a very low learning rate is used for iteration; at the same time, the feature extraction weights of the intermediate layers of the model are completely frozen to prevent damage or coverage to the past learned and well-generalized knowledge.

7. A digital-twin-based multi-park control system for implementing the method of any one of claims 1-6, characterized by, Comprise: A cloud server, wherein the cloud server is configured with: A digital twin module, configured to construct and maintain a four-level topological digital twin model containing membership and electrical connection relationship based on the project, building, space and equipment information of each park; An energy consumption prediction module connected to the digital twin module, configured to access the energy consumption, environment and micro-grid operation data of each park, and generate energy consumption prediction data for future period based on federated learning algorithm; A prediction scheduling decision module connected to the energy consumption prediction module and the digital twin module, configured to compare the energy consumption prediction data with the preset energy consumption budget, and generate energy saving control instruction set for heating, lighting, energy storage and charging pile system; An operation linkage module connected to the energy consumption prediction module, configured to automatically generate and dispatch inspection work orders to the operation service module when predicting device energy consumption anomaly; A plurality of edge collection control nodes respectively deployed in each park, configured to collect local data and upload to the cloud server, and receive and execute the energy saving control instruction set issued; A micro-grid monitoring module integrated in the cloud server or the plurality of edge collection control nodes, configured to monitor the operation state of photovoltaic, energy storage, charging pile and power transformation and distribution system in real time, and provide state data to the energy consumption prediction module and the prediction scheduling decision module; Wherein, the digital twin module, energy consumption prediction module, prediction scheduling decision module, operation linkage module, micro-grid monitoring module and the plurality of edge collection control nodes work together to form a control system for cross-park data collection, prediction analysis, optimization decision and closed-loop execution.

8. The control system of claim 7, wherein, The operation linkage module is configured to continuously monitor the device-level energy consumption anomaly prediction results and alarm events output by the prediction scheduling decision module; through an internal API interface, automatically call the work order creation function of the operation service module, convert the abnormal events into executable prediction inspection work orders, and automatically assign inspection priority for inspection.

Images