Power data driving-based energy consumption diagnosis method

Abstract

The invention discloses an energy consumption diagnosis method based on electric power data driving, and aims to solve the defects in the prior art through an energy consumption diagnosis method which performs classification modeling according to electricity-energy association intensity and is deeply coordinated with power grid operation. According to the method, the energy consumption characteristics of the manufacturing industry are accurately depicted through electricity-energy association intensity classification modeling in combination with the power grid operation state. The method includes establishing a device-level and multi-source energy coupling model, generating an energy efficiency rating report, and optimizing an energy saving strategy through anomaly detection. A real-time power grid load rate is introduced, a production state is analyzed by adopting dynamic clustering, and multi-level anomaly detection is realized in combination with LOF and MNNDAF algorithms. The optimization strategy is coded in a CIM / E format, equipment parameters are adjusted through a PI controller, cooperation of power grid peak regulation and demand response is ensured, a GAN enhancement model is further adopted for training, and the diagnosis precision and robustness are improved.

Description

Technical Field

[0001] This invention relates to energy consumption data processing technology, and more particularly to an energy consumption diagnosis method based on power data. Background Technology

[0002] In the field of energy management, existing technologies primarily rely on three core components: data acquisition, modeling and analysis, and optimized control. Traditional methods typically involve deploying smart meters, sensor networks, and energy management systems (EMS) to collect real-time energy consumption data from equipment. This data is then combined with statistical analysis tools or machine learning algorithms to build energy consumption models and generate energy-saving strategies. For example, time-series analysis-based energy prediction methods can identify periodic energy consumption patterns in equipment, while clustering-based energy consumption status classification techniques can categorize energy consumption characteristics in production processes. Furthermore, some advanced technologies have incorporated cloud computing and big data platforms to achieve integrated processing and remote diagnostics of multi-source data. These methods, to a certain extent, improve the precision of energy management and provide technical support for energy conservation and emission reduction.

[0003] However, existing technologies still have significant shortcomings. Chinese invention patent CN105512445B discloses a cloud-based platform for simulating energy consumption and diagnosing energy-saving problems in typical energy-consuming systems. However, this invention fails to adequately consider the differences in the strength of electricity-energy correlations, resulting in limitations in model accuracy and applicability. Specifically, this invention treats all energy-consuming links uniformly, failing to establish suitable modeling methods for links with strong and weak electricity-energy correlations, thus failing to accurately reflect the differences between equipment-level energy consumption characteristics and system-level energy coupling relationships. Furthermore, this invention lacks deep integration with the grid's operating status, failing to incorporate key factors such as regional grid load factor and substation operating parameters into the energy consumption diagnosis process, thus limiting its application value in grid peak shaving and demand response. Summary of the Invention

[0004] This invention aims to address the shortcomings of existing technologies by employing an energy consumption diagnosis method that classifies and models energy consumption based on the strength of the power-energy correlation and deeply coordinates with grid operation. By establishing differentiated power-energy correlation models, this invention achieves accurate characterization and efficient optimization of energy consumption characteristics, while enhancing its coordination with grid operation, thus providing technical support for energy conservation, emission reduction, and the safe and stable operation of the power grid.

[0005] This invention proposes an energy consumption diagnosis method for power grids. The method includes: establishing a device-level power-energy mapping function for strongly correlated links and constructing a multi-source energy coupling model for weakly correlated links based on the classification results of power-energy correlation strength; generating an energy efficiency rating report based on the production status of clustering and triggering energy-saving strategy optimization through anomaly detection; feeding back the strategy execution effect to the power grid dispatching system, updating equipment operating parameters and restarting the diagnosis process.

[0006] Preferably, this method performs data cleaning and feature extraction based on the substation bus voltage fluctuation rate, power plant output curve, and regional power grid frequency deviation to determine the strength of the power-energy correlation. Through multi-dimensional correlation analysis of power grid operating parameters and energy consumption data, the accuracy of power-energy correlation strength determination is improved, reducing the misjudgment rate to below 5%, thus providing a reliable basis for subsequent differentiated modeling.

