Distributed energy intelligent allocation method based on Internet of Things big data in smart city

Abstract

The invention relates to the field of distributed energy, in particular to a distributed energy intelligent allocation method based on Internet of Things big data in a smart city, which comprises three steps of data acquisition, processing and output. Firstly, photovoltaic, energy storage and charging pile real-time data are obtained through a multi-protocol analysis engine, and meteorological cloud charts, traffic thermodynamic diagrams and building energy consumption historical data are combined. And secondly, constructing a dynamic weight model by utilizing a space-time joint attention mechanism, generating a global wind and light output prediction network through federated learning, constructing an elastic scheduling strategy library by adopting a multi-target dynamic game algorithm, optimizing a scheduling model on a digital twin platform, and finally generating a real-time scheduling scheme through a two-stage increment optimization architecture. The method effectively improves the efficiency, economical efficiency, reliability and environmental protection property of an energy system, provides a powerful tool for urban managers, guarantees stable, economical and clean energy supply for residents, and promotes sustainable development of smart cities.

Description

Technical Field

[0001] The present invention relates to the field of distributed energy, and particularly to an intelligent dispatching method for distributed energy based on Internet of Things big data in a smart city. Background Art

[0002] With the acceleration of the urbanization process and the in-depth implementation of the concept of sustainable development, the construction of smart cities has become an important trend in global urban development. In this context, distributed energy systems are gradually becoming the core components of the energy infrastructure in smart cities due to their flexibility, environmental friendliness, and high efficiency. However, the high degree of dispersion, intermittency, and uncertainty of distributed energy systems pose great challenges to their efficient management and scheduling.

[0003] Currently, the management and scheduling of distributed energy in smart cities mainly rely on traditional centralized control methods and simple prediction models. These methods often fall short when dealing with large-scale and highly dynamic distributed energy systems. Traditional rule-based scheduling strategies are difficult to adapt to complex and changing energy supply and demand situations, resulting in low energy utilization efficiency and the inability to fully absorb renewable energy. Although simple machine learning prediction models have improved prediction accuracy to a certain extent, they are still difficult to capture the complex spatio-temporal relationships in the energy system, especially in extreme weather or emergency situations, where the prediction errors are relatively large.

[0004] In addition, existing distributed energy management systems generally suffer from the problem of data islands. It is difficult to effectively share and integrate data between different energy facilities and users, which not only limits the overall optimization ability of the system but also increases the risks of data security and privacy protection. At the same time, existing systems often adopt simplified processing or a divide-and-conquer approach when dealing with multi-objective optimization problems, and it is difficult to find the best balance among multiple objectives such as economy, environmental friendliness, and system stability.

[0005] In terms of system response speed and fault recovery ability, existing technologies also face severe challenges. Traditional scheduling systems often lag in responding to load changes and equipment failures, and it is difficult to adjust strategies in a timely manner, resulting in a decline in system stability and even triggering a chain reaction. Especially in the face of extreme weather or large-scale emergencies, the emergency handling ability of existing systems is stretched thin and it is difficult to quickly and effectively resume normal operation.

[0006] In view of the above problems, there is an urgent need for an innovative method that can comprehensively improve the management and scheduling level of distributed energy in smart cities. This method should be able to effectively integrate Internet of Things big data, achieve high-precision energy prediction and multi-objective optimization scheduling, and at the same time have fast response and adaptive capabilities to cope with complex and changing operating environments. Summary of the Invention

[0007] The intelligent distributed energy allocation method based on Internet of Things big data in the smart city of the present invention is precisely proposed to address the above technical problems. This method constructs a comprehensive, efficient, and flexible intelligent distributed energy allocation system by innovatively integrating advanced technologies such as a multi-protocol adaptive parsing engine, a spatio-temporal joint attention mechanism, federated learning, and a multi-objective dynamic game algorithm.

[0008] The present invention proposes an intelligent distributed energy allocation method based on Internet of Things big data in a smart city, including:

[0009] An acquisition step, including:

[0010] Obtain the real-time operation data of the photovoltaic array, energy storage power station, and charging pile through a multi-protocol adaptive parsing engine;

[0011] Obtain meteorological satellite cloud images, traffic flow heat maps, and building energy consumption historical data;

[0012] A processing step, including:

[0013] Based on the real-time operation data, meteorological satellite cloud images, traffic flow heat maps, and building energy consumption historical data, construct a dynamic weight allocation model using a spatio-temporal joint attention mechanism;

[0014] According to the dynamic weight allocation model, generate a global wind and solar power output prediction network through a region energy portrait system driven by federated learning;

[0015] Based on the global wind and solar power output prediction network, construct an elastic scheduling strategy library using a multi-objective dynamic game algorithm;

[0016] According to the elastic scheduling strategy library, simulate extreme scenarios on a digital twin simulation platform to optimize the scheduling model;

[0017] Based on the optimized scheduling model, generate a real-time scheduling plan through a two-level incremental optimization architecture;

[0018] An output step, including:

[0019] Output the real-time scheduling plan for the intelligent allocation of the smart city distributed energy system.

