Equipment energy consumption analysis method based on new energy station

Abstract

The invention discloses an equipment energy consumption analysis method based on a new energy station, and relates to the technical field of energy monitoring, and the method comprises the analysis steps: S1, collecting the energy consumption data and operation condition data of various types of equipment of the new energy station, and sorting out the judgment rules of the energy consumption abnormity of the equipment; s2, constructing an equipment energy consumption characteristic model, training the energy consumption characteristic model locally based on a federated learning framework, and storing energy consumption data, working condition data and an anomaly judgment rule into an analysis database according to equipment types in a classified manner; and S3, the system collects equipment energy consumption and working condition data in real time, calls the energy consumption characteristic model to perform intelligent analysis, and performs phase synchronous analysis based on working conditions and energy consumption. According to the method, dual mechanisms of federated learning model intelligent analysis and working condition and energy consumption phase synchronization verification are combined, energy consumption time sequence characteristics are captured, dynamic linkage logic of the two mechanisms is verified, the anomaly recognition accuracy is greatly improved, misjudgment caused by single dependence on a model is reduced, and manpower and equipment loss cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of energy monitoring technology, specifically to a method for analyzing equipment energy consumption at new energy power plants. Background Technology

[0002] As the global energy structure transitions towards clean and low-carbon energy, the installed capacity of new energy power plants such as wind power, photovoltaics, and energy storage continues to expand, and the types of equipment are becoming increasingly diverse. The precise control of their operating energy consumption directly affects the power generation efficiency, operating costs, and equipment lifespan of the power plants, becoming one of the core demands for the high-quality development of the new energy industry. However, in terms of equipment energy consumption analysis, existing technologies mostly rely on single data collection and lack intelligent analysis capabilities. Power plant personnel need to manually integrate energy consumption data and judge the rationality of energy consumption, which is a cumbersome process and prone to analysis bias due to human factors. It is also difficult to detect abnormal equipment energy consumption in a timely manner, affecting the energy utilization efficiency of the power plants. To address this, we propose a method for equipment energy consumption analysis based on new energy power plants. Summary of the Invention

[0003] To address the aforementioned technical problems, a method for analyzing equipment energy consumption in new energy power plants is provided. This technical solution resolves the problems described above.

[0004] To achieve the above objectives, the technical solution adopted by this invention is: a method for analyzing equipment energy consumption in new energy power stations, the analysis steps of which are as follows: S1. Collect energy consumption data and operating condition data of various equipment in new energy power stations, and sort out the judgment rules for abnormal equipment energy consumption; S2. Construct an equipment energy consumption characteristic model. Based on the federated learning framework, train the energy consumption characteristic model locally and store the energy consumption data, operating condition data and anomaly judgment rules into the analysis database according to equipment type. S3. The system collects equipment energy consumption and operating condition data in real time, calls the energy consumption characteristic model for intelligent analysis, and identifies equipment with abnormal energy consumption based on the phase synchronization analysis of operating conditions and energy consumption. S4. Employs a sliding window and incremental update mode to generate real-time dynamic reports and push optimization and maintenance suggestions.

[0005] Preferably, in step S1, energy consumption data is collected based on power sensors and flow sensors, and operating conditions are collected in real time using temperature, pressure, vibration, and speed sensors; after preprocessing the collected data, it is transmitted to the terminal via wired and wireless means to obtain equipment data.

[0006] Preferably, in step S1, the rules for judging abnormal equipment energy consumption are sorted out by establishing a dynamic energy consumption baseline using statistical methods based on equipment data according to equipment type and different working conditions; the abnormality types are refined, and a hierarchical judgment logic combining static thresholds and dynamic conditions with time dimension is set. The formula for establishing a dynamic energy consumption baseline using statistical methods is as follows: Normal energy consumption range baseline = [μ-kσ, μ+kσ]; Where μ is the average energy consumption under a certain operating condition group, σ is the standard deviation, and k is the coefficient.

[0007] Preferably, step S2 specifically includes: In terms of framework deployment, a horizontal federated architecture is adopted, with local training nodes set up at each site and aggregation nodes set up on the management platform; The model is constructed as a hierarchical architecture, using LSTM to capture energy consumption time-series features, extracting core related features based on the working condition attention mechanism, and outputting feature vectors and anomaly probabilities. Federated aggregation uses the FedAvg weighted average algorithm based on data volume to aggregate global parameters and then distribute them to each node. The convergence model is deployed locally, and the results are output in real time and added to the database to form a data loop.

