Intelligent Energy Consumption Assessment Method for Photovoltaic Power Generation Equipment Based on Big Data

Abstract

This invention discloses a big data-based intelligent energy consumption assessment method for photovoltaic (PV) power generation equipment, belonging to the field of intelligent PV power generation assessment technology. The method includes: timestamping and cleaning real-time DC power generation sequences and irradiance time-series data of PV power plants to construct standardized string power conversion curves; extracting spatial cloud movement trajectories and identifying cloud shading and load change interference events by combining auxiliary equipment start / stop signals; inputting these two types of events as external interference factors into a power decomposition model to separate the equipment's baseline performance degradation component, cloud instantaneous fluctuation component, and auxiliary equipment internal loss component from the power sequence; calculating the interference-free theoretical power generation efficiency through the baseline performance degradation component and comparing it with the power predicted by the standardized curve to obtain the efficiency deviation coefficient. This invention can effectively isolate meteorological and internal loss interference, accurately reflect the aging and degradation of the equipment itself, and improve the accuracy and reliability of energy consumption assessment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent assessment technology for photovoltaic power generation, specifically a method for intelligent assessment of energy consumption of photovoltaic power generation equipment based on big data. Background Technology

[0002] Current assessments of the energy consumption and health status of photovoltaic power generation equipment mostly involve collecting DC power generation and irradiance data from photovoltaic strings, constructing power conversion curves using regression algorithms after simple time-series matching, and determining the equipment's energy consumption status based on the difference between actual and theoretical power. The assessment process only performs basic processing on power and irradiance, without specifically distinguishing between meteorological interference and power supply interference from internal equipment in the power station.

[0003] Conventional assessment methods cannot eliminate power fluctuations caused by cloud cover and the start-up and shutdown of auxiliary equipment. The instantaneous power changes caused by cloud cover and the power consumption generated by the start-up and shutdown of auxiliary equipment will be superimposed on the aging and degradation of the equipment itself, resulting in distorted power data. It is impossible to accurately separate the energy consumption changes caused by the degradation of the equipment itself, and the assessment results cannot truly reflect the actual health level of photovoltaic power generation equipment.

[0004] To address the aforementioned problems, this invention extracts the spatial cloud movement trajectory from meteorological system cloud cover information, identifies cloud obstruction events and equipment start-up / shutdown load change events by combining auxiliary equipment power consumption start-up / shutdown signals, and inputs the two types of events as external interference factors into the power decomposition model. It separates the baseline performance degradation component, instantaneous fluctuation component, and internal loss component from the real-time DC power generation sequence, calculates the theoretical power generation efficiency through the baseline performance degradation component, and obtains the efficiency deviation coefficient by comparing it with the standardized power conversion curve. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art;

[0006] To this end, the present invention proposes a smart energy consumption assessment method for photovoltaic power generation equipment based on big data, including:

[0007] The real-time DC power generation sequence of photovoltaic power station equipment and the collected irradiance time series data are timestamped and cleaned to form string power and irradiance data pairs;

[0008] Based on string power and irradiance data pairs, a standardized power conversion curve for each photovoltaic string is constructed using a regression analysis algorithm;

[0009] The spatial cloud movement trajectory is extracted from the sky cloud cover change information released synchronously by the meteorological system. Combined with the equipment start-up and shutdown signals in the power consumption records of the power station auxiliary equipment, the cloud cover obstruction events and equipment start-up and shutdown load change events that affect the real-time DC power generation sequence are identified.

[0010] The identified cloud cover events and the equipment start-up and shutdown load change events are used as external interference factors and input into a pre-trained power decomposition model. The power decomposition model is used to separate the baseline performance degradation component caused by the aging of the equipment itself, the instantaneous fluctuation component caused by cloud cover, and the internal friction component caused by the load change of auxiliary equipment from the real-time DC power generation sequence.

[0011] Using the baseline performance degradation component output by the power decomposition model, the theoretical power generation efficiency of the photovoltaic power generation equipment under the condition of no external interference is calculated, and the theoretical power generation efficiency is compared with the theoretical power predicted by the standardized power conversion curve of the corresponding photovoltaic string to obtain the efficiency deviation coefficient characterizing the health of the equipment.

[0012] Furthermore, the process of timestamping and cleaning the real-time DC power generation sequence of the photovoltaic power station equipment with the collected irradiance time-series data to form string power and irradiance data pairs includes:

[0013] The time stamp formats of the real-time DC power generation sequence and the irradiance time series data are respectively parsed and uniformly converted into millisecond-level timestamps under the standard time zone.

[0014] Based on the timeline, a unified time index sequence containing all data collection time points is established;

[0015] For each time point in the unified time index sequence, find and match data points in the real-time DC power generation sequence and the irradiance time series data that are within a preset time tolerance range to form candidate data pairs;

[0016] For each candidate data pair, check whether its power value and irradiance value are within the preset physical reasonable threshold range, and mark the candidate data pairs that exceed the physical reasonable threshold range as anomalies;

[0017] All candidate data pairs marked as outliers are removed, and the remaining valid candidate data pairs are arranged in chronological order to form string power and irradiance data pairs.

[0018] Furthermore, the construction of a standardized power conversion curve for each photovoltaic string based on string power and irradiance data pairs using a regression analysis algorithm includes:

[0019] The subset of samples with irradiance higher than the clear weather threshold and no external interference records in the string power and irradiance data pairs are selected as training data.

[0020] On the training data, with irradiance as the independent variable and real-time DC power generation as the dependent variable, a quadratic polynomial curve is fitted using the least squares method.

[0021] Calculate the goodness of fit of the quadratic polynomial curve. If the goodness of fit is lower than the predetermined standard, add higher-order terms or perform piecewise fitting on the training data until the goodness of fit reaches the standard, and obtain the preliminary power conversion model.

[0022] The preliminary power conversion model is compensated and corrected for the temperature effect, which is calculated by using the ambient temperature data recorded by the environmental monitoring station and the theoretical temperature coefficient of the photovoltaic string.

[0023] The model after temperature compensation correction is determined as the standardized power conversion curve of the photovoltaic string. The standardized power conversion curve is stored in function form and describes the mapping relationship between irradiance and theoretical power generation under ideal conditions.