[0020] Preferably, the construction of the multi-protocol adaptive parsing engine specifically includes:

[0021] Adopt a protocol intelligent recognition model based on meta-learning to support multiple communication protocols such as Modbus, IEC 61850, and MQTT;

[0022] According to the device type, divide data collection into acquisition-based data and request-based data;

[0023] For the collected data, the device data is actively obtained through the corresponding protocol;

[0024] For the requested data, the data request of the device is responded through the corresponding protocol.

[0025] Preferably, the construction of the spatio-temporal joint attention mechanism specifically includes:

[0026] Use the long short-term memory network to construct a time-domain graph and extract the short-term dependence features in the time domain;

[0027] Use feature map convolution to construct a spatial-domain graph and extract the correlation features between neighborhoods;

[0028] Design an association feature extraction module based on the attention mechanism to dynamically adjust the weights between output feature representations;

[0029] Combine the temporal dependence features and spatial features to construct a regional energy portrait model.

[0030] Preferably, the construction of the regional energy portrait system driven by federated learning specifically includes:

[0031] Train a local LSTM prediction model on the edge computing node;

[0032] Use the homomorphic encryption technology to encrypt the gradients of each node;

[0033] Aggregate the encrypted gradients on the central server to generate a global wind and light output prediction network;

[0034] Distribute the updated global model parameters to each edge node.

[0035] Preferably, the implementation of the multi-objective dynamic game algorithm specifically includes:

[0036] Convert the energy supply and demand into a multi-objective optimization problem of minimizing cost and maximizing sustainability;

[0037] Use the NSGA-III algorithm to generate the Pareto optimal solution set;

[0038] Combine the reinforcement learning technology to construct an elastic scheduling policy library suitable for different scenarios;

[0039] When detecting a sudden drop in photovoltaic output or a sharp increase in the load of charging piles, automatically switch to the optimal compensation plan.

[0040] Preferably, the construction of the digital twin simulation platform specifically includes:

[0041] Create a regional microgrid panoramic digital twin system including GIS maps, meteorological satellite cloud maps, building energy consumption time series maps, and traffic flow heat maps;

[0042] Adopt a conditional generative adversarial network to generate simulated extreme meteorological scenarios based on historical data;

[0043] Simulate equipment failures and system anomalies in a virtual environment to train the fault tolerance ability of the scheduling model;

[0044] Utilize digital twin technology to combine real-time data with simulation analysis results to optimize the scheduling strategy.

[0045] Preferably, the implementation of the two-stage incremental optimization architecture specifically includes:

[0046] Generate a Pareto optimal solution set based on the NSGA-III algorithm in the cloud and construct a set of prediction distribution curves;

[0047] Adopt a lightweight online learning algorithm at the edge to update the prediction curve set in real time;

[0048] Use a causal inference model to identify the correlation characteristics between meteorological mutations and equipment aging;

[0049] Dynamically adjust the scheduling strategy based on the updated prediction curve set and correlation characteristics.

[0050] Preferably, it further includes an equipment health assessment step:

[0051] Obtain multi-dimensional data such as the vibration spectrum, temperature curve, and charge-discharge efficiency of the equipment;

[0052] Based on the obtained multi-dimensional data, calculate the membership degree and weight of the key parameters of different equipment;

[0053] Generate an equipment health decision matrix and construct a multi-objective dynamic game framework based on probability evolution;

[0054] According to the game framework, dynamically adjust the scheduling priority weight of the energy storage system.

[0055] Preferably, it further includes an elastic resource coordination step:

[0056] Construct an energy demand model under different scenarios and generate a historical scheduling knowledge graph;

[0057] Based on deep Q-network technology, develop a backup control strategy for the microgrid in island operation;

[0058] Utilize the historical scheduling knowledge graph to quickly generate an emergency scheduling plan;

[0059] Continuously optimize the emergency scheduling strategy through an online learning method.

[0060] According to the method described in claim 1, it further includes a multi-level energy storage management step:

[0061] Construct an energy storage management model with a multi - layer structure according to the load characteristics and the characteristics of energy storage devices;

[0062] Utilize real - time communication and control algorithms to coordinate the charging and discharging behaviors of energy storage devices at different levels;

[0063] Based on the evaluation results of device health, dynamically adjust the scheduling strategy of the energy storage system;

[0064] Optimize the economy and reliability of the energy storage system according to energy supply - demand prediction and electricity price information.

[0065] The method of the present invention has significant technical advantages and beneficial effects. First of all, this method realizes the unified management and data collection of heterogeneous Internet of Things devices, effectively solves the problem of data islands, and lays a solid foundation for global optimization. Secondly, through the spatio - temporal joint attention mechanism and the regional energy portrait system driven by federated learning, the accuracy of load prediction and renewable energy power generation prediction is greatly improved, providing a reliable basis for precise scheduling.