[0008] Preferably, the specific steps for identifying devices with abnormal energy consumption in step S3 are as follows: The system collects energy consumption and operating condition data in real time at a sampling frequency of 1 minute, through three methods: direct sensor sampling, device communication interface and synchronization with third-party systems, and transmits the data to the local edge node via MQTT encryption. The federated learning energy consumption feature model is invoked, and a dedicated sub-model is matched according to the device type and real-time operating conditions. The core features are extracted and the energy consumption feature vector and initial anomaly probability are output. The threshold is set at 0.8. If the value is greater than the threshold, it is marked as a suspected anomaly. Calculate the correlation of the rates of change and the phase difference between the two, and verify them. When phase decoupling or divergence occurs, it is determined to be an abnormal phase synchronization. Based on the two-level judgment, the final output includes the device ID, abnormal time, type, and confidence level information, and is then pushed to the system.

[0009] Preferably, data transmission uses the MQTT 3.1.1 protocol and is securely transmitted via a TLS 1.3 encrypted channel. A lightweight MQTT Broker is deployed on the local edge node to perform integrity verification after receiving data. Data that fails verification is temporarily stored in a local cache for 24 hours. Data that passes verification is immediately associated with the device ID, collection timestamp, and collection method identifier and stored in the local time-series database of the edge node.

[0010] Preferably, the edge nodes perform secondary adaptation on the preprocessed real-time data according to the unified standard of the federated training phase; the sub-model matching mechanism locks the model category according to the device type; and real-time analysis of operating parameters matches the exclusive sub-model for the corresponding sub-scenario of operating condition. The model is based on LSTM and a working condition attention mechanism, and extracts features in three layers, specifically including: Time-series features are extracted to show the moving average, variance, and peak / valley ratio of energy consumption over the past 10 minutes, thus capturing the time-series patterns of energy consumption. By analyzing the correlation features of operating conditions, we can extract the bias, rate of change, and response delay between energy consumption and real-time operating conditions, and capture the degree of linkage and matching between operating conditions and energy consumption. Baseline deviation characteristics are extracted, including the percentage deviation of the current energy consumption from the dynamic baseline of stage S1 and the duration of continuous deviation, and associated with previous anomaly judgment rules. Output a 128-dimensional fixed-length energy consumption feature vector; The model output layer uses the Sigmoid activation function to map the feature vectors to initial anomaly probabilities in the 0-1 range; The threshold of 0.8 was determined based on historical data from the federated training phase. Through testing with 10,000+ normal samples and 2,000+ labeled abnormal samples, the optimal threshold value was found to be ≥92% for anomaly identification and ≤5% for false positive rate when the threshold was set to 0.8.

[0011] Preferably, the correlation calculation of the rate of change is performed by selecting sampling data from the past 30 minutes, which is divided into 30 sample points. The rate of change of operating conditions and the rate of change of energy consumption are calculated for each sample point. The rate of change of operating conditions is obtained by subtracting the value of the previous sample point from the operating parameters of the current sample point and then dividing by the 60-second sampling interval. Similarly, the rate of change of energy consumption is obtained by dividing the difference between the energy consumption data of the current sample point and the previous sample point by 60 seconds. The phase difference is calculated using the cross-correlation function method. First, a delay step size of 0 to 10 sampling times is set. The sum of the products of the operating condition change rate sequence and the corresponding delayed energy consumption change rate sequence at each step size is calculated. The optimal delay step size that maximizes this sum is found. The optimal step size is multiplied by 60 seconds to obtain the actual phase difference. Positive numbers represent energy consumption lagging operating conditions, and negative numbers represent energy consumption leading operating conditions. The judgment criteria are as follows: the trend is judged based on the correlation of the rate of change of the data in the past 30 minutes. A strong positive correlation is normal, a weak positive correlation needs to be verified, and no correlation or negative correlation is directly judged as phase disconnection or divergence. For samples with weak positive correlation, the phase difference is checked according to the device type, and those exceeding the threshold or with energy consumption ahead are judged as disconnected. An anomaly is identified if any one of the conditions is met.