[0024] Furthermore, the extraction of spatial cloud movement trajectories from the cloud cover change information synchronously released by the meteorological system, combined with equipment start-up and shutdown signals from the power consumption records of the power station auxiliary equipment, identifies cloud obstruction events affecting the real-time DC power generation sequence, including:

[0025] The cloud cover change information is analyzed in consecutive frames, and each frame contains cloud type encoding and spatial coordinate information for different cloud areas;

[0026] Between consecutive frames, feature point matching and tracking are performed on cloud areas with similar cloud type encoding and spatial coordinates to form multiple cloud movement trajectory line segments;

[0027] Based on the direction and speed of the cloud movement trajectory line segment, the Kalman filter algorithm is used to predict its subsequent spatial position and generate a complete cloud movement trajectory.

[0028] Calculate the ground projection area covered by the cloud movement trajectory, and perform spatial intersection analysis between the ground projection area and the electronic fence map of the photovoltaic power station to obtain the expected time period of the cloud passing through the power station and the range of the affected sub-area.

[0029] By combining the sudden drop and recovery patterns of the irradiance time series data, the time period during which the cloud is expected to pass over the power station is verified, and the verified time period and sub-region range are marked as the cloud occlusion event.

[0030] Furthermore, the process of identifying equipment start-up and shutdown load change events affecting the real-time DC power generation sequence by combining equipment start-up and shutdown signals from the power consumption records of power plant auxiliary equipment includes:

[0031] Acquire power consumption records of power plant auxiliary equipment, the power consumption records including multi-channel current and voltage timing data of each equipment, as well as status change signals actively reported by the equipment controller;

[0032] Extract the state change signal from the power consumption record. The state change signal includes at least the device start time, the shutdown time, and the device operating mode switching time. Use the device start time, shutdown time, and device operating mode switching time as the initial time markers of the device start-up and shutdown events.

[0033] Within the neighborhood time window of the initial time stamp, the multi-channel current and voltage timing data are parsed, and the rising and falling edges of the current or power are identified by the threshold judgment method, thereby locating the actual load connection or disconnection time of the device, and correcting the time point to the final device start-stop signal timestamp.

[0034] Based on the timestamp of the equipment start-stop signal, locate the corresponding time in the real-time DC power generation sequence, extract power data segments of a preset duration before and after the timestamp of the equipment start-stop signal, and calculate the mean change of the power data segments;

[0035] The mean change of the power data segment is compared with a preset load power impact threshold for different equipment types. If the mean change exceeds the corresponding load power impact threshold, it is determined that a device start-stop load change event has occurred, and the start time, end time, affected power change amount, and associated equipment identifier of the device start-stop load change event are recorded.

[0036] Furthermore, the training and output of the power decomposition model include:

[0037] Collect multi-dimensional raw operating data from historical operating cycles as a training set. The training set has been labeled with time period tags for known baseline performance degradation events, cloud cover events, and equipment start-up and shutdown load change events.

[0038] Construct a decomposition model containing a multi-layer neural network. The input of the model is the real-time DC power generation sequence, irradiance time series data and equipment status signal within a specified time window. The output of the model is the estimated value sequence of the baseline performance degradation component, the instantaneous fluctuation component and the internal friction component.

[0039] The decomposition model is trained under supervision using the training set, and the error between the model output estimate and the true component corresponding to the known event label is minimized by the backpropagation algorithm.

[0040] After training, the network parameters of the decomposition model are fixed, and it is deployed to an online evaluation system.

[0041] The real-time DC power generation sequence, irradiance time series data, and equipment status signal to be analyzed are input into the deployed decomposition model, which directly outputs the separated baseline performance degradation component, instantaneous fluctuation component, and internal friction component.

[0042] Furthermore, the calculation of the theoretical power generation efficiency of the photovoltaic power generation equipment under conditions without external interference includes:

[0043] Extract a time-continuous and stationary data segment from the baseline performance degradation component output by the decomposition model.

[0044] Calculate the average power value of the data segment and use it as the actual benchmark output power of the photovoltaic power generation equipment under the current healthy state;

[0045] From the standardized power conversion curve, the average value of the irradiance data within the same time period is taken and input into the standardized power conversion curve to calculate the theoretical expected power of the photovoltaic string under the current environmental conditions.

[0046] Divide the actual benchmark output power by the theoretical expected power to obtain the theoretical power generation efficiency of the photovoltaic power generation equipment under the current state;

[0047] The curve of the theoretical power generation efficiency changing over time is continuously calculated and recorded.

[0048] Furthermore, the efficiency deviation coefficient characterizing the health of the equipment includes:

[0049] Obtain the theoretical power generation efficiency of the photovoltaic string during the stable operation phase after initial commissioning, and use it as the benchmark efficiency;

[0050] Extract the average theoretical power generation efficiency within the most recent statistical period from the currently calculated theoretical power generation efficiency change curve;

[0051] Calculate the percentage relative deviation between the average theoretical power generation efficiency and the benchmark efficiency;

[0052] The relative deviation percentage is compared with multiple preset aging level thresholds to determine the current aging level of the device;

[0053] The relative deviation percentage is combined with the aging level and encoded to generate an efficiency deviation coefficient that characterizes the current health of the device.

[0054] Furthermore, the method also includes a device-level energy consumption attribution step based on energy flow tracing:

[0055] Using the baseline performance degradation component, the instantaneous fluctuation component, and the internal friction component output by the power decomposition model, the theoretical power generation curve, the external disturbance loss power curve, and the auxiliary system power consumption curve of the photovoltaic power generation equipment in any time period are reconstructed.

[0056] At the power station level, the power consumption curves of the auxiliary systems of all equipment are summed to obtain the total auxiliary energy consumption curve of the power station, and cross-validated with the mains power consumption record or the fuel consumption record of the standby generator set input to the power station.

[0057] At the string level, the theoretical power generation curve is compared with the external disturbance loss power curve to calculate the average energy loss rate caused by identifiable external factors.

[0058] An energy consumption attribution report is generated for each photovoltaic string. The report includes at least the statistical characteristics of the theoretical power generation curve, the time distribution pattern of the external disturbance loss power curve, the composition of the auxiliary system power consumption curve, and the calculated average energy loss rate.

[0059] Furthermore, the generation of an energy consumption attribution report for each photovoltaic string includes:

[0060] Set a fixed report generation cycle, and automatically trigger the report generation process at the end of the cycle;

[0061] During the report generation period, the corresponding theoretical power generation curve, external disturbance loss power curve, auxiliary system power consumption curve, average energy loss rate, and efficiency deviation coefficient are collected.

[0062] According to a predefined data structure, the collected theoretical power generation curve, external disturbance loss power curve, auxiliary system power consumption curve, average energy loss rate, and efficiency deviation coefficient are integrated to form a structured data block.