[0007] Preferably, the strongly correlated step includes: establishing a nonlinear power-energy mapping function for individual devices; and performing joint parameter calibration of the device group using the Levenberg-Marquardt algorithm. This solves the parameter calibration problem caused by the nonlinear coupling of the device group, improves the model convergence speed by 40%, controls the joint calibration error to within 3%, and achieves accurate energy consumption prediction for the strongly correlated step.

[0008] Preferably, the weakly correlated step includes: extracting the daily periodic component of steam / gas energy using wavelet transform; and establishing a grey relational model between ambient temperature and compressor energy consumption. By separating the energy cycle characteristics from external environmental interference, the energy consumption modeling error of the weakly correlated step is reduced by 18%, and the prediction accuracy of summer air conditioning system energy consumption is improved to 92%.

[0009] Preferably, this method constructs a grid collaborative evaluation index based on energy use intensity and the energy consumption ratio during peak and off-peak electricity price periods. This achieves coordinated optimization of energy consumption costs and grid load, reducing the proportion of energy consumption during peak periods by 12% and reducing overall electricity costs by 8.5%.

[0010] Preferably, this method employs dynamic density clustering based on grid load factor to determine production status; and uses a cascaded verification method for transient impact anomalies and collective mode anomalies for anomaly detection. Dynamic density clustering accurately identifies different production statuses, improving the targeting of energy efficiency assessments and increasing the matching degree of optimization strategies by 25%; cascaded verification reduces the false alarm rate to below 3%.

[0011] Preferably, this method employs an improved OPTICS algorithm to introduce a dynamic reachability distance threshold to classify production states; it uses the LOF algorithm to identify transient impact anomalies and sets a dynamic density threshold; and it uses the MNNDAF algorithm to detect collective anomalies and correlate them with substation fault recording data. The improved OPTICS algorithm increases the efficiency of production state classification by 30%, and the accuracy of correlation between anomaly detection and power grid faults exceeds 90%, avoiding disorderly shutdowns caused by misjudgments.

[0012] Preferably, this method encodes the optimization strategy into a grid dispatch instruction in CIM / E format, and adjusts the reactive power compensation of the equipment through a PI controller. This achieves seamless integration between the energy consumption optimization strategy and grid dispatch, reducing the reactive power compensation response time to 200ms and increasing the bus voltage qualification rate to 99.2%.

[0013] Preferably, when the regional power grid load rate is detected to exceed 90%, the method prioritizes the implementation of peak-shaving load optimization strategies; the energy-saving strategy generation module is coupled with the electricity market pricing system interface. This allows for rapid reduction of peak loads by 15%-20% in power grid overload scenarios, while simultaneously maximizing demand-side response benefits through the electricity market interface.

[0014] Preferably, this method establishes an energy consumption diagnosis reliability evaluation model. When the reliability is below 0.8, it triggers GAN-based sample augmentation training. Through dynamic reliability assessment and data augmentation, the model's diagnostic accuracy is improved by 12% in scenarios with insufficient samples, avoiding strategy failure due to data bias.

[0015] The present invention also has the following beneficial effects:

[0016] 1. Achieve energy efficiency improvement: Through electricity-energy correlation modeling and dynamic cluster analysis, accurately identify high energy-consuming links and generate optimization strategies, significantly reduce the comprehensive power consumption per ton of steel, and improve the power factor to over 0.93.

[0017] 2. Achieve grid coordination, optimize strategies and deeply coordinate with the grid dispatch system to achieve load transfer of 1.2MW, effectively supporting the peak-shaving needs of the regional power grid.

[0018] 3. Implement anomaly early warning: the dual-factor anomaly detection mechanism can locate the insulation deterioration problem of the 10kV feeder in the substation 72 hours in advance, avoid the loss of continuous casting billet defects, and reduce the scrap rate by 1.3%.