[0066] In terms of scheduling optimization, the multi - objective dynamic game algorithm of the present invention can simultaneously consider multiple objectives such as economy, environmental protection and system stability, and find the optimal balance point. This not only improves the energy utilization efficiency, but also significantly increases the consumption rate of renewable energy, making important contributions to the sustainable development of the city.

[0067] Especially worth mentioning is that the two - level incremental optimization architecture and the elastic scheduling strategy library of the present invention greatly improve the response speed and adaptive ability of the system. In the face of load mutations or equipment failures, the system can respond within seconds, quickly adjust the strategy, and maintain system stability. This fast response ability is particularly crucial in extreme weather and emergencies, which can effectively reduce system risks and improve the reliability of urban energy supply.

[0068] In addition, the digital twin simulation platform of the present invention provides a powerful tool for system optimization and fault prevention. By simulating various extreme situations and fault scenarios, the system can formulate countermeasures in advance, greatly shortening the fault recovery time and enhancing the resilience of the entire energy system.

[0069] Generally speaking, the method of the present invention realizes the comprehensive improvement of the smart city distributed energy system in terms of efficiency, economy, reliability and environmental protection through the organic combination of a number of innovative technologies. This not only provides a powerful energy management tool for urban managers, but also brings a more stable, economic and clean energy supply to residents, promoting the sustainable development of smart cities. With the continuous optimization and improvement of technology, the method of the present invention is expected to play a more important role in the future construction of smart cities, making positive contributions to addressing global challenges such as climate change and energy transformation... Brief Description of the Drawings

[0070] Figure 1 It is the logic block diagram of the system of the present invention and its core modules.

[0071] Figure 2 It is the diagram of the data acquisition module of the present invention.

[0072] Figure 3 It is the diagram of the multi-modal data fusion module of the present invention.

[0073] Figure 4 It is the diagram of the risk assessment module of the present invention. Specific implementation manners

[0074] Please refer to the appendix Figures 1-4 , the present invention relates to a distributed energy intelligent allocation method based on Internet of Things big data in a smart city, and specifically belongs to the technical field of smart energy management. The method aims to solve problems such as data islands, low prediction accuracy, and rigid scheduling strategies in traditional distributed energy management systems, and realizes the efficient, flexible, and reliable operation of the smart city energy system by innovatively integrating a number of advanced technologies.

[0075] The method of the present invention includes the following steps:

[0076] First, in the acquisition step, this method obtains the real-time operation data of photovoltaic arrays, energy storage power stations, and charging piles through the multi-protocol adaptive parsing engine 1. Preferably, this engine supports multiple communication protocols such as Modbus, IEC 61850, and MQTT, and can adapt to the diverse device communication requirements in a smart city. At the same time, this method also obtains meteorological satellite cloud images, traffic flow heat maps, and building energy consumption historical data to lay a foundation for subsequent analysis and prediction.

[0077] Next, in the processing step, this method first constructs a dynamic weight allocation model based on the obtained real-time operation data, meteorological satellite cloud images, traffic flow heat maps, and building energy consumption historical data by using the spatio-temporal joint attention mechanism 2. The innovation of this mechanism lies in its ability to capture features in both the time and space dimensions simultaneously, improving the prediction accuracy of the model. For example, when considering photovoltaic power generation prediction, not only the time trend of historical power generation data will be considered, but also the spatial distribution characteristics of meteorological cloud images will be combined.

[0078] According to the constructed dynamic weight allocation model, this method generates a global wind and light output prediction network through the federated learning-driven regional energy portrait system 3. This federated learning-based method can protect data privacy while making full use of distributed data resources, improving the accuracy and generalization ability of prediction. For example, photovoltaic power stations in different regions can train models locally and then only share model parameters without sharing the original data, thus achieving collaborative learning while protecting business secrets.

[0079] Based on the generated global wind and solar power output prediction network, this method uses a multi-objective dynamic game algorithm 4 to construct an elastic scheduling strategy library. This algorithm takes into account multiple objectives such as economy, environmental protection, and system stability, and can adapt to the scheduling requirements in different scenarios. For example, during peak electricity price periods, the algorithm is more inclined to discharge from the energy storage system; while during off-peak electricity consumption periods, it will give priority to charging or consuming renewable energy.

[0080] To further improve the robustness of the system, this method simulates extreme scenarios on the digital twin simulation platform 5 to optimize the scheduling model. This step can help the system anticipate possible extreme weather or equipment failure situations in advance and improve the system's emergency response ability. For example, by simulating the photovoltaic power generation fluctuations under typhoon weather, the system can formulate corresponding backup plans in advance.

[0081] Finally, based on the optimized scheduling model, this method generates a real-time scheduling plan through a two-level incremental optimization architecture 6. This architecture divides the optimization process into two levels: the cloud and the edge. It not only ensures the effect of global optimization but also can quickly respond to local changes. For example, the cloud can generate macroscopic scheduling strategies based on long-term predictions, while the edge can make fine-tuning according to real-time data to achieve refined control.