[0063] The structured data blocks are populated into a pre-designed report template, which includes a time range, device identification, visualization charts of each power curve, a summary of key indicators, and a historical trend comparison area.

[0064] After the data is populated, the report will be output as a document in the specified format and stored in the preset report archive directory. At the same time, a notification that the report generation is complete will be sent to the associated management system.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] By extracting the spatial cloud movement trajectory from the cloud cover change information synchronously released by the meteorological system, and combining it with the equipment start-up and shutdown signals in the power consumption records of the power station's auxiliary equipment, cloud cover obstruction events and equipment start-up and shutdown load change events that affect the real-time DC power generation sequence can be accurately identified. Natural meteorological interference and power consumption interference of internal equipment in the power station can be independently defined, distinguishing between two different types of power fluctuation factors, avoiding confusion between interference from different sources, and allowing the source of abnormal power changes to be directly located.

[0067] By inputting cloud cover events and equipment start-up / shutdown load change events as external disturbance factors into a pre-trained power decomposition model, the model can separate the baseline performance degradation component caused by equipment aging, the instantaneous fluctuation component caused by cloud cover, and the internal loss component caused by auxiliary equipment load changes from the real-time DC power generation sequence. These three components are independent of each other and correspond to different sources of power influence. The theoretical power generation efficiency under the condition of no external disturbance is calculated based on the baseline performance degradation component. The theoretical power generation efficiency is compared with the theoretical power predicted by the standardized power conversion curve of the corresponding photovoltaic string. The efficiency deviation coefficient obtained can directly correspond to the health status of the equipment itself, thus eliminating the influence of external factors on the equipment energy consumption assessment results. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the steps of the intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data as described in this invention.

[0069] Figure 2 A flowchart for aligning and cleaning timestamps to form string power and irradiance data pairs;

[0070] Figure 3 A flowchart for identifying cloud occlusion events. Detailed Implementation

[0071] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] See Figure 1 This invention provides a method for intelligent energy consumption assessment of photovoltaic power generation equipment based on big data, and its overall implementation scheme is as follows:

[0073] The real-time DC power generation sequence of photovoltaic power plant equipment and the collected irradiance time-series data are timestamped and cleaned. This process unifies the timestamp format of different source data and removes outliers, ultimately forming a time-synchronized and reliable string power and irradiance data pair. Based on the cleaned data pair, a standardized power conversion curve for each photovoltaic string is constructed using regression analysis algorithms. This curve represents the theoretical mapping relationship between irradiance and power generation under ideal conditions. The spatial cloud movement trajectory is extracted from the cloud cover change information synchronously released by the meteorological system. Combined with equipment start-up and shutdown signals from the power consumption records of the power plant's auxiliary equipment, cloud shading events and equipment start-up and shutdown load change events affecting the power generation sequence are identified. These identified events are used as external interference factors and input into a pre-trained power decomposition model. This model can separate three key components from the real-time DC power generation sequence: the baseline performance degradation component caused by the aging of the equipment itself, the instantaneous fluctuation component caused by cloud shading, and the internal friction component caused by changes in the auxiliary equipment load. By using the baseline performance degradation component output by the power decomposition model, the theoretical power generation efficiency of the photovoltaic power generation equipment under the condition of no external interference is calculated. This theoretical power generation efficiency is then compared with the theoretical power predicted by the standardized power conversion curve of the corresponding photovoltaic string, thereby obtaining the efficiency deviation coefficient that can characterize the health status of the equipment.

[0074] In one embodiment of the present invention, the timestamp alignment and cleaning process is as follows: (See attached document) Figure 2 The time stamp formats of real-time DC power generation sequences and irradiance time series data are analyzed separately and converted into millisecond-level timestamps in a standard time zone. A unified time index sequence containing all acquisition time points is established based on the time axis. For each time point in the unified time index sequence, data points within a preset time tolerance range in the real-time DC power generation sequence and irradiance time series data are searched and matched. These matching pairs constitute candidate data pairs. For each candidate data pair, its power and irradiance values ​​are checked to see if they are within a preset physically reasonable threshold range. Candidate data pairs exceeding the physically reasonable threshold range are marked as outliers. All candidate data pairs marked as outliers are removed, and the remaining valid candidate data pairs are arranged in chronological order to form the final string power and irradiance data pairs.

[0075] The process of constructing a standardized power conversion curve involves selecting a subset of samples from the string power and irradiance data pairs where the irradiance is above the clear skies threshold and there are no external interference records, as training data. On the training data, a quadratic polynomial curve is fitted using the least squares method with irradiance as the independent variable and real-time DC power generation as the dependent variable. The goodness of fit of this quadratic polynomial curve is calculated. If the goodness of fit is lower than a predetermined standard, higher-order terms are added or the training data is piecewise fitted until the goodness of fit reaches the standard, thus obtaining a preliminary power conversion model. Temperature-based compensation corrections are then applied to this preliminary power conversion model. The temperature effect is calculated using ambient temperature data recorded by an environmental monitoring station and the theoretical temperature coefficient of the photovoltaic string. The model after temperature compensation correction is determined as the standardized power conversion curve of the photovoltaic string. This standardized power conversion curve is stored in function form, describing the mapping relationship between irradiance and theoretical power generation under ideal conditions.

[0076] In practical implementation, the process of aligning and cleaning the real-time DC power generation sequence of photovoltaic power plant equipment with the collected irradiance time-series data to form string power and irradiance data pairs is as follows: The real-time DC power generation sequence is recorded by the power plant monitoring system in the format "YYYY-MM-DDHH:MM:SS", and the irradiance time-series data is recorded by the environmental monitoring station in the format "YY / MM / DDHH:MM". In practice, the cleaning process parses the time stamp formats of the real-time DC power generation sequence and the irradiance time-series data respectively, and converts them into millisecond-level timestamps in Coordinated Universal Time (UTC). Based on the time axis, a unified time index sequence containing all collection time points is established, with an interval of one second. For each time point in the unified time index sequence, data points in the real-time DC power generation sequence and the irradiance time-series data that are within the preset time tolerance range are searched and matched. The preset time tolerance range is set to ±5 seconds, thus forming candidate data pairs. In some embodiments, for each candidate data pair, its power value and irradiance value are checked to see if they are within a preset physically reasonable threshold range. The physically reasonable threshold range for power value is zero to the nominal power of the photovoltaic string, and the physically reasonable threshold range for irradiance value is zero to 1,500 watts per square meter. Candidate data pairs that exceed the physically reasonable threshold range are marked as outliers, and all candidate data pairs marked as outliers are removed. It can be understood that the remaining valid candidate data pairs are arranged in chronological order to ultimately form the cleaned string power and irradiance data pairs.