[0019] 4. The model has self-learning capabilities and a dynamic adjustment mechanism for model parameters based on feedback data, ensuring continuous improvement in diagnostic accuracy and keeping energy consumption prediction error within 2% for a long period of time. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0021] Figure 2 This is a flowchart of the quantitative description and evaluation index of energy consumption characteristics of the present invention;

[0022] Figure 3 This is a flowchart of the energy consumption diagnosis modeling method of the present invention. Detailed Implementation

[0023] Example 1

[0024] according to Figure 1 As shown, this invention proposes an energy consumption diagnosis method based on power data, which includes:

[0025] First, data such as the voltage fluctuation rate of the 220kV substation bus, regional power grid frequency regulation commands, and peak-valley electricity price schedules were collected from the power grid side. Simultaneously, real-time data on the harmonic distortion rate of the three-phase current of the electric arc furnace, the cooling water temperature of the continuous casting machine, and the reactive power of the rolling mill were obtained from the manufacturing equipment side. These data were then combined with environmental parameters such as workshop ambient temperature and air compressor station exhaust pressure for time synchronization and data repair. In the data preprocessing stage, a sliding window interpolation method was used to repair missing values, and a wavelet denoising algorithm was used to eliminate measurement noise, ensuring that the data quality met the modeling requirements.

[0026] After data preprocessing, the strength of the power-energy correlation was further quantitatively determined. Specifically, based on time-series data from the grid and equipment sides within a sliding window, the Pearson correlation coefficient between power parameters (current, voltage, power factor) and energy consumption indicators was calculated, with a threshold ρ ≥ 0.7 set as the condition for strong correlation. For high-energy-consuming equipment such as electric arc furnaces, the correlation coefficient between current harmonic distortion rate and electricity consumption per ton of steel reached 0.91, significantly exceeding the threshold, and was classified as a strongly correlated link; while the correlation coefficient between air compressor station exhaust pressure and grid frequency was only 0.35, classifying it as a weakly correlated system. Simultaneously, the time-series characteristics of regional grid frequency regulation commands were introduced, and the alignment between them and equipment energy consumption curves was analyzed using the Dynamic Time Warping (DTW) algorithm, further verifying the reliability of the correlation strength classification results. This determination process provides a crucial basis for subsequent differentiated modeling.

[0027] In the power-energy correlation modeling stage, models were established for strongly correlated equipment and weakly correlated systems based on the classification results of power-energy correlation strength. For electric arc furnaces, a strongly correlated equipment, the 3rd, 5th, and 7th harmonic features of its current harmonic components were extracted, and a nonlinear mapping model between harmonic components and power consumption per ton of steel was established. The Levenberg-Marquardt algorithm was used to jointly calibrate the parameters of the equipment group, and the calibration error was controlled within 2.1%, significantly improving the prediction accuracy of the model. For air compressor stations, a weakly correlated system, wavelet transform was used to separate the grid frequency regulation response component in air pressure fluctuations, and its correlation characteristics with the regional grid frequency were analyzed. At the same time, a grey relational model of ambient temperature and air compressor energy consumption was established, and the calculated grey relational degree was 0.83, indicating that ambient temperature has a significant impact on air compressor energy consumption.

[0028] On the other hand, for weakly correlated systems, such as air compressor stations, this invention employs wavelet transform technology to separate different frequency components in air pressure fluctuations, particularly the power grid frequency regulation response component, to analyze its correlation with the regional power grid frequency. Furthermore, this invention constructs a grey relational model of ambient temperature and air compressor energy consumption, finding that ambient temperature has a significant impact on air compressor energy consumption, with a grey relational degree reaching 0.83. This analysis not only reveals the potential impact of environmental factors on equipment energy consumption but also provides an important basis for subsequent energy efficiency optimization strategies.