[0082] In the output step, this method outputs real-time scheduling plans for the intelligent allocation of the smart city's distributed energy system. These scheduling plans can be directly applied to the real-time control of distributed energy facilities such as photovoltaic power generation systems, energy storage devices, and electric vehicle charging stations, realizing the efficient utilization and intelligent management of energy.

[0083] Furthermore, the construction of the multi-protocol adaptive parsing engine 1 of the present invention specifically includes the following steps: First, a protocol intelligent recognition model based on meta-learning is adopted to support multiple communication protocols such as Modbus, IEC 61850, and MQTT. The advantage of this method is that it can quickly adapt to new protocol types and improve the scalability of the system. Second, according to the device type, data collection is divided into collected data and requested data. For collected data, device data is actively obtained through the corresponding protocol; for requested data, the system responds to the device's data request through the corresponding protocol. This classification method can improve the efficiency and flexibility of data collection.

[0084] For example, for a photovoltaic inverter using the Modbus protocol, the system can regularly and actively read parameters such as its output power, voltage, and current; while for a smart meter using the MQTT protocol, the system can subscribe to relevant topics to receive electricity consumption data in real time. In this way, this method can achieve unified management and data collection of devices of different types and different protocols.

[0085] In one embodiment of the present invention, the construction of the spatio-temporal joint attention mechanism 2 specifically includes the following steps: First, use a long short-term memory network (LSTM) to construct a time-domain graph and extract short-term dependence features in the time domain. The advantage of the LSTM network is that it can effectively capture long-term dependence relationships and is particularly suitable for processing time series data. Second, use feature map convolution to construct a spatial-domain graph and extract correlation features between neighborhoods. This step can make full use of spatial information, such as the correlation of energy consumption patterns between different regions.

[0086] Next, design an association feature extraction module based on the attention mechanism to dynamically adjust the weights between output feature representations. The introduction of the attention mechanism enables the model to adaptively focus on the most relevant features and improve the prediction accuracy. Finally, combine the time series dependence features and spatial features to construct a regional energy portrait model. This spatio-temporal joint method can comprehensively capture the dynamic characteristics of the energy system and provide a reliable basis for subsequent prediction and scheduling.

[0087] For example, when predicting the electricity load of a certain region, the model will simultaneously consider the time trend of the region's historical electricity consumption data (captured by LSTM) and the electricity consumption conditions of surrounding regions (captured by feature map convolution). If it is found that the electricity consumption at a certain time point is extremely high, the attention mechanism will automatically increase the attention to the relevant features at that time point, thereby improving the prediction accuracy.

[0088] In the method of the present invention, the construction of the federated learning-driven regional energy portrait system 3 specifically includes the following steps: First, train a local LSTM prediction model on edge computing nodes. This distributed training method can make full use of local computing resources and reduce the burden on the central server. Second, use homomorphic encryption technology to encrypt the gradients of each node. The use of homomorphic encryption ensures data security during the model aggregation process and prevents the leakage of sensitive information.

[0089] Then, aggregate the encrypted gradients on the central server to generate a global wind and solar power output prediction network. This step realizes knowledge sharing between different nodes and improves the overall performance of the model. Finally, distribute the updated global model parameters to each edge node. In this way, each node can maintain data locality and privacy while benefiting from the global model.

[0090] For example, when predicting the photovoltaic power generation of different regions in a city, each regional photovoltaic power station can train a local model based on its own historical data. The gradient information of these models is encrypted and sent to the central server, and the server aggregates this information to generate a global model. Finally, each region can obtain a prediction model that takes into account both local features and global information, greatly improving the prediction accuracy and generalization ability.

[0091] Through the above method, the present invention realizes the efficient and intelligent allocation of distributed energy in the smart city, significantly improves the energy utilization efficiency, reduces the operation cost, and at the same time enhances the reliability and flexibility of the system. The innovation and practicality of this method make it have important application value in the construction and sustainable development of the smart city.

[0092] In a preferred embodiment of the present invention, the implementation of the multi-objective dynamic game algorithm 4 specifically includes the following steps: First, convert the energy supply and demand into a multi-objective optimization problem of minimizing cost and maximizing sustainability. This conversion enables the system to seek a balance between economic benefits and environmental benefits and meet the sustainable development requirements of the smart city.

[0093] Secondly, this method uses the NSGA-III algorithm to generate a Pareto optimal solution set. The NSGA-III algorithm is an efficient multi-objective optimization algorithm, especially suitable for dealing with complex optimization problems with three or more objectives. In the present invention, this algorithm can consider multiple objectives such as cost, sustainability, and system stability at the same time, generate a series of non-dominated solutions, and provide diverse choices for decision-making.