[0012] Preferably, in step S4, the core parameters of the sliding window are set, including the window duration and the sliding step size, and the window data is stored in groups according to device type and working condition scenario; Synchronize incremental data every unit of minute to remove duplicate data and invalid noise; The incremental data is linked with the historical data within the window to complete the data chain of equipment operating status, anomaly records, and changes in operating conditions, and the statistical results of the indicators are updated synchronously.

[0013] Preferably, the dynamic report in step S4 includes an overview of equipment operation, the number of each type of equipment operating in the window, the percentage of normal and abnormal equipment, and a summary of abnormal details; the report can be viewed in real time on a PC monitoring platform.

[0014] An energy consumption analysis system for equipment in renewable energy power plants includes an energy consumption analysis module, a feature model module, an abnormal equipment module, and an analysis module. The energy consumption analysis module collects energy consumption data and operating condition data of various equipment in new energy power stations, and analyzes the rules for judging abnormal equipment energy consumption. The feature model module constructs a device energy consumption feature model. Based on the federated learning framework, it trains the energy consumption feature model locally and stores energy consumption data, operating condition data and anomaly judgment rules into the analysis database according to device type. The abnormal equipment module collects equipment energy consumption and operating condition data in real time, calls the energy consumption characteristic model for intelligent analysis, and identifies equipment with abnormal energy consumption based on phase synchronization analysis of operating conditions and energy consumption. The analysis module uses a sliding window and incremental update mode to generate real-time dynamic reports and push optimization and maintenance suggestions.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention presents a method for analyzing equipment energy consumption in new energy power plants. It combines intelligent analysis using a federated learning model with a dual mechanism of synchronized verification of operating conditions and energy consumption phases. This captures both the temporal characteristics of energy consumption and verifies the dynamic linkage logic between the two, significantly improving the accuracy of anomaly identification and reducing misjudgments from relying solely on a single model. Furthermore, the real-time acquisition and analysis mode enables rapid response to equipment anomalies, preventing the escalation of faults. It provides highly targeted operation and maintenance decisions, reducing operating costs. Dynamic reports are generated through sliding windows and incremental updates, ensuring real-time effectiveness of information. Ineffective inspections are reduced, lowering manpower and equipment wear and tear costs. Attached Figure Description

[0016] Figure 1 This is a flowchart of the analysis steps in an embodiment of the present invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Reference Figure 1 As shown, a method for analyzing equipment energy consumption in new energy power stations is described, and the analysis steps are as follows: S1. Collect energy consumption data and operating condition data of various equipment in new energy power stations, and sort out the judgment rules for abnormal equipment energy consumption; S2. Construct an equipment energy consumption characteristic model. Based on the federated learning framework, train the energy consumption characteristic model locally and store the energy consumption data, operating condition data and anomaly judgment rules into the analysis database according to equipment type. S3. The system collects equipment energy consumption and operating condition data in real time, calls the energy consumption characteristic model for intelligent analysis, and identifies equipment with abnormal energy consumption based on the phase synchronization analysis of operating conditions and energy consumption. S4. Employs a sliding window and incremental update mode to generate real-time dynamic reports and push optimization and maintenance suggestions.

[0019] In step S1, energy consumption data is collected based on power sensors and flow sensors, while operating conditions are collected in real time using temperature, pressure, vibration, and speed sensors. After preprocessing, the collected data is transmitted to the terminal via wired and wireless means to obtain equipment data.

[0020] S2 uses a federated learning framework to enable local data training and transmit only model parameters. This not only integrates data from multiple sites to improve the model's generalization ability but also avoids the privacy risks of transmitting raw data across sites. At the same time, it stores data and decision rules according to device type to ensure standardized data management and lay a reliable foundation for subsequent analysis. The anomaly identification is accurate and efficient, reducing false alarms and missed detections. The S3 combines intelligent analysis with federated learning models and a dual mechanism of synchronous verification of operating conditions and energy consumption phases. It not only captures the time-series characteristics of energy consumption but also verifies the dynamic linkage logic between the two, which greatly improves the accuracy of anomaly identification and reduces misjudgments that rely on a single model. Moreover, the real-time acquisition and analysis mode can quickly respond to equipment anomalies and prevent the fault from escalating. The S4 generates dynamic reports through incremental updates via a sliding window to ensure that the information is effective in real time.

