Power system management method combining big data and intelligent algorithm

Abstract

The invention discloses a power system management method combining big data and an intelligent algorithm, and the method comprises the steps: 1, covering the distribution-transmission-distribution-utilization-storage deployment of an IoT intelligent sensor, employing an edge computing gateway to process high-frequency data nearby, and then uploading the high-frequency data to a power big data center; step 2, carrying out quality optimization on the uploaded data, wherein the quality optimization comprises noise cleaning, missing value completion and data standardization; 3, constructing a short-term load prediction model, a new energy output prediction model and an equipment fault early warning model, wherein each model is matched with scenarized input, an adaptive algorithm and dynamic optimization; 4, an algorithm and a service system are connected, and prediction-decision-execution automation is achieved; 5, closed-loop monitoring and algorithm iteration are carried out; the method has the advantages of being firm in data foundation, accurate in long-term prediction of the algorithm model, timely and effective in scheduling strategy and intelligent and advanced in management method, and safe, stable, efficient and economical operation of the whole chain of transmission, transmission, distribution, utilization and storage of the novel power system can be achieved.

Description

Technical Field

[0001] This invention belongs to the field of power system management technology, specifically relating to a power system management method that combines big data and intelligent algorithms. Background Technology

[0002] The core of power system management is to achieve safe, stable, efficient and economical operation of the entire power generation, transmission, distribution and consumption chain through the coordination of four dimensions: technology, economy, dispatch and security.

[0003] However, the current power system management still has the following problems: (1) Data silos are serious. Data on power generation, transmission, distribution and user side are scattered and contain noise and missing values, which makes it impossible to support advanced algorithm modeling. (2) The short-term load forecasting error exceeds 15%, the new energy output forecasting error exceeds 20%, and most equipment failures still rely on manual inspection, resulting in a high rate of unplanned power outages. (3) The algorithm is disconnected from the business, resulting in problems such as delayed scheduling response (more than 1 hour), low operation and maintenance efficiency (average 4 hours), and accurate prediction of new energy consumption but poor utilization. (4) The algorithm model is prone to failure due to system changes, the management effect lacks quantitative monitoring, and there is no basis for continuous optimization; To address the aforementioned issues, it is essential to develop a power system management method that combines big data and intelligent algorithms. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a power system management method that combines big data and intelligent algorithms, which has a solid data foundation, accurate long-term prediction of algorithm models, timely and effective scheduling strategies, and intelligent and advanced management methods. This method can realize the safe, stable, efficient and economical operation of the entire chain of power generation, transmission, distribution, consumption and storage in the new power system.

[0005] The objective of this invention is achieved as follows: a power system management method combining big data and intelligent algorithms, comprising the following steps: Step 1, full-chain data collection: covering five links of generation, transmission, distribution, consumption and storage, deploying IoT smart sensors to collect full-chain data in real time, and using edge computing gateways to process high-frequency data locally, and then uploading the processed data to the power big data center via 5G and / or fiber optics; Step 2, Data Quality Optimization: The uploaded data is optimized in the power big data center, including noise cleaning, missing value completion, and data standardization; Step 3, Multi-scenario intelligent algorithm modeling: Based on the optimized data, short-term load forecasting model, new energy output forecasting model and equipment fault early warning model are constructed respectively. Each model is matched with scenario-based input, adapted algorithm and dynamic optimization. Step 4, Intelligent Decision-Making and Business Implementation: Build a power intelligent management platform to connect algorithms with scheduling, operation and maintenance, and new energy consumption to achieve automation of prediction, decision-making, and execution; Step 5, Closed-loop monitoring and algorithm iteration: Quantify the monitoring system and conduct dynamic algorithm iteration and management scheme optimization regularly or irregularly.

[0006] Preferably, in step 1, the power generation side is equipped with wind power / photovoltaic output sensors, thermal power boiler temperature / pressure sensors, and hydropower water level sensors; the transmission side is equipped with line current / voltage transformers, tower tilt sensors, and insulator pollution detectors; the distribution side is equipped with distribution area load collectors and transformer oil temperature / level sensors; the user side is equipped with industrial / commercial / residential smart meters; and the energy storage side is equipped with SOC sensors and charge / discharge power detectors.