[0029] Through these differentiated modeling methods, this invention can more accurately characterize the energy consumption characteristics of different energy-consuming stages, laying the foundation for subsequent energy consumption optimization and management. Combining the OPTICS algorithm with grid load factor correction, this invention performs dynamic clustering analysis on production states, introducing the real-time grid load factor as an adjustment factor, and classifying production states into three typical modes: high load during normal periods, peak-valley production surges, and emergency load reduction. This classification not only helps this invention better understand the energy consumption characteristics in the production process but also enables refined energy consumption management and optimization.

[0030] Based on the above model, the OPTICS algorithm with grid load factor correction is used to perform dynamic clustering analysis on the production status. By introducing the real-time grid load factor as a reachability distance threshold correction factor, the production status is divided into three typical modes: high load during normal periods, impact production during valley periods, and emergency load reduction. Among them, the high load during normal periods has a 92% matching degree with the historical template and can be directly rated by calling the historical energy efficiency baseline; the impact production during valley periods is the first occurrence, so a new status template needs to be created and included in the knowledge base. In the anomaly detection stage, the LOF algorithm (an algorithm for detecting abnormal data points, which identifies anomalies by comparing the local density of data points with the density of other points in its neighborhood) is used to identify the transient impact anomaly of the rolling mill at 12:05, whose anomaly factor exceeds the standard by 2.8 times. Further, the MNNDAF algorithm (a method for detecting collective anomalies of data, which identifies the phenomenon of collective deviation from the normal pattern by analyzing the relative distance between multiple data points) is used to detect a collective anomaly pattern in the continuous casting cooling system. Correlation analysis shows that this anomaly is highly correlated with the voltage sag event at the substation at 23:17.

[0031] In the classification of production states, the identification and management of high-load periods during flat periods are particularly important. Through comparative analysis of historical data, this invention can identify the energy efficiency baseline under these conditions, providing enterprises with a clear energy efficiency reference. Optimization and improvement based on this baseline can significantly enhance overall energy efficiency. For production states experiencing peak demand during off-peak periods, since this is the first time such a condition has occurred, a new state template needs to be established and incorporated into the knowledge base. This process involves not only in-depth analysis of current anomalies but also combining historical data and industry standards to ensure the accuracy and applicability of the new template. The emergency load reduction mode is primarily used to cope with sudden changes in grid load. By adjusting equipment operating parameters in real time, it reduces energy consumption and ensures the continuity and stability of production.

[0032] Anomaly detection is a key component of this method. By combining the LOF and MNNDAF algorithms, abnormal phenomena in the production process can be effectively identified and analyzed. The LOF algorithm can quickly identify transient impact anomalies in the rolling mill at specific time points, while the MNNDAF algorithm further reveals collective anomaly patterns in the continuous casting cooling system. This multi-layered anomaly detection mechanism not only improves the accuracy of anomaly identification but also provides reliable data support for subsequent anomaly handling and optimization.

[0033] During the grid collaborative optimization phase, energy-saving strategies were generated based on cluster analysis and anomaly detection results. For the electric arc furnace, a high-energy-consuming piece of equipment, its preheating period was adjusted from 10:00 to 23:30 to avoid peak grid load periods. Simultaneously, the reactive power compensation of the rolling mill was dynamically adjusted via a PI controller, improving the power factor from 0.89 to 0.93. The optimization strategy was encoded into grid dispatch instructions in CIM / E format and sent to the enterprise energy management system (EMS) for execution. After strategy execution, feedback data was collected in real time and the effects were evaluated. The results showed a 6.7% reduction in comprehensive electricity consumption per ton of steel, from 582 kWh / t to 543 kWh / t. Simultaneously, the peak-hour load transfer in the plant area reached 1.2 MW, accounting for 14% of the total plant load, effectively alleviating the regional grid peak-shaving pressure.