[0021] In step S1, the rules for judging abnormal equipment energy consumption are sorted out by establishing a dynamic energy consumption baseline by statistical methods based on equipment type and different working conditions; the abnormality types are refined, and a hierarchical judgment logic combining static thresholds and dynamic conditions with time dimension is set. The formula for establishing a dynamic energy consumption baseline using statistical methods is as follows: Normal energy consumption range baseline = [μ-kσ, μ+kσ]; Where μ is the average energy consumption under a certain operating condition group, σ is the standard deviation, and k is the coefficient.

[0022] The criteria for determining the time dimension are: Abnormal duration: Level 1 anomaly (emergency): Energy consumption exceeds the baseline / static threshold and lasts for ≥5 minutes (core equipment) / 10 minutes (auxiliary equipment). Level 2 abnormality (general): Energy consumption exceeds the baseline, duration ≥3 minutes and <5 minutes (core equipment) / ≥5 minutes and <10 minutes (auxiliary equipment). Level 3 anomaly (early warning): Energy consumption exceeds the baseline, but the duration is <3 minutes (core equipment) / <5 minutes (auxiliary equipment), or the threshold is exceeded instantaneously multiple times (≥3 times within 1 hour).

[0023] If the same type of abnormality is triggered ≥3 times within 1 hour (even if the duration of each instance is short), it will be upgraded to a level 2 abnormality to avoid ignoring intermittent faults.

[0024] Hierarchical judgment logic closed loop Initially, if the energy consumption data simultaneously exceeds the dynamic baseline [μ-kσ, μ+kσ] or exceeds the static threshold, an anomaly warning is triggered. Time verification determines the duration of the anomaly and the frequency of triggering within one hour, matching the corresponding anomaly level. As a fallback check, if the energy consumption exceeds the static threshold, regardless of whether it is within the dynamic baseline or the duration, it is directly judged as a level one anomaly. The output clearly defines the anomaly level (Level 1 / Level 2 / Level 3), triggering conditions (dynamic baseline exceeding limits / static threshold exceeding limits), and duration / frequency, providing a priority basis for subsequent handling.

[0025] Step S2 specifically includes: In terms of framework deployment, a horizontal federated architecture is adopted, with local training nodes set up at each site and aggregation nodes set up on the management platform; The model is constructed as a hierarchical architecture, using LSTM to capture energy consumption time-series features, extracting core related features based on the working condition attention mechanism, and outputting feature vectors and anomaly probabilities. Federated aggregation uses the FedAvg weighted average algorithm based on data volume to aggregate global parameters and then distribute them to each node. The convergence model is deployed locally, and the results are output in real time and added to the database to form a data loop.

[0026] This application adopts a horizontal federated architecture, where data from each site is stored locally for training and only model parameters are transmitted, completely eliminating privacy leaks and compliance risks associated with the cross-site transmission of raw energy consumption / operating condition data; at the same time, it aggregates model parameters from multiple site data, breaking through the limitations of limited data volume and single scenario of a single site, allowing the model to learn common features under different sites and operating conditions, greatly improving generalization ability and adapting to diverse equipment operating scenarios; The model boasts high accuracy and is well-suited to the core characteristics of energy consumption data. Its layered architecture addresses key pain points in energy consumption analysis. LSTM is naturally adapted to the temporal dependencies of energy consumption, accurately capturing the correlation between historical energy consumption and real-time data. The operating condition attention mechanism highlights the impact weight of key operating condition parameters on energy consumption, filters out irrelevant interference, and makes the extraction of core related features more focused. The final output feature vector and anomaly probability are more in line with the actual operating logic of the equipment, providing a highly reliable basis for subsequent anomaly identification.

[0027] The specific steps for identifying devices with abnormal energy consumption in step S3 are as follows: The system collects energy consumption and operating condition data in real time at a sampling frequency of 1 minute, through three methods: direct sensor sampling, device communication interface and synchronization with third-party systems, and transmits the data to the local edge node via MQTT encryption. The federated learning energy consumption feature model is invoked, and a dedicated sub-model is matched according to the device type and real-time operating conditions. The core features are extracted and the energy consumption feature vector and initial anomaly probability are output. The threshold is set at 0.8. If the value is greater than the threshold, it is marked as a suspected anomaly. Calculate the correlation of the rates of change and the phase difference between the two, and verify them. When phase decoupling or divergence occurs, it is determined to be an abnormal phase synchronization. Based on the two-level judgment, the final output includes the device ID, abnormal time, type, and confidence level information, and is then pushed to the system.