[0007] Preferably, step 2 includes the following steps: Noise cleaning: Outliers are removed using the 3σ principle, and the moving average method is used to smooth the wind power output data that fluctuates frequently. Missing value completion: Short-term missing values ​​are completed using linear interpolation, while long-term missing values ​​are completed using the average of historical data within the same time period plus the daily load trend correction method. Data standardization: Min-Max normalization is used to convert data of different dimensions into values ​​in the [0,1] interval.

[0008] Preferably, in step 3, the short-term load forecasting model is set as follows: Algorithm selection: LSTM Long Short-Term Memory Network + Attention Mechanism. LSTM captures temporal patterns, and the attention mechanism strengthens the weights of key factors. Feature inputs: historical load data for the past 30 days, meteorological data for the next 24 hours, user attributes, and holiday tags; Training and optimization: The Adam optimizer is used to minimize the mean squared error (MSE) between the predicted and actual values. The model parameters are updated daily at 2:00 AM using the actual data from the previous day.

[0009] Preferably, in step 3, the new energy output prediction model is set as follows: Algorithm selection: GRU gated recurrent unit + numerical weather prediction (NWP) fusion. GRU simplifies the LSTM structure and improves computational efficiency, while NWP data serves as the core input. Scenario adaptation: Photovoltaic forecasting: Focus on solar irradiance + cloud cover rate, and add a photovoltaic panel temperature correction item; Wind power forecasting: Focus on wind speed + wind direction, and add historical wind turbine failure data; Deviation correction: Compare the actual output with the predicted value every 12 hours, calculate the deviation rate, and dynamically adjust the model coefficients.

[0010] Preferably, in step 3, the equipment fault early warning model is set as follows: Algorithm selection: CNN convolutional neural network + isolated forest algorithm. CNN extracts local features of device status data, and isolated forest identifies abnormal patterns that deviate from the normal range. Data input: Transmission line current / voltage fluctuation value, transformer oil temperature / partial discharge quantity, switch cabinet humidity / temperature, and the input data is updated every 5 minutes; Early warning classification: Three threshold levels are set: Yellow: deviation 10-20%, push notification; Orange: 20-30%, dispatch for inspection; Red: >30%, emergency repair. When a red warning is issued, the fault type is automatically matched.

[0011] Preferably, step 4 includes the following steps: Intelligent dispatch automation: The platform integrates load forecasting and renewable energy output forecasting to automatically generate unit combination schemes. It adopts a dispatch scheme that utilizes full wind / solar power generation, supplemented by hydropower for any shortfall, thermal power as peak-shaving power, and energy storage for discharging during peak hours and charging during off-peak hours. When the actual load deviates from the forecast by more than 5%, the platform automatically sends output adjustment commands to thermal power units and emergency discharge commands to energy storage systems to shorten response time. It also connects to the UHV dispatch system and automatically applies for external transmission channels when local renewable energy output is excessive. Precise Operation and Maintenance Closed Loop: After receiving a red alert, the platform generates a maintenance work order within 1 minute; the maintenance work order is automatically pushed to the responsible person through the operation and maintenance APP, and the progress is updated in real time, with the progress marked as dispatched - under maintenance - restored. If the work order has not been dispatched for more than 1 hour, a reminder is automatically pushed to the superior; after the fault is handled, the platform automatically records the fault cause, handling measures and recovery time, and incorporates them into the equipment fault knowledge base for optimizing subsequent early warning models; Optimization of renewable energy consumption: When the predicted output of wind power / solar power exceeds the grid's capacity, the system automatically sends a power reduction instruction to the power station and simultaneously instructs the surrounding energy storage power stations to increase their charging power; it pushes time-sharing incentives to industrial and commercial users and guides them to turn on high-energy-consuming equipment during this period by connecting to the enterprise's production system via API; it also calculates the regional renewable energy consumption rate in real time and automatically adjusts the incentive level if it is lower than 90%.

[0012] Preferably, step 5 includes the following steps: Quantitative monitoring system: Core indicators are set on the platform homepage and updated every 15 minutes; a "Daily Operational Deviation Report" is generated, and abnormal items are marked and the causes are automatically analyzed; Dynamic algorithm iteration: On the 1st of each month, the three major models are retrained using the full-chain data of the previous month to optimize parameters; when a certain indicator fails to meet the standard for 3 consecutive days, an emergency iteration is immediately started to supplement new features; the mature fault early warning model in region A is transferred and applied after adjusting the parameters based on the equipment characteristics in region B, thus shortening the model training cycle. Management solution optimization: Compare the effectiveness of intelligent management and traditional management every quarter and output "Optimization Suggestions"; if the demand response participation rate is found to be less than 30%, adjust the incentive policy and update the platform's user incentive rule library.