[0034] When the regional power grid load rate is detected to exceed 90%, a peak-shaving load optimization strategy is prioritized. In practice, the system monitors the regional power grid load rate in real time, collecting data using sensors installed in substations and on major electrical equipment. When the load rate exceeds the 90% threshold, the system automatically triggers the peak-shaving load optimization strategy. This strategy includes adjusting the operating times of high-energy-consuming equipment, postponing non-urgent production tasks, or dynamically adjusting equipment operating parameters to reduce instantaneous load. For example, the use of electric arc furnaces can be postponed to periods of lower load, or instantaneous power demand can be reduced by lowering the heating power of the electric arc furnaces. Simultaneously, the energy-saving strategy generation module is coupled with the electricity market pricing system interface to obtain real-time electricity market price signals and dynamically adjust the electricity consumption strategy based on price changes, thereby maximizing economic benefits.

[0035] An energy consumption diagnosis reliability evaluation model is established. When the reliability falls below 0.8, sample augmentation training based on GAN (Generative Adversarial Network) (a deep learning model that generates samples similar to the real data distribution through an adversarial process, often used for data augmentation and model training) is triggered. In implementation, the energy consumption diagnosis system first calculates the reliability of the energy consumption diagnosis by comparing historical and real-time data. The reliability evaluation model considers multiple factors, including data completeness, model prediction accuracy, and the deviation between actual and predicted energy consumption. When the system detects a reliability below 0.8, it indicates that the current model may face insufficient data or model bias. At this time, the system automatically triggers GAN-based sample augmentation training. GAN enriches the training set by generating new data samples, simulating different production and grid load conditions, thereby improving the robustness and accuracy of the model. After sample augmentation training, the system recalibrates the model parameters to ensure that the accuracy and reliability of energy consumption diagnosis remain at a high level.

[0036] The implementation of energy consumption management and optimization strategies has been further strengthened. In addition to using GANs for sample augmentation training to improve the robustness of the model, the system also introduces a continuous monitoring and feedback mechanism to ensure the dynamic adjustment of energy consumption optimization strategies. By integrating a real-time monitoring module into the enterprise energy management system (EMS), the system can continuously acquire data on equipment operating status, energy consumption indicators, and grid load.

[0037] The real-time monitoring module not only provides energy consumption data for the current production process but also identifies potential energy efficiency optimization opportunities. For example, the system can detect the idle status of certain equipment during off-peak hours and suggest scheduling maintenance or repairs for these devices during these periods to avoid unnecessary energy consumption. Furthermore, the system records and analyzes the effectiveness of each optimization strategy in detail, generating energy efficiency reports to help management assess the effectiveness of energy-saving measures.

[0038] In long-term energy optimization planning, predictive analytics is used to forecast future production demand and grid load changes. By combining historical data and market trends, the system can develop optimization strategies in advance to ensure optimal energy management under different production conditions. Predictive analytics can also help companies make more strategic decisions when participating in electricity market bidding, maximizing economic benefits by flexibly adjusting production plans and electricity consumption strategies.

[0039] Through this multi-level, multi-dimensional approach to energy consumption diagnosis and optimization, enterprises can not only achieve immediate energy savings but also improve their overall energy efficiency in the long term. This method provides enterprises with a comprehensive energy management framework, supporting them in maintaining a competitive edge in a highly competitive market environment. Furthermore, this method offers valuable insights and lessons for energy management in other industries, demonstrating broad application potential.