[0028] Data transmission adopts the MQTT 3.1.1 protocol and is securely transmitted through a TLS 1.3 encrypted channel. A lightweight MQTT Broker is deployed on the local edge node. After receiving data, it performs integrity verification. Data that fails verification is temporarily stored in the local cache for 24 hours. Data that passes verification is immediately associated with the device ID, collection timestamp and collection method identifier and stored in the local time-series database of the edge node.

[0029] Edge nodes perform secondary adaptation on preprocessed real-time data according to the unified standards of the federated training phase; the sub-model matching mechanism locks the model category according to the device type; real-time parsing of operating parameters matches the exclusive sub-model for the corresponding sub-scenario of operating condition; The model is based on LSTM and a working condition attention mechanism, and extracts features in three layers, specifically including: Time-series features are extracted to show the moving average, variance, and peak / valley ratio of energy consumption over the past 10 minutes, thus capturing the time-series patterns of energy consumption. By analyzing the correlation features of operating conditions, we can extract the bias, rate of change, and response delay between energy consumption and real-time operating conditions, and capture the degree of linkage and matching between operating conditions and energy consumption. Baseline deviation characteristics are extracted, including the percentage deviation of the current energy consumption from the dynamic baseline of stage S1 and the duration of continuous deviation, and associated with previous anomaly judgment rules. Output a 128-dimensional fixed-length energy consumption feature vector; The model output layer uses the Sigmoid activation function to map the feature vectors to initial anomaly probabilities in the 0-1 range; The threshold of 0.8 was determined based on historical data from the federated training phase. Through testing with 10,000+ normal samples and 2,000+ labeled abnormal samples, the optimal threshold value was found to be ≥92% for anomaly identification and ≤5% for false positive rate when the threshold was set to 0.8.

[0030] For the correlation calculation of the rate of change, 30 sample points were selected from the sampling data of the past 30 minutes. The rate of change of operating conditions and the rate of change of energy consumption were calculated for each sample point. The rate of change of operating conditions was obtained by subtracting the value of the previous sample point from the operating condition parameter of the current sample point and then dividing by the 60-second sampling interval. Similarly, the rate of change of energy consumption was obtained by dividing the difference between the energy consumption data of the current sample point and the previous sample point by 60 seconds. The phase difference is calculated using the cross-correlation function method. First, a delay step size of 0 to 10 sampling times is set. The sum of the products of the operating condition change rate sequence and the corresponding delayed energy consumption change rate sequence at each step size is calculated. The optimal delay step size that maximizes this sum is found. The optimal step size is multiplied by 60 seconds to obtain the actual phase difference. Positive numbers represent energy consumption lagging operating conditions, and negative numbers represent energy consumption leading operating conditions. The judgment criteria are as follows: the trend is judged based on the correlation of the rate of change of the data in the past 30 minutes. A strong positive correlation is normal, a weak positive correlation needs to be verified, and no correlation or negative correlation is directly judged as phase disconnection or divergence. For samples with weak positive correlation, the phase difference is checked according to the device type, and those exceeding the threshold or with energy consumption ahead are judged as disconnected. An anomaly is identified if any one of the conditions is met.

[0031] Multi-source real-time data acquisition and secure transmission The system uses a sampling frequency of 1 minute as the baseline. When the core equipment is at a load of ≥80% or the operating condition fluctuates by more than 10% per minute, the sampling frequency is automatically reduced to 10 seconds per sampling. The auxiliary equipment maintains the baseline frequency, balancing data granularity and computing power consumption. Data is collected through three methods: direct sampling from sensors to adapt to auxiliary equipment without built-in output, accurately capturing energy consumption and temperature data; communication between the equipment and the core equipment, parsing the core fields of energy consumption and operating condition in the built-in registers through a dedicated protocol; and synchronous integration of data from existing monitoring and energy management platforms from third-party systems, with cross-validation to remove duplicates. Data transmission adopts the MQTT 3.1.1 protocol and achieves end-to-end secure transmission based on the TLS 1.3 encrypted channel. It is equipped with dual authentication of client certificate and password to prevent data tampering or theft. A lightweight MQTTBroker is deployed on the local edge node. After receiving data, it first verifies the integrity and format of the fields. Failed data is temporarily stored in a local cache of more than 10GB. Data that passes the verification is immediately bound to the device ID, collection timestamp and collection method identifier and stored in the local time series database to ensure low-latency access for subsequent analysis.