[0094] Preferably, the present invention combines reinforcement learning technology to construct a flexible scheduling strategy library suitable for different scenarios. The introduction of reinforcement learning enables the system to continuously optimize the decision-making strategy through continuous interaction with the environment and improve the adaptive ability of the system. For example, the system can learn the best scheduling strategies under different weather conditions, load levels, and electricity prices and store these strategies in the strategy library for quick invocation.

[0095] A remarkable feature of the present invention is that when a sudden drop in photovoltaic output or a sharp increase in charging pile load is detected, the system can automatically switch to the optimal compensation plan. This rapid response mechanism greatly improves the stability and reliability of the system. For example, if sudden cloud cover causes a sharp drop in photovoltaic output, the system can quickly invoke the corresponding plan in the strategy library, such as increasing energy storage discharge or starting a backup power source, to maintain the balance of the power grid.

[0096] In the construction of the digital twin simulation platform 5, the present invention adopts the following innovative methods: First, create a regional microgrid panoramic digital twin system including GIS maps, meteorological satellite cloud maps, building energy consumption time series maps, and traffic flow heat maps. This comprehensive digital representation provides a rich data basis for subsequent simulation and optimization.

[0097] Secondly, this method uses a conditional generative adversarial network (CGAN) to generate simulated extreme meteorological scenarios based on historical data. The use of CGAN enables the system to generate more realistic and diverse extreme scenarios, such as rare heavy rain, heatwaves, or cold snaps. These generated scenarios are crucial for testing and optimizing the extreme performance of the system.

[0098] In a virtual environment, the present invention simulates equipment failures and system anomalies to train the fault tolerance ability of the scheduling model. This method allows the system to fully test and optimize response strategies under various abnormal conditions without affecting actual operation. For example, it is possible to simulate a sudden increase in load caused by the simultaneous charging of a large number of electric vehicles, or a major transmission line failure, etc., to evaluate the system's response ability.

[0099] Finally, this method uses digital twin technology to combine real-time data with simulation analysis results to optimize the scheduling strategy. This method of combining the real and virtual can, while ensuring real-time performance, make full use of historical data and model predictions to achieve more intelligent and forward-looking scheduling decisions.

[0100] In the implementation of the two-level incremental optimization architecture 6 of the present invention, the following innovative methods are adopted: First, a Pareto optimal solution set is generated based on the NSGA-III algorithm in the cloud to construct a set of prediction distribution curves. This method can consider multiple objectives at the global level, generate a series of possible optimal solutions, and provide a basis for subsequent refined scheduling.

[0101] Secondly, a lightweight online learning algorithm is adopted at the edge side to update the prediction curve set in real time. This distributed learning method can quickly adapt to local changes, such as a sudden change in load in a certain area or a fluctuation in the output of renewable energy, improving the system's response speed and flexibility.

[0102] A key innovation of the present invention is to use a causal inference model to identify the correlation characteristics between meteorological mutations and equipment aging. This method can deeply understand the causal relationship behind system behavior, rather than just the surface correlation. For example, the system can identify the pattern of performance degradation of specific equipment under certain meteorological conditions, and thus take preventive measures in advance.

[0103] Finally, based on the updated prediction curve set and correlation characteristics, this method dynamically adjusts the scheduling strategy. This adaptive scheduling method can, while ensuring global optimality, quickly respond to local changes and achieve refined management of the distributed energy system.

[0104] In the equipment health assessment step of the present invention, first, multi-dimensional data such as the vibration spectrum, temperature curve, and charge-discharge efficiency of the equipment are obtained. These multi-source heterogeneous data provide a solid foundation for comprehensively evaluating the health status of the equipment. For example, for energy storage batteries, it is possible to simultaneously monitor the internal temperature change, the attenuation of charge-discharge efficiency, and the change in electrochemical impedance spectrum to comprehensively evaluate its health status.

[0105] Next, based on the obtained multi-dimensional data, this method calculates the membership degrees and weights of key parameters of different devices. This step adopts fuzzy logic theory, which can effectively handle the uncertainty and ambiguity of data. For example, the membership function of battery capacity attenuation can be defined to map continuous capacity values to discrete states such as "healthy", "sub-healthy", and "faulty".

[0106] Then, the present invention generates a device health decision matrix and constructs a multi-objective dynamic game framework based on probability evolution. This method combines device health assessment with system operation decision-making, and can optimize system performance while considering the device state. For example, for a energy storage device with a poor health state, the system may reduce its frequency of participating in peak shaving and frequency modulation to extend its service life.

[0107] Finally, according to the game framework, this method dynamically adjusts the scheduling priority weights of the energy storage system. This dynamic adjustment mechanism ensures the reliability and economy of system operation. For example, when the health state of a certain energy storage unit deteriorates, the system will automatically reduce its usage frequency and increase the usage proportion of other units with better health states, so as to balance system performance and device life.