[0077] In specific implementation, the process of constructing a standardized power conversion curve for each photovoltaic string based on string power and irradiance data pairs using regression analysis algorithms is as follows: A subset of samples from the string power and irradiance data pairs with irradiance exceeding the clear weather threshold and without external interference records is selected as training data. The clear weather threshold is set to 800 watts per square meter. On the training data, irradiance is used as the independent variable, and real-time DC power generation is used as the dependent variable. A quadratic polynomial curve is fitted using the least squares method. The goodness of fit of the quadratic polynomial curve is calculated, and the goodness of fit is measured by the coefficient of determination. In some embodiments, if the goodness of fit is lower than a predetermined standard (set to 0.95), higher-order terms are added or the training data is piecewise fitted until the goodness of fit reaches the standard, resulting in a preliminary power conversion model. Optionally, adding higher-order terms means upgrading the fitted polynomial from quadratic to cubic. Temperature-related compensation corrections are applied to the preliminary power conversion model. The temperature effect is calculated using ambient temperature data recorded by an environmental monitoring station and the theoretical temperature coefficient of the photovoltaic string. In a specific calculation, the theoretical power output value... Compared with the output value of the initial power conversion model The relationship between them can be expressed by the following formula:

[0078]

[0079] in: This represents the corrected theoretical power output under standard test conditions. This indicates that the preliminary power conversion model is based on irradiance. Calculated power, It is the theoretical power temperature coefficient of the photovoltaic string. It is the ambient temperature recorded by the environmental monitoring station. The reference temperature is 25 degrees Celsius under standard test conditions. It can be understood that the model after temperature compensation correction is defined as the standardized power conversion curve of the photovoltaic string, which is stored in function form. The standardized power conversion curve describes the mapping relationship between irradiance and theoretical power generation under ideal conditions.

[0080] In one embodiment of the present invention, the process of extracting the spatial cloud movement trajectory and identifying cloud obstruction events from the sky cloud cover change information synchronously released by the meteorological system is as follows: (See [link to relevant documentation]). Figure 3The system performs continuous frame analysis of cloud cover changes, with each frame containing cloud type codes and spatial coordinates for different cloud regions. Feature point matching and tracking are performed between consecutive frames for cloud regions with similar cloud type codes and spatial coordinates, forming multiple cloud movement trajectory segments. Based on the movement direction and velocity of these cloud trajectory segments, a Kalman filter algorithm is used to predict their subsequent spatial positions, generating complete cloud movement trajectories. The ground projection area covered by the cloud movement trajectory is calculated, and spatial intersection analysis is performed between this ground projection area and the electronic fence map of the photovoltaic power station. This yields the expected time period of cloud passage over the power station and the affected sub-region range. The expected time period of cloud passage over the power station is verified by combining the sudden drop and recovery patterns of irradiance time-series data, and the verified time period and sub-region range are marked as cloud shading events. Accurate identification of cloud shading events and equipment start-up / shutdown load change events provides crucial input for subsequent steps to distinguish different sources of power fluctuations. The identified events will serve as explicit interference factors, guiding the power decomposition model to separate the corresponding interference components from the total power.

[0081] The process of identifying equipment start-up and shutdown load change events from the power consumption records of power plant auxiliary equipment involves: acquiring the power consumption records of the power plant auxiliary equipment, which contain multi-channel current and voltage time-series data of each device, as well as status change signals actively reported by the device controller. The status change signals are extracted from the power consumption records. These signals include at least the device start-up time, shutdown time, and device operating mode switching time, and are used as the initial timestamps for the equipment start-up and shutdown events. The multi-channel current and voltage time-series data are parsed within the neighborhood time window of the initial timestamps. A threshold judgment method is used to identify the rising and falling edges of current or power surges, thereby locating the actual load connection or disconnection time of the device. This time point is then corrected to the final equipment start-up and shutdown signal timestamp. Based on the equipment start-up and shutdown signal timestamp, the corresponding time is located in the real-time DC power generation sequence, and power data segments of a preset duration before and after this timestamp are extracted. The mean change of this power data segment is then calculated. The mean change of power data segments is compared with preset load power impact thresholds for different equipment types. If the mean change exceeds the corresponding load power impact threshold, an equipment start-stop load change event is determined to have occurred, and the start time, end time, affected power change, and associated equipment identifier of the event are recorded.

[0082] In practice, the spatial cloud movement trajectory is extracted from the cloud cover change information synchronously released by the meteorological system. Combined with equipment start-up and shutdown signals from the power plant's auxiliary equipment power consumption records, cloud obstruction events affecting the real-time DC power generation sequence are identified. The cloud cover change information released by the meteorological system is provided in satellite cloud image sequences at five-minute intervals. Each frame of cloud image data contains cloud type codes for different cloud regions and spatial coordinate information based on latitude and longitude. In practice, the cloud cover change information is analyzed in consecutive frames. Feature point matching and tracking are performed on cloud regions with similar cloud type codes and spatial coordinates between consecutive frames. Feature point matching uses a scale-invariant feature transformation method. Successfully matched feature points in consecutive frames are connected to form multiple cloud movement trajectory segments. Based on the direction and velocity of the cloud movement trajectory segments, the Kalman filter algorithm is used to predict their subsequent spatial position, generating a complete cloud movement trajectory. It can be understood that the state vector of the Kalman filter algorithm... Including the location of the cloud center and speed Its prediction equation is as follows:

[0083]

[0084] in: Indicates at time State prediction value, It is the state transition matrix. It is a moment The state estimate is calculated. The ground projection area covered by the complete cloud movement trajectory is calculated, and spatial intersection analysis is performed between the ground projection area and the electronic fence map of the photovoltaic power station. In some embodiments, the spatial intersection analysis is performed by performing polygon overlay analysis through a geographic information system database to obtain the time period during which the cloud is expected to pass over the power station and the affected sub-region range. The time period during which the cloud is expected to pass over the power station is verified by combining the sudden drop and recovery pattern of irradiance time series data. When the irradiance drops by more than 500 watts per square meter within ten minutes and recovers within the following twenty minutes, the verified time period and sub-region range are marked as a cloud shading event.