[0040] Example 2

[0041] according to Figure 2 , Figure 3 As shown, this invention proposes an energy consumption diagnosis method based on power data, which includes:

[0042] First, based on the correlation characteristics of "electricity-energy," for energy-consuming links with strong "electricity-energy" correlation, a functional expression is established to represent the "electricity-energy" mapping relationship of individual devices, and a data fitting method is used to establish the "electricity-energy" mapping relationship of device groups. For energy-consuming links with weak "electricity-energy" correlation, the periodicity and volatility of other types of energy and their correlation with the external environment are analyzed, thus forming a quantitative description method for typical energy consumption characteristics. For different energy-consuming links, the applicability of energy consumption characteristic evaluation indicators such as energy use intensity (EUI) (used to measure the energy consumption level per unit area or unit output, usually used to evaluate the energy efficiency of buildings or production processes), energy consumption ratio (ECR), energy cost ratio (ECR), carbon emission intensity, energy efficiency ratio (COP) (referring to the energy efficiency level of equipment or system under specific conditions, usually expressed as the ratio of output energy to input energy), and load factor is analyzed, and a comprehensive evaluation index for typical energy consumption characteristics is established.

[0043] Secondly, this study investigates energy consumption distribution models for typical multi-level scenarios. It employs an energy consumption distribution analysis method based on spatial density clustering and deep learning. Spatial density clustering is used to perform cluster analysis on daily energy consumption time series, and production states are categorized based on the clustering results. Attribute analysis is conducted on energy consumption curves under various production states, including correlation, fluctuation, impact, periodicity, seasonality, and interruptibility. Based on the correlation between energy consumption links and total load, they can be categorized into peak-season, peak-avoidance, and stable types. The annual variation characteristics (seasonality) and daily periodicity of equipment are analyzed. For fluctuation, wavelet transform can be used for analysis. For impact attributes, the LOF (Local Outlier Factor) algorithm can be used for identification. For interruptibility, it can be determined based on production characteristics and load importance. A typical scenario set is constructed based on these attributes to analyze the spatiotemporal distribution of energy consumption.

[0044] Typical energy consumption diagnosis mainly focuses on two aspects. First, it assesses energy consumption under normal operating conditions. Based on cluster analysis results, it calculates energy consumption evaluation indicators and identifies extreme points of these indicators under various production conditions. This then determines the input-output indicators of energy-consuming links / equipment. Data Envelopment Analysis (DEA) (a non-parametric method for evaluating the relative efficiency of multiple decision-making units such as equipment or processes, which constructs an "efficiency frontier" to compare the efficiency of each decision-making unit) is used to determine efficiency differences between similar energy-consuming links / equipment, thereby determining the user's energy efficiency level. Second, it establishes typical energy efficiency anomaly detection methods. The Local Outlier Factor (LOF) method is used to identify abnormal data mutations, and the Mean Nearest Neighbor Distance Anomaly Factor (MNNDAF) algorithm is used to detect collective anomalies in the data, thereby identifying typical energy consumption anomalies.

[0045] During the energy consumption rating process, the system identifies extreme points of energy consumption assessment indicators under different production conditions through in-depth analysis of clustering results. These extreme points reflect the energy efficiency and potential energy-saving potential under different conditions. For example, under peak production conditions, the system may find that the coefficient of performance (COP) of some equipment reaches its maximum value, while under peak-shaving conditions, energy intensity (EUI) may be significantly reduced. By identifying these extreme points, enterprises can better understand under which production conditions optimizing energy consumption management can bring the greatest benefits.

[0046] Data Envelopment Analysis (DEA) is used to assess efficiency differences between similar energy-consuming processes or equipment. DEA is a non-parametric method that compares the relative efficiency of various devices or energy-consuming processes by constructing an "efficiency frontier." This analysis not only helps companies identify inefficient equipment or processes but also provides a basis for developing improvement strategies. For example, through DEA ​​analysis, companies may discover that certain equipment performs poorly under specific production conditions, thus deciding to upgrade equipment or improve processes to enhance overall energy efficiency.

[0047] In energy efficiency anomaly detection, the LOF (Local Outlier Factor) algorithm is used to identify anomalous changes in data. This algorithm identifies outliers that deviate from normal patterns by comparing local data densities. For example, if the energy consumption of a piece of equipment suddenly spikes without a corresponding increase in production demand, the equipment may be malfunctioning or being operated improperly. This information can help companies intervene before the problem escalates.