[0108] In the elastic resource coordination step of the present invention, first, an energy demand model under different scenarios is constructed to generate a historical scheduling knowledge graph. This knowledge graph-based method can effectively capture and represent the operation laws of complex energy systems, providing a strong knowledge basis for subsequent intelligent decision-making. For example, the knowledge graph can include the impacts of different weather conditions, load levels, and energy prices on system operation, as well as the scheduling strategies for successfully coping with various emergencies in history.

[0109] Secondly, based on the deep Q-network technology, this method develops a backup control strategy for the microgrid in island operation. The deep Q-network is an advanced form of reinforcement learning, which is particularly suitable for dealing with complex decision-making problems with large state spaces and action spaces. In the highly dynamic and uncertain scenario of microgrid island operation, the deep Q-network can gradually optimize the control strategy through continuous trial and error and learning, improving the system's adaptability and robustness.

[0110] Preferably, the present invention uses the historical scheduling knowledge graph to quickly generate an emergency scheduling plan. This method combines the advantages of rule-based expert systems and data-driven machine learning methods, and can make decisions quickly in case of emergencies. For example, when it is detected that a large-scale power outage may occur in a certain area, the system can quickly query the treatment plans for similar historical events and make appropriate adjustments in combination with the current situation to minimize the impact of the power outage.

[0111] Finally, through an online learning method, this approach continuously optimizes the emergency dispatch strategy. This continuous learning mechanism ensures that the system can continuously learn from new experiences and adapt to the changing environment and requirements. For example, the system can record the handling process and effects of each emergency event and use this new data to update and optimize the knowledge graph and decision-making model, so as to make better decisions in future similar situations.

[0112] In the multi-level energy storage management step of the present invention, first, according to the load characteristics and energy storage device characteristics, an energy storage management model with a multi-layer structure is constructed. This hierarchical management method can effectively handle different types and scales of energy storage devices and achieve refined management from the system level to the device level. For example, the energy storage system can be divided into multiple levels such as large power station-level energy storage, community-level energy storage, and household-level energy storage, and each level has its specific management strategy and optimization goal.

[0113] Secondly, this method uses real-time communication and control algorithms to coordinate the charging and discharging behaviors of energy storage devices at different levels. This coordination mechanism ensures that the entire energy storage system can operate efficiently as a whole while also retaining the autonomy and flexibility of each level. For example, when the power grid frequency fluctuates, the system can first mobilize the battery energy storage system with a fast response speed for rapid frequency regulation, and then gradually start large pumped-storage power stations for continuous power balance.

[0114] Based on the evaluation results of device health, the present invention dynamically adjusts the dispatch strategy of the energy storage system. This method combines device management with system operation optimization, which can extend the service life of devices and reduce maintenance costs while ensuring system performance. For example, for battery packs in a relatively poor health state, the system will reduce their deep charge and discharge times and use battery packs in a good health state more for daily peak shaving and frequency regulation operations.

[0115] Finally, this method optimizes the economy and reliability of the energy storage system according to energy supply and demand forecasts and electricity price information. This forward-looking optimization strategy can make full use of the electricity price difference for arbitrage and also provide support for the consumption of renewable energy. For example, when it is predicted that there will be a large amount of wind power output the next day, the system can arrange the energy storage device to charge during the low electricity price period at night in advance to free up capacity for the consumption of wind power the next day.

[0116] Through the above steps, the method of the present invention realizes the all-round and multi-level intelligent management of the smart city distributed energy system. From data collection, prediction analysis, strategy optimization to real-time control, a complete closed-loop system is constructed. This method not only improves energy utilization efficiency, reduces operating costs, but also enhances the reliability and flexibility of the system, providing strong technical support for the sustainable development of the smart city.

[0117] In practical applications, the method of the present invention can be flexibly adjusted and optimized according to the characteristics and requirements of specific cities. For example, in cities with rich solar energy resources, the optimization of photovoltaic power generation prediction and scheduling can be strengthened; while in cities with a high penetration rate of electric vehicles, more attention can be paid to the optimization of charging load management and vehicle-grid interaction strategies. This flexibility and scalability make the method of the present invention have wide applicability and practical value.

[0118] In order to verify the superiority of the "Intelligent Distributed Energy Allocation Method Based on Internet of Things Big Data in Smart Cities" of the present invention, a set of simulation experiments were designed, and a typical smart city scenario was selected for simulation. This scenario includes multiple distributed energy facilities, such as photovoltaic power stations, wind turbines, electric vehicle charging stations, and large-scale energy storage systems.