[0032] Edge nodes first perform secondary adaptation on the preprocessed data, select device-specific core features according to federated training standards, and then align the data distribution through Min-Max scaling to ensure consistency with the model training input format. Sub-model matching follows the logic of "locking the major category by device type → parsing real-time operating parameters to match the sub-scenarios". It quickly matches the exclusive sub-model based on the built-in three-dimensional index table. High-frequency operating condition sub-models reside in memory, while low-frequency ones are loaded on demand. The loading and matching time is controlled within 100 milliseconds.

[0033] The model extracts features in three layers based on LSTM and operating condition attention mechanism: time-series features capture the time series patterns of energy consumption moving average and fluctuation variance over the past 10 minutes; operating condition correlation features extract the bias between energy consumption and operating condition, and the linkage matching index of response delay; baseline deviation features calculate the percentage deviation of the current energy consumption from the S1 dynamic baseline and the duration of continuous deviation, associate it with the previous judgment rules, and output a 128-dimensional energy consumption feature vector, which is mapped to the initial anomaly probability in the 0-1 interval by the Sigmoid activation function.

[0034] The correlation calculation of the rate of change selects 30 sampling points in the past 30 minutes. The difference between the current and previous sampling point's operating conditions and energy consumption data is divided by 60 seconds to obtain two types of rate of change. Then, the trend matching degree is judged by the correlation coefficient: 0.6-1.0 is a strong positive correlation (normal), 0.3-0.6 is a weak positive correlation (needs to be verified), -0.3-0.3 is no correlation (phase disconnection), and -1.0-0.3 is a negative correlation (phase divergence).

[0035] The phase difference is determined using the cross-correlation function method. A sampling delay step of 0-10 is set to find the optimal step that maximizes the sum of the products of the operating conditions and the energy consumption change rate sequence. Multiplying this step by 60 seconds yields the actual phase difference (positive values ​​indicate energy consumption lag, negative values ​​indicate energy consumption lead). For samples with weak positive correlation, the phase difference is verified according to equipment type: wind turbines ≤ 30 seconds, photovoltaic inverters ≤ 15 seconds. If the phase difference exceeds the threshold or is ahead of the threshold, it is determined to be phase disconnection.

[0036] The two-level judgment logic is as follows: Level 1 judgment (high confidence anomaly) is based on an anomaly probability ≥ the threshold and phase synchronization anomaly, which is directly confirmed; Level 2 judgment (suspected anomaly review) is based on a single indicator meeting the standard, extracting the time series data of the past 10 minutes and comparing it with the historical normal curve under the same operating conditions. If the similarity is <85%, it is upgraded to confirmed anomaly. The final output includes auxiliary information such as device ID, anomaly time, type, confidence level, operating condition parameters, and correlation coefficient. The results are pushed to multiple channels, including monitoring platform pop-ups, APP messages, and SMS, with the same anomaly updated every 10 minutes to avoid duplicate pushes.

[0037] In step S4, the core parameters of the sliding window are set, including the window duration and sliding step size. The window data is grouped and stored according to device type and operating scenario. Synchronize incremental data every unit of minute to remove duplicate data and invalid noise; The incremental data is linked with the historical data within the window to complete the data chain of equipment operating status, anomaly records, and changes in operating conditions, and the statistical results of the indicators are updated synchronously.

[0038] The dynamic report in step S4 includes an overview of equipment operation, the number of each type of equipment in operation within the window, the percentage of normal and abnormal equipment, and a summary of abnormal details; the report can be viewed in real time on the PC monitoring platform.

[0039] This application groups and stores window data according to equipment type and operating conditions to ensure the accuracy of the analysis objects and avoid the distortion of results caused by the mixing of data from different equipment and operating conditions. The combination of sliding step size and incremental update only adds valid data synchronously and removes duplicates and noise, without having to repeatedly calculate the full amount of data, which greatly reduces computing power consumption. At the same time, it ensures that the data is updated in real time every unit minute, so that the analysis results keep up with the changes in equipment operating status and balance real-time response and resource conservation.