[0013] Preferably, the core indicators are load forecast accuracy, renewable energy curtailment rate, unplanned power outage duration, and unit power supply cost.

[0014] Due to the adoption of the above technical solution, the beneficial effects of the present invention are: (1) This invention can effectively cover and collect data across the entire chain by deploying IoT smart sensors in the five links of generation, transmission, distribution, use and storage, thereby achieving unified collection of data across the entire chain, breaking down data silos, and effectively improving data quality through quality optimization processing such as noise reduction and completion of collected data, providing a solid data foundation for subsequent analysis; (2) By constructing three major models, this invention greatly reduces the short-term load forecasting error, greatly reduces the wind power / photovoltaic output forecasting error, effectively improves the early warning rate of equipment failure, thereby providing a precise load benchmark for dispatching, reducing the amount of wind and solar curtailment, transforming post-event maintenance into pre-event warning, and reducing the unplanned power outage rate. (3) By building a smart power management platform, this invention promotes the full implementation of smart dispatching, precise operation and maintenance and new energy consumption, which significantly improves power supply stability, effectively improves equipment maintenance efficiency, and effectively shortens the average repair time, thereby reducing industrial economic losses caused by power outages and significantly increasing the utilization rate of new energy power generation in the region, thus helping to achieve the dual carbon target. (4) This invention enables the algorithm model to maintain stable prediction accuracy over a long period of time through real-time effect monitoring, dynamic algorithm iteration and management scheme optimization, avoids model failure caused by system changes, realizes a closed loop of management effect, algorithm optimization and effect improvement, and maintains the advanced nature of intelligent management methods in the long term. In summary, this invention has the advantages of a solid data foundation, accurate long-term prediction of the algorithm model, timely and effective scheduling strategy, and intelligent and advanced management method, which can realize the safe, stable, efficient and economical operation of the entire chain of power generation, transmission, distribution, consumption and storage in the new power system. Detailed Implementation

[0015] The technical solution of the present invention will be further described in detail below through embodiments.

[0016] This invention provides a power system management method that combines big data and intelligent algorithms, comprising the following steps: The first step is full-chain data collection: covering five links from generation to transmission, distribution, consumption and storage, deploying IoT smart sensors to collect data from the entire chain in real time, and using edge computing gateways (such as Huawei OceanConnect) to process high-frequency data (1 time / second) locally, filter redundant data (such as steady-state current data without fluctuations), and upload the processed data to the power big data center via 5G / fiber. The big data center adopts a Hadoop+HBase distributed architecture, supports PB-level data storage, and meets the data backtracking needs of more than 3 years.

[0017] On the power generation side, wind power / photovoltaic output sensors, thermal power boiler temperature / pressure sensors, and hydropower water level sensors are deployed. On the transmission side, line current / voltage transformers, tower tilt sensors, and insulator pollution detectors are deployed. On the distribution side, substation load collectors and transformer oil temperature / level sensors are deployed. On the user side, smart meters for industrial, commercial, and residential use (sampling every 15 minutes) are deployed. On the energy storage side, SOC (state of charge) sensors and charge / discharge power detectors are deployed.

[0018] The second step is data quality optimization: The uploaded data is optimized in the power big data center, including noise cleaning, missing value completion, and data standardization.

[0019] Noise cleaning: The 3σ principle is used to remove outliers (such as a sudden increase in line current to 10 times the normal value, which is judged as a sensor failure), and the moving average method is used to smooth the wind power output data with frequent fluctuations.

[0020] Missing value completion: Short-term missing values ​​are completed using linear interpolation, while long-term missing values ​​are completed using the average of historical data for the same period plus the daily load trend correction method. For example, if the morning peak on a weekday is missing, the average of the morning peak over the past 7 weekdays plus the daily temperature influence coefficient are used as a reference.

[0021] Data standardization: Through Min-Max normalization, data of different dimensions such as voltage (kV), power (MW), and temperature (°C) are converted into values ​​in the range of [0,1] to avoid interference from dimensions during algorithm training.

[0022] The third step is multi-scenario intelligent algorithm modeling: Based on the optimized data, short-term load forecasting model (1-24 hours), new energy output forecasting model (1-72 hours), and equipment fault early warning model (real-time monitoring) are constructed respectively. Each model is matched with scenario-based input, adapted algorithm, and dynamically optimized.