[0119] The simulation conditions are set as follows:

[0120] 1. Time span: 30 consecutive days;

[0121] 2. Weather conditions: including various weather types such as sunny, cloudy, and rainy;

[0122] 3. Load characteristics: Simulate the typical electricity consumption patterns of residential areas, commercial areas, and industrial areas;

[0123] 4. Energy prices: Consider two mechanisms of peak-valley electricity prices and real-time electricity prices;

[0124] 5. Emergency events: Simulate emergency situations such as extreme weather and equipment failures;

[0125] The following three schemes were compared:

[0126] Example 1: The intelligent allocation method of the present invention;

[0127] Comparative Example 1: Traditional rule-based scheduling method;

[0128] Comparative Example 2: Simple machine learning prediction + fixed scheduling strategy;

[0129] The main test indicators include: energy utilization efficiency, cost savings rate, renewable energy consumption rate, load prediction accuracy, system response time, and fault recovery time. The test methods and standards for these indicators are as follows:

[0130] 1. Energy utilization efficiency: Calculate the ratio of total energy output to input, and the higher the standard, the better.

[0131] 2. Cost savings rate: The percentage reduction in operating costs compared to the baseline case, and the higher the standard, the better.

[0132] 3. Renewable energy consumption rate: The proportion of the actual use of renewable energy in the total power generation, and the higher the standard, the better.

[0133] 4. Load forecasting accuracy: The mean absolute percentage error (MAPE) between the predicted value and the actual value, and the lower the standard, the better.

[0134] 5. System response time: The average time from detecting a load change to adjusting the power generation output, and the shorter the standard, the better.

[0135] 6. Fault recovery time: The average time from the occurrence of a fault to the system returning to normal operation, and the shorter the standard, the better.

[0136] The simulation test results are shown in the following table:

[0137] Index Example 1 Comparative Example 1 Comparative Example 2 Energy utilization efficiency 92.5 78.3 85.1 Cost savings rate 18.7 5.2 10.4 Renewable energy consumption rate 95.3 76.8 84.2 Load forecasting accuracy (MAPE) 3.2 12.5 7.8 System response time 2.3 15.7 8.5 Fault recovery time 5.6 28.3 17.2

[0138] It can be seen from the test results that the method of the present invention is significantly superior to the comparative scheme in all indicators. The specific analysis is as follows:

[0139] In terms of energy utilization efficiency, the method of the present invention reaches 92.5%, which is 14.2 percentage points higher than the traditional method and 7.4 percentage points higher than the simple machine learning method. This is mainly due to the multi-objective dynamic game algorithm and the two-stage incremental optimization architecture adopted by the present invention, which can better coordinate various energy resources and reduce losses.

[0140] In terms of cost savings rate, the method of the present invention achieves a savings of 18.7%, which is much higher than the other two methods. This reflects the advantage of this method in economic dispatch, especially in the ability to utilize the peak-valley electricity price difference and optimize the operation of the energy storage system.

[0141] The renewable energy consumption rate reaches 95.3%, which is at least 11 percentage points higher than other methods. This shows that the method of the present invention can better predict the output of renewable energy and maximize the use of renewable energy through flexible dispatch strategies and the coordination of the energy storage system.

[0142] In terms of load forecasting accuracy, the method of the present invention controls the MAPE at 3.2%, which is significantly better than other methods. This is due to the spatio-temporal joint attention mechanism and the federated learning-driven regional energy portrait system adopted by this method, which can better capture the complex patterns of load changes.

[0143] The system response time is only 2.3 seconds, which is nearly 7 times faster than the traditional method and nearly 4 times faster than the simple machine learning method. This reflects the advantage of the elastic dispatch strategy library and the two-stage incremental optimization architecture of the present invention in fast response.

[0144] In terms of the fault recovery time, the method of the present invention only takes 5.6 minutes, which is much faster than other methods. This is mainly due to the digital twin simulation platform and the elastic resource coordination mechanism of this method, which can quickly identify faults and generate effective recovery strategies.

[0145] In summary, the method of the present invention shows obvious advantages in many aspects such as energy efficiency, economy, utilization of renewable energy, prediction accuracy, response speed and reliability. These advantages stem from the integration of a number of innovative technologies in the present invention, such as a multi-protocol adaptive parsing engine, a spatio-temporal joint attention mechanism, federated learning, a multi-objective dynamic game algorithm, etc. The organic combination of these technologies enables this method to grasp the complex dynamics of the smart city energy system more comprehensively and accurately, so as to make better scheduling decisions.

[0146] It is particularly worth noting that the present invention performs excellently in dealing with emergencies and extreme situations, which is of great significance for improving the resilience and reliability of the smart city energy system. In the future, when facing the challenges brought by climate change and energy transformation, this high adaptability and strong self-repair ability will play a key role.

[0147] The best embodiment of the present invention further optimizes the simulation accuracy of the digital twin simulation platform and enhances the intelligence level of the elastic resource coordination mechanism on the basis of the original scheme. This makes the system perform even better under extreme weather conditions, the fault recovery time is further shortened to 4.2 minutes, and the renewable energy consumption rate is increased to 97.1%. This improvement fully reflects the scalability and continuous optimization potential of the method of the present invention, and points out the direction for the future development of the smart city energy system.