[0040] Incremental data is linked with historical data within the window to complete the entire chain of information on equipment operating status, anomaly records, and changes in operating conditions, avoiding data fragmentation; the statistical results of core indicators are updated synchronously, so that the report not only presents the current status, but also reflects the dynamic changes of indicators, providing process and result-based data support for the derivation of subsequent operation and maintenance suggestions, and improving the credibility of the analysis conclusions; The structured reporting system enhances both operational visualization and decision-making efficiency. Dynamic reports focus on core value information, clearly presenting the equipment's operational status, the quantity of each type of equipment, and the percentage of normal / abnormal equipment. It also summarizes anomaly details, allowing operations personnel to quickly grasp the overall operational status and key anomalies without sifting through massive amounts of data. Real-time viewing on PC monitoring platforms breaks down the time and space limitations of information acquisition, enabling operations teams to respond to anomalies instantly and preventing the escalation of faults due to information delays. This significantly improves the targeting and timeliness of operations decisions.

[0041] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for analyzing equipment energy consumption in new energy power plants, characterized in that, The analysis steps are as follows: S1. Collect energy consumption data and operating condition data of various equipment in new energy power stations, and sort out the judgment rules for abnormal equipment energy consumption; S2. Construct an equipment energy consumption characteristic model. Based on the federated learning framework, train the energy consumption characteristic model locally and store the energy consumption data, operating condition data and anomaly judgment rules into the analysis database according to equipment type. S3. The system collects equipment energy consumption and operating condition data in real time, calls the energy consumption characteristic model for intelligent analysis, and identifies equipment with abnormal energy consumption based on the phase synchronization analysis of operating conditions and energy consumption. S4. Employs a sliding window and incremental update mode to generate real-time dynamic reports and push optimization and maintenance suggestions.

2. The method for analyzing equipment energy consumption in a new energy power station according to claim 1, characterized in that: In step S1, energy consumption data is collected based on power sensors and flow sensors, while operating conditions are collected in real time using temperature, pressure, vibration, and speed sensors. After preprocessing, the collected data is transmitted to the terminal via wired and wireless means to obtain equipment data.

3. The method for analyzing equipment energy consumption in a new energy power station according to claim 1, characterized in that: In step S1, the rules for judging abnormal equipment energy consumption are sorted out by establishing a dynamic energy consumption baseline by statistical methods based on equipment type and different working conditions; the abnormality types are refined, and a hierarchical judgment logic combining static thresholds and dynamic conditions with time dimension is set. The formula for establishing a dynamic energy consumption baseline using statistical methods is as follows: Normal energy consumption range baseline = [μ-kσ, μ+kσ]; Where μ is the average energy consumption under a certain operating condition group, σ is the standard deviation, and k is the coefficient.

4. The method for analyzing equipment energy consumption in a new energy power station according to claim 1, characterized in that, Step S2 specifically includes: In terms of framework deployment, a horizontal federated architecture is adopted, with local training nodes set up at each site and aggregation nodes set up on the management platform; The model is constructed as a hierarchical architecture, using LSTM to capture energy consumption time-series features, extracting core related features based on the working condition attention mechanism, and outputting feature vectors and anomaly probabilities. Federated aggregation uses the FedAvg weighted average algorithm based on data volume to aggregate global parameters and then distribute them to each node. The convergence model is deployed locally, and the results are output in real time and added to the database to form a data loop.

5. The method for analyzing equipment energy consumption in a new energy power station according to claim 1, characterized in that, The specific steps for identifying devices with abnormal energy consumption in step S3 are as follows: The system collects energy consumption and operating condition data in real time at a sampling frequency of 1 minute, through three methods: direct sensor sampling, device communication interface and synchronization with third-party systems, and transmits the data to the local edge node via MQTT encryption. The federated learning energy consumption feature model is invoked, and a dedicated sub-model is matched according to the device type and real-time operating conditions. The core features are extracted and the energy consumption feature vector and initial anomaly probability are output. The threshold is set at 0.

8. If the value is greater than the threshold, it is marked as a suspected anomaly. Calculate the correlation of the rates of change and the phase difference between the two, and verify them. When phase decoupling or divergence occurs, it is determined to be an abnormal phase synchronization. Based on the two-level judgment, the final output includes the device ID, abnormal time, type, and confidence level information, and is then pushed to the system.

6. The method for analyzing equipment energy consumption in a new energy power station according to claim 5, characterized in that: Data transmission uses MQTT The 3.1.1 protocol is based on a TLS 1.3 encrypted channel for secure transmission. A lightweight MQTT Broker is deployed on the local edge node. After receiving data, it performs integrity verification. Data that fails verification is temporarily stored in the local cache for 24 hours. Data that passes verification is immediately associated with the device ID, collection timestamp, and collection method identifier and stored in the local time-series database of the edge node.