[0085] In specific implementation, by combining the equipment start-up and shutdown signals in the power consumption records of power plant auxiliary equipment, the load change events affecting the real-time DC power generation sequence are identified. Power consumption records of the power plant auxiliary equipment are acquired, containing multi-channel current and voltage timing data for each device, as well as status change signals actively reported by the device controller. Status change signals are extracted from the power consumption records. These signals include at least the device start-up time, shutdown time, and device operating mode switching time, and are used as the initial timestamps for the equipment start-up and shutdown events. In some embodiments, the multi-channel current and voltage timing data are parsed within a neighborhood time window of the initial timestamp, with the neighborhood time window set to 30 seconds before and after the initial timestamp. A threshold judgment method is used to identify the rising and falling edges of current or power, with the threshold set to 20% of the historical average operating current value of the device, thereby locating the actual load connection or disconnection time of the device, and correcting the time point to the final device start-up and shutdown signal timestamp. Based on the timestamp of the equipment start / stop signal, the corresponding time is located in the real-time DC power generation sequence, and power data segments of a preset duration (five minutes) before and after the timestamp are extracted. Optionally, the mean change of the power data segments is calculated, which is the difference between the average power in the five minutes after the event and the average power in the five minutes before the event. The mean change of the power data segments is compared with preset load power impact thresholds for different equipment types, which are set according to the rated power of the equipment. It can be understood that if the mean change exceeds the corresponding load power impact threshold, an equipment start / stop load change event is determined to have occurred, and the start time, end time, affected power change, and associated equipment identifier of the equipment start / stop load change event are recorded.

[0086] In one embodiment of the present invention, the training and output of the power decomposition model includes the following steps. The power decomposition model is the core of this method for achieving accurate evaluation. The model is trained using external interference events identified in the preceding steps as supervisory signals. Its core function is to decouple and output an independent baseline performance degradation component from the raw, mixed DC power generation sequence. This component is the sole data basis for subsequent equipment performance evaluation. Multi-dimensional raw operating data from historical operating cycles is collected as a training set, which is labeled with time period tags for known baseline performance degradation events, cloud cover events, and equipment start-up / shutdown load change events. A decomposition model containing a multi-layer neural network is constructed. The input to this model is the real-time DC power generation sequence, irradiance time-series data, and equipment status signals within a specified time window. The model output is a sequence of estimated values ​​for the baseline performance degradation component, instantaneous fluctuation component, and internal friction component. The decomposition model is trained under supervision using the training set, and the error between the model output estimates and the true components corresponding to the known event tags is minimized through a backpropagation algorithm. After training, the network parameters of the decomposition model are fixed, and it is deployed to an online evaluation system. The real-time DC power generation sequence, irradiance time series data and equipment status signal to be analyzed are input into the deployed decomposition model, and the separated baseline performance degradation component, instantaneous fluctuation component and internal friction component are directly output.

[0087] In practical implementation, the training and output of the power decomposition model includes the following steps: First, collect multi-dimensional raw operational data from a historical operating cycle as the training set. The historical operating cycle is typically a full year after the power plant is put into operation. The training set is labeled with time period tags for known baseline performance degradation events, cloud cover events, and equipment start-up / shutdown load change events. These tags are obtained through manual verification and labeling using historical maintenance logs and synchronized meteorological reports. The multi-dimensional raw operational data includes at least the real-time DC power generation sequence of the photovoltaic strings, irradiance time-series data, ambient temperature sequence, and status signal sequence of the power plant's auxiliary equipment. Second, construct a decomposition model containing a multi-layer neural network. The specific architecture of the model includes an input layer, several long short-term memory (LSTM) network layers, a fully connected layer, and an output layer. The model input consists of the real-time DC power generation sequence, irradiance time-series data, and equipment status signals within a specified time window. The time window length is set to six hours, corresponding to 1440 input data points. The model output is a sequence of estimated values ​​for the baseline performance degradation component, instantaneous fluctuation component, and internal friction component. The length of the output sequence is the same as the length of the input time window. In some embodiments, the training set data structure used for model training is shown in Table 1:

[0088] Table 1: Training Set Data Structure Table Used for Model Training

[0089] field name Data types illustrate timestamp Timestamp Data collection time point power_actual floating-point numbers Real-time DC power generation irradiance floating-point numbers Irradiance data flag_baseline Integer Reference performance degradation event tag flag_cloud Integer Cloud obscuring event tags flag_load Integer Equipment start / stop load change event tags

[0090] The decomposition model is trained under supervision using a training set, and the error between the model's output estimate and the true components corresponding to the known event labels is minimized through backpropagation. This can be understood as the loss function during training. Defined as the sum of the mean square errors between the estimated values ​​and actual values ​​of each component:

[0091]

[0092] in: Indicates in model parameters The total loss below Indicates the number of training samples. , , They represent the model for the first... Estimates of the baseline performance degradation component, instantaneous fluctuation component, and internal friction component for each sample. , , These represent the corresponding true component label values. After training, the network parameters of the decomposition model are fixed, and it is deployed to the online evaluation system. Optionally, the model deployment can be encapsulated as an independent service that can be called via an application programming interface. The real-time DC power generation sequence to be analyzed, irradiance time-series data, and equipment status signals are input into the deployed decomposition model. The model directly calculates and outputs the separated baseline performance degradation component, instantaneous fluctuation component, and internal friction component via forward propagation. In some embodiments, the online evaluation system uses a streaming processing method, triggering a model inference process every six hours of new data received. It can be understood that the output of the power decomposition model is three time-aligned power sequences, corresponding to the contributions of equipment aging, cloud cover, and auxiliary equipment load changes to the total power, respectively.

[0093] In one embodiment of the present invention, the process of calculating the theoretical power generation efficiency of a photovoltaic power generation device under conditions without external interference involves extracting a time-continuous and stable data segment from the baseline performance degradation component output by the power decomposition model. This step is the final calculation stage of the evaluation, and its innovation lies in the fact that the input data used is not the original total power, but a baseline performance degradation component generated by the power decomposition model that only reflects the aging trend of the device itself. Since this component has eliminated instantaneous fluctuations caused by clouds and load, the theoretical power generation efficiency calculated based on it can stably and directly characterize the performance level of the device itself, and the evaluation results are not affected by external accidental factors. The average power value of this data segment is calculated and used as the actual baseline output power of the photovoltaic power generation device under the current healthy state. The average value of irradiance data within the same time period is taken from the standardized power conversion curve and input as the curve to calculate the theoretical expected power of the photovoltaic string under the current environmental conditions. The actual baseline output power is divided by the theoretical expected power to obtain the theoretical power generation efficiency of the photovoltaic power generation device under the current state, and the curve of the change of the theoretical power generation efficiency over time is continuously calculated and recorded.