[0048] The Mean Nearest Neighbor Distance Anomaly Factor (MNNDAF) algorithm is used to detect collective anomalies in data. This method identifies phenomena that collectively deviate from normal patterns by analyzing the relative distances between multiple data points. For example, when multiple devices simultaneously exhibit abnormal energy consumption concentrated within a specific time period, it may indicate a collective impact of power grid fluctuations or changes in the external environment on the production system. Identifying these collective anomalies helps enterprises adjust production plans and energy consumption strategies at a macro level.

[0049] This dual diagnostic approach allows companies to gain a more comprehensive understanding of energy consumption characteristics and anomalies in their production processes, thereby improving the precision of energy efficiency management. This not only helps reduce production costs and carbon emissions but also provides companies with greater flexibility and a competitive advantage when participating in the electricity market. In practical applications, companies can adjust production plans, optimize equipment operating parameters, and develop long-term energy-saving improvement strategies based on the diagnostic results to achieve sustainable development goals.

Claims

1. An energy consumption diagnosis method based on power data, characterized in that, The method includes: Based on the classification results of the power-energy correlation strength, a device-level power-energy mapping function is established for strongly correlated links, and a multi-source energy coupling model is constructed for weakly correlated links. Energy efficiency rating reports are generated based on production status categorized by clustering, and energy-saving strategy optimization is triggered through anomaly detection. The strategy execution effect is fed back to the power grid dispatching system to update equipment operating parameters and restart the diagnostic process.

2. The energy consumption diagnosis method based on power data as described in claim 1, characterized in that, The method describes a process for data cleaning and feature extraction based on the substation bus voltage fluctuation rate, power plant output curve, and regional power grid frequency deviation, in order to determine the strength of the power-energy correlation.

3. The energy consumption diagnosis method based on power data as described in claim 1, characterized in that, The strongly correlated steps include: establishing a nonlinear power-energy mapping function for individual devices; and performing joint parameter calibration of the device group using the Levenberg-Marquardt algorithm.

4. The energy consumption diagnosis method based on power data as described in claim 1, characterized in that, The weakly correlated links include: using wavelet transform to extract the daily periodic components of steam / gas energy; and establishing a grey relational model between ambient temperature and compressor energy consumption.

5. A power data-driven energy consumption diagnosis method according to claim 1, 3, or 4, characterized in that, The method constructs a grid collaborative evaluation index based on energy use intensity and peak-valley electricity price period energy consumption ratio.

6. The energy consumption diagnosis method based on power data as described in claim 1, characterized in that, The method employs dynamic density clustering based on grid load rate to determine production status; and uses a cascaded verification method of transient impact anomalies and collective pattern anomalies for anomaly detection.

7. The energy consumption diagnosis method based on power data as described in claim 1 or 6, characterized in that, The method employs an improved OPTICS algorithm to introduce a dynamic reachability threshold to classify production states; it uses the LOF algorithm to identify transient impact anomalies, sets a dynamic density threshold, and uses the MNNDAF algorithm to detect collective anomalies, linking them to substation fault recording data.

8. The energy consumption diagnosis method based on power data as described in claim 1, characterized in that, The method encodes the optimization strategy into a power grid dispatch instruction in CIM / E format, and adjusts the reactive power compensation of the equipment through a PI controller.

9. The energy consumption diagnosis method based on power data as described in claim 1, characterized in that, When the method detects that the regional power grid load rate exceeds 90%, it prioritizes the implementation of peak-shaving load optimization strategies; the energy-saving strategy generation module is coupled with the interface of the electricity market quotation system.

10. A power data-driven energy consumption diagnosis method according to claim 1 or 9, characterized in that, The method establishes an energy consumption diagnosis reliability evaluation model. When the reliability is lower than 0.8, it triggers GAN-based sample augmentation training.

Images