7. The method for analyzing equipment energy consumption in a new energy power station according to claim 5, characterized in that: Edge nodes perform secondary adaptation on the preprocessed real-time data according to the unified standards of the federated training phase; the sub-model matching mechanism locks the model category according to the device type. Real-time analysis of operating parameters to match specific sub-models for corresponding operating conditions and scenarios; The model is based on LSTM and a working condition attention mechanism, and extracts features in three layers, specifically including: Time-series features are extracted to show the moving average, variance, and peak / valley ratio of energy consumption over the past 10 minutes, thus capturing the patterns in the energy consumption time series. By analyzing the correlation features of operating conditions, we can extract the bias, rate of change, and response delay between energy consumption and real-time operating conditions, and capture the degree of linkage and matching between operating conditions and energy consumption. Baseline deviation characteristics are extracted, including the percentage deviation of the current energy consumption from the dynamic baseline of stage S1 and the duration of continuous deviation, and associated with previous anomaly judgment rules. Output a 128-dimensional fixed-length energy consumption feature vector; The model output layer uses the Sigmoid activation function to map the feature vectors to initial anomaly probabilities in the 0-1 range; The threshold of 0.8 was determined based on historical data from the federated training phase. Through testing with 10,000+ normal samples and 2,000+ labeled abnormal samples, the optimal threshold value was found to be ≥92% for anomaly identification and ≤5% for false positive rate when the threshold was set to 0.

8.

8. The method for analyzing equipment energy consumption in a new energy power station according to claim 5, characterized in that: For the correlation calculation of the rate of change, 30 sample points were selected from the sampling data of the past 30 minutes. The rate of change of operating conditions and the rate of change of energy consumption were calculated for each sample point. The rate of change of operating conditions was obtained by subtracting the value of the previous sample point from the operating condition parameter of the current sample point and then dividing by the 60-second sampling interval. Similarly, the rate of change of energy consumption was obtained by dividing the difference between the energy consumption data of the current sample point and the previous sample point by 60 seconds. The phase difference is calculated using the cross-correlation function method. First, a delay step size of 0 to 10 sampling times is set. The sum of the products of the operating condition change rate sequence and the corresponding delayed energy consumption change rate sequence at each step size is calculated. The optimal delay step size that maximizes this sum is found. The optimal step size is multiplied by 60 seconds to obtain the actual phase difference. Positive numbers represent energy consumption lagging operating conditions, and negative numbers represent energy consumption leading operating conditions. The judgment criteria are as follows: the trend is judged based on the correlation of the rate of change of the data in the past 30 minutes. A strong positive correlation is normal, a weak positive correlation needs to be verified, and no correlation or negative correlation is directly judged as phase disconnection or divergence. For samples with weak positive correlation, the phase difference is checked according to the device type, and those exceeding the threshold or with energy consumption ahead are judged as disconnected. An anomaly is identified if any one of the conditions is met.

9. The method for analyzing equipment energy consumption in a new energy power station according to claim 1, characterized in that: In step S4, the core parameters of the sliding window are set, including the window duration and sliding step size. The window data is stored in groups according to device type and operating scenario. Synchronize incremental data every unit of minute to remove duplicate data and invalid noise; The incremental data is linked with the historical data within the window to complete the data chain of equipment operating status, anomaly records, and changes in operating conditions, and the statistical results of the indicators are updated synchronously.

10. A system for analyzing the energy consumption of equipment in a new energy power station based on the method of claim 1, characterized in that, It includes an energy consumption analysis module, a feature model module, an abnormal device module, and an analysis module: The energy consumption analysis module collects energy consumption data and operating condition data of various equipment in new energy power stations, and analyzes the rules for judging abnormal equipment energy consumption. The feature model module constructs a device energy consumption feature model. Based on the federated learning framework, it trains the energy consumption feature model locally and stores energy consumption data, operating condition data and anomaly judgment rules into the analysis database according to device type. The abnormal equipment module collects equipment energy consumption and operating condition data in real time, calls the energy consumption characteristic model for intelligent analysis, and identifies equipment with abnormal energy consumption based on phase synchronization analysis of operating conditions and energy consumption. The analysis module uses a sliding window and incremental update mode to generate real-time dynamic reports and push optimization and maintenance suggestions.

Images