[0094] The process of obtaining the efficiency deviation coefficient, which characterizes the health of the equipment, involves obtaining the theoretical power generation efficiency of the photovoltaic string during its stable operation phase after initial commissioning and using it as the benchmark efficiency. The average theoretical power generation efficiency within the most recent statistical period is extracted from the currently calculated theoretical power generation efficiency change curve. The relative percentage deviation between this average theoretical power generation efficiency and the benchmark efficiency is calculated. This relative deviation percentage is compared with multiple preset aging level thresholds to determine the current aging level of the equipment. The relative deviation percentage and aging level are combined and encoded to generate the efficiency deviation coefficient, which characterizes the current health of the equipment. In the preceding steps, the power decomposition model has separated the real-time power sequence into a benchmark performance degradation component, an instantaneous fluctuation component, and an internal loss component. The core of this step lies in using the pure benchmark performance degradation component, stripped of external interference, for efficiency evaluation. This is the key to the breakthrough in accuracy achieved by this method compared to traditional schemes. Traditional evaluation methods directly use the original total power, which is subject to multiple disturbances, to calculate efficiency. The results are easily affected by sudden weather changes and equipment start-ups and shutdowns, leading to inaccurate judgments about the aging state of the equipment itself. This method, however, is based on a benchmark component that only reflects the slow changes in the equipment's own performance, thus ensuring the stability and authenticity of the evaluation basis.

[0095] In practical implementation, the process of calculating the theoretical power generation efficiency of photovoltaic (PV) power generation equipment under conditions without external interference involves extracting a time-continuous and stationary data segment from the baseline performance degradation component output by the power decomposition model. The length of this data segment is set to thirty minutes, and its stationarity is judged by the variance of the calculated segment, which must be below a preset threshold. The average power value of the data segment is calculated and used as the actual baseline output power of the PV power generation equipment under the current healthy state. The average value of irradiance data within the same time period is taken from the standardized power conversion curve and input into the standardized power conversion curve to calculate the theoretical expected power of the PV string under the current environmental conditions. This can be understood as the theoretical power generation efficiency... The calculation formula is the ratio of the actual reference output power to the theoretical expected power, and it is expressed as follows:

[0096]

[0097] in: This represents the calculated theoretical power generation efficiency. This represents the average power value of the data segment extracted from the reference performance degradation component, i.e., the actual reference output power. This represents the theoretical expected power calculated based on the average irradiance from the standardized power conversion curve. Dividing the actual reference output power by the theoretical expected power yields the theoretical power generation efficiency of the photovoltaic power generation equipment under the current conditions. The curve of the theoretical power generation efficiency changing over time is continuously calculated and recorded, with one data point calculated and stored every hour.

[0098] The process of obtaining the efficiency deviation coefficient, which characterizes the health of the equipment, involves acquiring the theoretical power generation efficiency of the photovoltaic string during its stable operation phase after initial commissioning, and using this as the benchmark efficiency. The stable operation phase after initial commissioning is defined as the third month after the equipment is connected to the grid, and the benchmark efficiency is the average of all theoretical power generation efficiency data within that month. The average theoretical power generation efficiency within the most recent statistical period is extracted from the currently calculated theoretical power generation efficiency change curve; the statistical period is set to thirty days. In some embodiments, the relative percentage deviation between the average theoretical power generation efficiency and the benchmark efficiency is calculated. (Relative deviation percentage) The calculation method is to divide the difference between the average theoretical power generation efficiency and the baseline efficiency by the baseline efficiency, and then multiply by 100%. In some embodiments, multiple preset aging level thresholds are shown in Table 2:

[0099] Table 2: Aging Level Threshold Table

[0100] Aging level Relative deviation percentage range Level coding healthy H Slight attenuation M1 Moderate attenuation M2 Severe attenuation S

[0101] The relative deviation percentage is compared with multiple preset aging level thresholds to determine the current aging level of the equipment. Optionally, the comparison process involves matching the calculated relative deviation percentage value with the "Relative Deviation Percentage Range" column in the table level by level. In essence, the relative deviation percentage and aging level are combined and coded to generate an efficiency deviation coefficient characterizing the current health of the equipment. The format of the combined code is "Level Code - Percentage Value". For example, if the calculated relative deviation percentage of a photovoltaic string is -3.2%, and its corresponding aging level is "Slight Degradation (M1)", then the generated efficiency deviation coefficient is "M1-3.2". The efficiency deviation coefficient is then stored in the database along with a timestamp and the equipment identifier.

[0102] In one embodiment of the present invention, the implementation of the device-level energy consumption attribution step based on energy flow tracing includes reconstructing the theoretical power generation curve, external disturbance loss power curve, and auxiliary system power consumption curve of the photovoltaic power generation equipment over any time period using the baseline performance degradation component, instantaneous fluctuation component, and internal friction component output by the power decomposition model. At the power plant level, the auxiliary system power consumption curves of all equipment are summed to obtain the total auxiliary energy consumption curve of the power plant, and cross-validated with the mains power consumption record or standby generator fuel consumption record input to the power plant. At the string level, the theoretical power generation curve and the external disturbance loss power curve are compared to calculate the average energy loss rate caused by identifiable external factors. An energy consumption attribution report is generated for each photovoltaic string, which includes at least the statistical characteristics of the theoretical power generation curve, the temporal distribution pattern of the external disturbance loss power curve, the composition of the auxiliary system power consumption curve, and the calculated average energy loss rate. The data basis for all indicators in the energy consumption attribution report, especially the theoretical power generation curve and the theoretical power generation efficiency calculated accordingly, is the baseline performance degradation component stripped of external disturbances. This allows the report to not only clearly show the composition of energy loss, but also to provide a reliable performance assessment that reflects only the health status of the equipment itself, thus providing a direct and accurate basis for decision-making regarding preventive maintenance and performance optimization of the power plant.

[0103] The specific process for generating an energy consumption attribution report involves setting a fixed report generation cycle, which is automatically triggered at the end of the cycle. Within the generation cycle, the corresponding theoretical power generation curve, external disturbance loss power curve, auxiliary system power consumption curve, average energy loss rate, and efficiency deviation coefficient are collected. These curves and indicators are then integrated into a structured data block according to a predefined data structure. This structured data block is then populated into a pre-designed report template, which includes a time range, equipment identification, visualization charts of each power curve, a summary of key indicators, and a historical trend comparison area. After completion, the report is output as a document in a specified format and stored in a pre-defined report archive directory. Simultaneously, a notification indicating report generation is complete is sent to the associated management system.