[0023] The specific settings for the short-term load forecasting model are as follows: Algorithm selection: LSTM Long Short-Term Memory Network + Attention Mechanism. LSTM captures temporal patterns (such as weekday / weekend load differences), and the attention mechanism strengthens the weight of key factors (such as temperature, holidays, and industrial production plans). Feature inputs: historical load data for the past 30 days, meteorological data for the next 24 hours (temperature, precipitation, wind speed), user attributes (proportion of industrial / commercial / residential users), and holiday tags (e.g., Spring Festival = 1, ordinary day = 0). Training and optimization: The Adam optimizer is used to minimize the mean square error (MSE) between the predicted and actual values. The model parameters are updated daily at 2:00 AM with the actual data from the previous day to ensure adaptation to load change trends, such as the increase in air conditioning load during the summer.

[0024] The specific settings for the new energy output prediction model are as follows: Algorithm selection: GRU gated recurrent unit + numerical weather prediction (NWP) fusion. GRU simplifies the LSTM structure and improves computational efficiency, while NWP data (irradiance and wind speed for the next 3 days) serves as the core input. Scene adaptation: Photovoltaic forecast: Focus on solar irradiance + cloud cover, and add a photovoltaic panel temperature correction factor (because excessive temperature will reduce conversion efficiency). Wind power forecasting: Focus on wind speed and wind direction, and add historical data on wind turbine failures (e.g., if a wind turbine fails 3 times in the past month, reduce its output forecast weight). Deviation correction: Compare the actual output with the predicted value every 12 hours, calculate the deviation rate, and dynamically adjust the model coefficients. If the deviation rate exceeds 15%, the weight of NWP data can be increased.

[0025] The specific settings for the equipment fault early warning model are as follows: Algorithm selection: CNN convolutional neural network + isolated forest algorithm. CNN extracts local features of equipment status data (such as high-frequency abnormal components of transformer vibration signals), and isolated forest identifies abnormal patterns that deviate from the normal range. Data input: Transmission line current / voltage fluctuation value, transformer oil temperature / partial discharge quantity, switch cabinet humidity / temperature, and the input data is updated every 5 minutes; Early warning classification: Set three threshold levels (yellow: deviation 10-20%, push notification; orange: 20-30%, dispatch order for inspection; red: >30%, emergency repair). When a red warning is issued, the fault type is automatically matched (such as transformer insulation aging, line icing).

[0026] The fourth step is intelligent decision-making and business implementation: build an intelligent power management platform to connect algorithms with scheduling, operation and maintenance, and new energy consumption to achieve automation of prediction, decision-making, and execution.

[0027] Intelligent dispatch automation: The platform integrates load forecasting and renewable energy output forecasting, automatically generating unit combination schemes, prioritizing full-capacity wind / solar power generation, supplementing the shortfall with hydropower, and using thermal power as peak-shaving power. Energy storage discharges during peak electricity consumption (18:00-22:00) and charges during off-peak electricity consumption (0:00-6:00). When the actual load deviates from the forecast by more than 5%, the platform automatically sends output adjustment commands to thermal power units (such as increasing by 50MW) and emergency discharge commands to energy storage systems, shortening response time and dynamically adjusting the load. It also connects to the UHV dispatch system, automatically applying for external transmission channels (such as transmitting electricity to the east via the West-to-East Power Transmission Line) when local renewable energy output is excessive.

[0028] Precise Operation and Maintenance Closed Loop: After receiving a red alert, the platform automatically generates a maintenance work order within 1 minute, including the GIS location of the faulty equipment, the fault type, recommended tools (such as an insulation tester), and nearby spare parts inventory information; the maintenance work order is automatically pushed to the responsible person through the operation and maintenance APP, and the progress is updated in real time (departed - under maintenance - restored). If it has not been dispatched for more than 1 hour, a reminder is automatically pushed to the superior; after the fault is handled, the platform automatically records the fault cause, handling measures and recovery time, and incorporates them into the equipment fault knowledge base for optimizing subsequent early warning models.