[0148] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A distributed energy intelligent allocation method based on IoT big data in smart cities, characterized in that: include: The acquisition steps include: Acquire real-time operating data of photovoltaic arrays, energy storage power stations and charging piles through a multi-protocol adaptive parsing engine; Obtain meteorological satellite cloud images, traffic flow heat maps and historical data on building energy consumption; Processing steps include: Based on the real-time operation data, meteorological satellite cloud images, traffic flow heat maps and building energy consumption historical data, a dynamic weight allocation model is constructed using a spatiotemporal joint attention mechanism; According to the dynamic weight allocation model, a global wind and solar output prediction network is generated through a regional energy profiling system driven by federated learning; Based on the global wind and solar power output prediction network, a flexible scheduling strategy library is constructed using a multi-objective dynamic game algorithm; According to the elastic scheduling strategy library, extreme scenarios are simulated on the digital twin simulation platform to optimize the scheduling model; Based on the optimized scheduling model, a real-time scheduling solution is generated through a two-level incremental optimization architecture; Output steps include: The real-time dispatching scheme is output for intelligent deployment of distributed energy systems in smart cities.

2. The method according to claim 1, characterized in that The construction of the multi-protocol adaptive parsing engine specifically includes: Adopts a meta-learning-based protocol intelligent identification model to support multiple communication protocols such as Modbus, IEC 61850 and MQTT; According to the device type, data collection is divided into collection data and request data; For collected data, actively obtain device data through corresponding protocols; For requested data, respond to the device's data request through the corresponding protocol.

3. The method according to claim 1, characterized in that The construction of the spatiotemporal joint attention mechanism specifically includes: Use long short-term memory networks to construct time domain graphs and extract short-term dependency features in the time domain; Use feature graph convolution to construct a spatial domain graph and extract correlation features between neighborhoods; Design a correlation feature extraction module based on the attention mechanism to dynamically adjust the weights between output feature representations; Combining time-dependent features with spatial features, a regional energy portrait model is constructed.

4. The method according to claim 1, characterized in that: The construction of the regional energy profiling system driven by federated learning specifically includes: Train a local LSTM prediction model on the edge computing node; Homomorphic encryption technology is used to encrypt the gradient of each node; Aggregate encrypted gradients on a central server to generate a global wind and solar output prediction network; Distribute the updated global model parameters to each edge node.

5. The method according to claim 1, characterized in that The implementation of the multi-objective dynamic game algorithm specifically includes: Transforming energy supply and demand into a multi-objective optimization problem of minimizing costs and maximizing sustainability; The NSGA-III algorithm is used to generate the Pareto optimal solution set; Combined with reinforcement learning technology, a flexible scheduling strategy library that adapts to different scenarios is built; When a sudden drop in photovoltaic output or a surge in charging pile load is detected, it automatically switches to the optimal compensation solution.

6. The method according to claim 1, characterized in that The construction of the digital twin simulation platform specifically includes: Create a panoramic digital twin system of regional microgrids that includes GIS maps, meteorological satellite cloud maps, building energy consumption time series diagrams, and traffic flow heat maps; Using conditional generative adversarial networks to generate simulated extreme weather scenarios based on historical data; Simulate equipment failures and system anomalies in a virtual environment to train the fault tolerance of the scheduling model; Using digital twin technology, real-time data is combined with simulation analysis results to optimize scheduling strategies.

7. The method according to claim 1, characterized in that The implementation of the two-level incremental optimization architecture specifically includes: Generate a Pareto optimal solution set based on the NSGA-III algorithm in the cloud and construct a set of prediction distribution curves; A lightweight online learning algorithm is used at the edge to update the prediction curve set in real time; Use causal inference models to identify the correlation characteristics between meteorological mutations and equipment aging; The scheduling strategy is dynamically adjusted based on the updated set of prediction curves and associated features.

8. The method according to claim 1, characterized in that It also includes the equipment health assessment steps: Obtain multi-dimensional data such as vibration spectrum, temperature curve, and charge and discharge efficiency of the equipment; Based on the acquired multi-dimensional data, calculate the membership and weight of key parameters of different equipment; Generate equipment health decision matrix and build a multi-objective dynamic game framework based on probability evolution; According to the game framework, the dispatch priority weight of the energy storage system is dynamically adjusted.

9. The method according to claim 1, characterized in that: It also includes elastic resource coordination steps: Build energy demand models under different scenarios and generate historical scheduling knowledge graphs; Based on deep Q network technology, develop a backup control strategy for microgrid island operation; Using the historical dispatch knowledge graph, an emergency dispatch plan is quickly generated; Continuously optimize emergency dispatch strategies through online learning methods.

10. The method according to claim 1, characterized in that It also includes multi-level energy storage management steps: According to the load characteristics and energy storage equipment characteristics, a multi-layer energy storage management model is constructed; Use real-time communication and control algorithms to coordinate the charging and discharging behaviors of energy storage devices at different levels; Dynamically adjust the dispatch strategy of the energy storage system based on the equipment health assessment results; Optimize the economy and reliability of energy storage systems based on energy supply and demand forecasts and electricity price information.

Images