[0029] Optimization of renewable energy consumption: When the predicted output of wind / solar power exceeds the grid's capacity, the system automatically sends output reduction instructions to the power plants (e.g., reducing solar power from 100MW to 80MW), and simultaneously instructs nearby energy storage power stations to increase their charging power to absorb excess electricity; it also pushes time-of-use incentives to industrial and commercial users (e.g., reducing the electricity price to 0.3 yuan / kWh when renewable energy is in surplus), and guides them to start high-energy-consuming equipment (e.g., steelmaking furnaces in factories) during this period by connecting to the enterprise's production system via API; it also calculates the regional renewable energy consumption rate in real time, and if it is lower than 90%, it automatically adjusts the incentive level (e.g., further reducing the price by 0.05 yuan / kWh).

[0030] The fifth step is closed-loop monitoring and algorithm iteration: quantify the monitoring system and conduct dynamic algorithm iteration and management scheme optimization regularly or irregularly.

[0031] Quantitative monitoring system: The platform sets core indicators (including load forecast accuracy, renewable energy curtailment rate, unplanned power outage duration, unit power supply cost, etc.) on the homepage of the KPI dashboard in real time, and updates them once every 15 minutes; it generates an "Operational Deviation Report" every day and marks abnormal items (such as wind curtailment rate of 6% in a certain area, exceeding the target by 1%), and automatically analyzes the reasons (such as congestion of external transmission channels, insufficient energy storage and charging capacity).

[0032] Dynamic algorithm iteration: On the 1st of each month, the three major models are retrained using the full-chain data of the previous month, and the parameters are optimized (such as adjusting the number of hidden layer nodes of LSTM from 64 to 128); when a certain indicator fails to meet the standard for 3 consecutive days (such as load prediction accuracy <75%), an emergency iteration is immediately launched to supplement new features (such as adding regional large-scale event labels to cope with temporary load surges); the mature fault early warning model in region A is transferred and applied after adjusting the parameters in combination with the equipment characteristics of region B (such as high altitude), thus shortening the model training cycle.

[0033] Management solution optimization: Compare the benefits of intelligent management and traditional management every quarter (e.g., how much the unit power supply cost is reduced, how much carbon emissions are reduced), and output "Optimization Suggestions"; if the demand response participation rate is found to be <30%, adjust the incentive policy (e.g., extend the incentive period, increase subsidies), and update the platform's user incentive rule library.

[0034] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A power system management method combining big data and intelligent algorithms, characterized in that, Includes the following steps: Step 1, full-chain data collection: covering five links of generation, transmission, distribution, consumption and storage, deploying IoT smart sensors to collect full-chain data in real time, and using edge computing gateways to process high-frequency data locally, and then uploading the processed data to the power big data center via 5G and / or fiber optics; Step 2, Data Quality Optimization: The uploaded data is optimized in the power big data center, including noise cleaning, missing value completion, and data standardization; Step 3, Multi-scenario intelligent algorithm modeling: Based on the optimized data, short-term load forecasting model, new energy output forecasting model and equipment fault early warning model are constructed respectively. Each model is matched with scenario-based input, adapted algorithm and dynamic optimization. Step 4, Intelligent Decision-Making and Business Implementation: Build a power intelligent management platform to connect algorithms with scheduling, operation and maintenance, and new energy consumption to achieve automation of prediction, decision-making, and execution; Step 5, Closed-loop monitoring and algorithm iteration: Quantify the monitoring system and conduct dynamic algorithm iteration and management scheme optimization regularly or irregularly.

2. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, In step 1, the power generation side is equipped with wind power / photovoltaic output sensors, thermal power boiler temperature / pressure sensors, and hydropower water level sensors; the transmission side is equipped with line current / voltage transformers, tower tilt sensors, and insulator pollution detectors; the distribution side is equipped with distribution area load collectors and transformer oil temperature / level sensors; the user side is equipped with industrial / commercial / residential smart meters; and the energy storage side is equipped with SOC sensors and charge / discharge power detectors.

3. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, Step 2 includes the following steps: Noise cleaning: Outliers are removed using the 3σ principle, and the moving average method is used to smooth the wind power output data that fluctuates frequently. Missing value completion: Short-term missing values ​​are completed using linear interpolation, while long-term missing values ​​are completed using the average of historical data within the same time period plus the daily load trend correction method. Data standardization: Min-Max normalization is used to convert data of different dimensions into values ​​in the [0,1] interval.

4. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, In step 3, the short-term load forecasting model is set up as follows: Algorithm selection: LSTM Long Short-Term Memory Network + Attention Mechanism. LSTM captures temporal patterns, and the attention mechanism strengthens the weights of key factors. Feature inputs: historical load data for the past 30 days, meteorological data for the next 24 hours, user attributes, and holiday tags; Training and optimization: The Adam optimizer is used to minimize the mean squared error (MSE) between the predicted and actual values. The model parameters are updated daily at 2:00 AM using the actual data from the previous day.

5. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, In step 3, the specific settings for the new energy output prediction model are as follows: Algorithm selection: GRU gated recurrent unit + numerical weather prediction (NWP) fusion. GRU simplifies the LSTM structure and improves computational efficiency, while NWP data serves as the core input. Scenario adaptation: Photovoltaic forecasting: Focus on solar irradiance + cloud cover rate, and add a photovoltaic panel temperature correction item; Wind power forecasting: Focus on wind speed + wind direction, and add historical wind turbine failure data; Deviation correction: Compare the actual output with the predicted value every 12 hours, calculate the deviation rate, and dynamically adjust the model coefficients.

6. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, In step 3, the equipment fault early warning model is set up as follows: Algorithm selection: CNN convolutional neural network + isolated forest algorithm. CNN extracts local features of device status data, and isolated forest identifies abnormal patterns that deviate from the normal range. Data input: Transmission line current / voltage fluctuation value, transformer oil temperature / partial discharge quantity, switch cabinet humidity / temperature, and the input data is updated every 5 minutes; Warning levels: Three thresholds are set, yellow: deviation 10-20%, push notification alert; Orange: 20-30%, dispatch for inspection; Red: >30%, emergency repair, automatic fault type matching when red warning occurs.

7. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, Step 4 includes the following steps: Intelligent dispatch automation: The platform integrates load forecasting and renewable energy output forecasting to automatically generate unit combination schemes. It adopts a dispatch scheme that utilizes full wind / solar power generation, supplemented by hydropower for any shortfall, thermal power as peak-shaving power, and energy storage for discharging during peak hours and charging during off-peak hours. When the actual load deviates from the forecast by more than 5%, the platform automatically sends output adjustment commands to thermal power units and emergency discharge commands to energy storage systems to shorten response time. It also connects to the UHV dispatch system and automatically applies for external transmission channels when local renewable energy output is excessive. Precise Operation and Maintenance Closed Loop: After receiving a red alert, the platform generates a maintenance work order within 1 minute; the maintenance work order is automatically pushed to the responsible person through the operation and maintenance APP, and the progress is updated in real time, with the progress marked as dispatched - under maintenance - restored. If the work order has not been dispatched for more than 1 hour, a reminder is automatically pushed to the superior; after the fault is handled, the platform automatically records the fault cause, handling measures and recovery time, and incorporates them into the equipment fault knowledge base for optimizing subsequent early warning models; Optimization of renewable energy consumption: When the predicted output of wind power / solar power exceeds the grid's capacity, the system automatically sends a power reduction instruction to the power station and simultaneously instructs the surrounding energy storage power stations to increase their charging power; it pushes time-sharing incentives to industrial and commercial users and guides them to turn on high-energy-consuming equipment during this period by connecting to the enterprise's production system via API; it also calculates the regional renewable energy consumption rate in real time and automatically adjusts the incentive level if it is lower than 90%.

8. The power system management method combining big data and intelligent algorithms according to claim 1, characterized in that, Step 5 includes the following steps: Quantitative monitoring system: Core indicators are set on the platform homepage and updated every 15 minutes; a "Daily Operational Deviation Report" is generated, and abnormal items are marked and the causes are automatically analyzed; Algorithm dynamic iteration: On the 1st of each month, the three major models are retrained using the full-chain data of the previous month to optimize parameters; When a certain indicator fails to meet the standard for 3 consecutive days, an emergency iteration is immediately initiated to add new features; the mature fault early warning model in region A is transferred and applied after adjusting the parameters according to the equipment characteristics in region B, thereby shortening the model training cycle. Management solution optimization: Compare the effectiveness of intelligent management and traditional management every quarter and output "Optimization Suggestions"; if the demand response participation rate is found to be less than 30%, adjust the incentive policy and update the platform's user incentive rule library.

9. The power system management method combining big data and intelligent algorithms according to claim 8, characterized in that: The core indicators are load forecast accuracy, renewable energy curtailment rate, unplanned power outage duration, and unit power supply cost.