[0104] In practical implementation, the device-level energy consumption attribution step based on energy flow tracing includes reconstructing the theoretical power generation curve, external disturbance loss power curve, and auxiliary system power consumption curve of photovoltaic power generation equipment within any time period using the baseline performance degradation component, instantaneous fluctuation component, and internal friction component output by the power decomposition model. The theoretical power generation curve is directly constructed from the baseline performance degradation component, the external disturbance loss power curve is constructed from the instantaneous fluctuation component, and the auxiliary system power consumption curve is constructed from the internal friction component. At the power plant level, the auxiliary system power consumption curves of all photovoltaic power generation equipment are summed. The summation process involves accumulating the internal friction component values ​​of all photovoltaic strings at the same time to obtain the total auxiliary energy consumption curve of the power plant. In some embodiments, the total auxiliary energy consumption curve of the power plant is cross-validated with the mains power consumption record or standby generator fuel consumption record input by the power plant. Cross-validation is performed by comparing the total energy consumption value calculated by integrating the total auxiliary energy consumption curve of the power plant within the same time period with the equivalent energy consumption value converted from the mains power meter reading or generator fuel consumption, and calculating the consistency ratio between the two.

[0105] At the string level, the theoretical power generation curve is compared with the power loss curve due to external disturbances to calculate the average energy loss rate caused by identifiable external factors. This can be understood as the average energy loss rate. The calculation formula is expressed as follows:

[0106]

[0107] in: Indicates the average energy loss rate. This represents the numerical value of the external disturbance power loss curve at time t. This represents the theoretical power generation curve value at time t. This indicates the statistical time period for comparison; the integral operation represents the time period... The power curves are accumulated internally. An energy consumption attribution report is generated for each photovoltaic string. This report includes at least the statistical characteristics of the theoretical power generation curve, the temporal distribution pattern of the external disturbance loss power curve, the composition of the auxiliary system power consumption curve, and the calculated average energy loss rate. The statistical characteristics include the maximum, minimum, average, and standard deviation of the theoretical power generation curve, and the temporal distribution pattern describes the frequency and duration of high-loss events in the external disturbance loss power curve.

[0108] The specific process for generating an energy consumption attribution report involves setting a fixed report generation cycle, either weekly or monthly, with the report generation process automatically triggered at the end of the cycle. Within this cycle, the corresponding theoretical power generation curve, external disturbance loss power curve, auxiliary system power consumption curve, average energy loss rate, and efficiency deviation coefficient are collected. Following a predefined data structure, these collected curves are integrated into a structured data block. This predefined data structure uses key-value pairs, where the key is the indicator name and the value is a numerical value or an array. The structured data block is then populated into a pre-designed report template, which includes a time range, equipment identification, visualization charts for each power curve, a summary of key indicators, and a historical trend comparison area. Optionally, the visualization charts display the changes of the three power curves over time as line charts, and the summary of key indicators lists values ​​such as the average energy loss rate and efficiency deviation coefficient in tabular form. In some embodiments, after the report is populated, it is output as a document in a specified format, such as Portable Document Format or Hypertext Markup Language, and stored in a preset report archive directory. Simultaneously, the report generation process sends a notification to the associated management system indicating that the report generation is complete. This notification includes the report storage path, device identifier, and report time range.

[0109] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for intelligent energy consumption assessment of photovoltaic power generation equipment based on big data, characterized in that, include: The real-time DC power generation sequence of photovoltaic power station equipment and the collected irradiance time series data are timestamped and cleaned to form string power and irradiance data pairs; Based on string power and irradiance data pairs, a standardized power conversion curve for each photovoltaic string is constructed using a regression analysis algorithm; The spatial cloud movement trajectory is extracted from the sky cloud cover change information released synchronously by the meteorological system. Combined with the equipment start-up and shutdown signals in the power consumption records of the power station auxiliary equipment, the cloud cover obstruction events and equipment start-up and shutdown load change events that affect the real-time DC power generation sequence are identified. The identified cloud cover events and the equipment start-up and shutdown load change events are used as external interference factors and input into a pre-trained power decomposition model. The power decomposition model is used to separate the baseline performance degradation component caused by the aging of the equipment itself, the instantaneous fluctuation component caused by cloud cover, and the internal friction component caused by the load change of auxiliary equipment from the real-time DC power generation sequence. Using the baseline performance degradation component output by the power decomposition model, the theoretical power generation efficiency of the photovoltaic power generation equipment under the condition of no external interference is calculated, and the theoretical power generation efficiency is compared with the theoretical power predicted by the standardized power conversion curve of the corresponding photovoltaic string to obtain the efficiency deviation coefficient characterizing the health of the equipment.

2. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 1, characterized in that, The process of aligning and cleaning the real-time DC power generation sequence of the photovoltaic power station equipment with the collected irradiance time-series data to form string power and irradiance data pairs includes: The time stamp formats of the real-time DC power generation sequence and the irradiance time series data are respectively parsed and uniformly converted into millisecond-level timestamps under the standard time zone. Based on the timeline, a unified time index sequence containing all data collection time points is established; For each time point in the unified time index sequence, find and match data points in the real-time DC power generation sequence and the irradiance time series data that are within a preset time tolerance range to form candidate data pairs; For each candidate data pair, check whether its power value and irradiance value are within the preset physical reasonable threshold range, and mark the candidate data pairs that exceed the physical reasonable threshold range as anomalies; All candidate data pairs marked as outliers are removed, and the remaining valid candidate data pairs are arranged in chronological order to form string power and irradiance data pairs.

3. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 1, characterized in that, The process of constructing a standardized power conversion curve for each photovoltaic string using regression analysis algorithms based on string power and irradiance data pairs includes: The subset of samples with irradiance higher than the clear weather threshold and no external interference records in the string power and irradiance data pairs are selected as training data. On the training data, with irradiance as the independent variable and real-time DC power generation as the dependent variable, a quadratic polynomial curve is fitted using the least squares method. Calculate the goodness of fit of the quadratic polynomial curve. If the goodness of fit is lower than the predetermined standard, add higher-order terms or perform piecewise fitting on the training data until the goodness of fit reaches the standard, and obtain the preliminary power conversion model. The preliminary power conversion model is compensated and corrected for the temperature effect, which is calculated by using the ambient temperature data recorded by the environmental monitoring station and the theoretical temperature coefficient of the photovoltaic string. The model after temperature compensation correction is determined as the standardized power conversion curve of the photovoltaic string. The standardized power conversion curve is stored in function form and describes the mapping relationship between irradiance and theoretical power generation under ideal conditions.

4. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 1, characterized in that, The process involves extracting the spatial cloud movement trajectory from the cloud cover change information synchronously released by the meteorological system, combining it with the equipment start-up and shutdown signals in the power consumption records of the power station's auxiliary equipment, and identifying cloud obstruction events affecting the real-time DC power generation sequence, including: The cloud cover change information is analyzed in consecutive frames, and each frame contains cloud type encoding and spatial coordinate information for different cloud areas; Between consecutive frames, feature point matching and tracking are performed on cloud areas with similar cloud type encoding and spatial coordinates to form multiple cloud movement trajectory line segments; Based on the direction and speed of the cloud movement trajectory line segment, the Kalman filter algorithm is used to predict its subsequent spatial position and generate a complete cloud movement trajectory. Calculate the ground projection area covered by the cloud movement trajectory, and perform spatial intersection analysis between the ground projection area and the electronic fence map of the photovoltaic power station to obtain the expected time period of the cloud passing through the power station and the range of the affected sub-area. By combining the sudden drop and recovery patterns of the irradiance time series data, the time period during which the cloud is expected to pass over the power station is verified, and the verified time period and sub-region range are marked as the cloud occlusion event.

5. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 1, characterized in that, The method of identifying equipment start-up and shutdown load change events affecting the real-time DC power generation sequence by combining equipment start-up and shutdown signals from the power consumption records of power station auxiliary equipment includes: Acquire power consumption records of power plant auxiliary equipment, the power consumption records including multi-channel current and voltage timing data of each equipment and status change signals actively reported by the equipment controller; Extract the state change signal from the power consumption record. The state change signal includes at least the device start time, the shutdown time, and the device operating mode switching time. Use the device start time, shutdown time, and device operating mode switching time as the initial time markers of the device start-up and shutdown events. Within the neighborhood time window of the initial time stamp, the multi-channel current and voltage timing data are parsed, and the rising and falling edges of the current or power are identified by the threshold judgment method, thereby locating the actual load connection or disconnection time of the device, and correcting the time point to the final device start-stop signal timestamp. Based on the timestamp of the equipment start-stop signal, locate the corresponding time in the real-time DC power generation sequence, extract power data segments of a preset duration before and after the timestamp of the equipment start-stop signal, and calculate the mean change of the power data segments; The mean change of the power data segment is compared with a preset load power impact threshold for different equipment types. If the mean change exceeds the corresponding load power impact threshold, it is determined that a device start-stop load change event has occurred, and the start time, end time, affected power change amount, and associated equipment identifier of the device start-stop load change event are recorded.

6. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 1, characterized in that, The training and output of the power decomposition model include: Collect multi-dimensional raw operating data from historical operating cycles as a training set. The training set has been labeled with time period tags for known baseline performance degradation events, cloud cover events, and equipment start-up and shutdown load change events. Construct a decomposition model containing a multi-layer neural network. The input of the model is the real-time DC power generation sequence, irradiance time series data and equipment status signal within a specified time window. The output of the model is the estimated value sequence of the baseline performance degradation component, the instantaneous fluctuation component and the internal friction component. The decomposition model is trained under supervision using the training set, and the error between the model output estimate and the true component corresponding to the known event label is minimized by the backpropagation algorithm. After training, the network parameters of the decomposition model are fixed, and it is deployed to an online evaluation system. The real-time DC power generation sequence, irradiance time series data, and equipment status signal to be analyzed are input into the deployed decomposition model, which directly outputs the separated baseline performance degradation component, instantaneous fluctuation component, and internal friction component.

7. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 6, characterized in that, The calculation of the theoretical power generation efficiency of the photovoltaic power generation equipment under the condition of no external interference includes: Extract a time-continuous and stationary data segment from the baseline performance degradation component output by the decomposition model. Calculate the average power value of the data segment and use it as the actual benchmark output power of the photovoltaic power generation equipment under the current healthy state; From the standardized power conversion curve, the average value of the irradiance data within the same time period is taken and input into the standardized power conversion curve to calculate the theoretical expected power of the photovoltaic string under the current environmental conditions. Divide the actual benchmark output power by the theoretical expected power to obtain the theoretical power generation efficiency of the photovoltaic power generation equipment under the current state; The curve of the theoretical power generation efficiency changing over time is continuously calculated and recorded.

8. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 7, characterized in that, The efficiency deviation coefficient, which characterizes the health of the equipment, includes: Obtain the theoretical power generation efficiency of the photovoltaic string during the stable operation phase after initial commissioning, and use it as the benchmark efficiency; Extract the average theoretical power generation efficiency within the most recent statistical period from the currently calculated theoretical power generation efficiency change curve; Calculate the percentage relative deviation between the average theoretical power generation efficiency and the benchmark efficiency; The relative deviation percentage is compared with multiple preset aging level thresholds to determine the current aging level of the device; The relative deviation percentage is combined with the aging level and encoded to generate an efficiency deviation coefficient that characterizes the current health of the device.

9. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 1, characterized in that, The method also includes a device-level energy consumption attribution step based on energy flow tracing: Using the baseline performance degradation component, the instantaneous fluctuation component, and the internal friction component output by the power decomposition model, the theoretical power generation curve, the external disturbance loss power curve, and the auxiliary system power consumption curve of the photovoltaic power generation equipment in any time period are reconstructed. At the power station level, the power consumption curves of the auxiliary systems of all equipment are summed to obtain the total auxiliary energy consumption curve of the power station, and cross-validated with the mains power consumption record or the fuel consumption record of the standby generator set input to the power station. At the string level, the theoretical power generation curve is compared with the external disturbance loss power curve to calculate the average energy loss rate caused by identifiable external factors. An energy consumption attribution report is generated for each photovoltaic string. The report includes at least the statistical characteristics of the theoretical power generation curve, the time distribution pattern of the external disturbance loss power curve, the composition of the auxiliary system power consumption curve, and the calculated average energy loss rate.

10. The intelligent energy consumption assessment method for photovoltaic power generation equipment based on big data according to claim 9, characterized in that, The process of generating an energy consumption attribution report for each photovoltaic string includes: Set a fixed report generation cycle, and automatically trigger the report generation process at the end of the cycle; During the report generation period, the corresponding theoretical power generation curve, external disturbance loss power curve, auxiliary system power consumption curve, average energy loss rate, and efficiency deviation coefficient are collected. According to a predefined data structure, the collected theoretical power generation curve, external disturbance loss power curve, auxiliary system power consumption curve, average energy loss rate, and efficiency deviation coefficient are integrated to form a structured data block. The structured data blocks are populated into a pre-designed report template, which includes a time range, device identification, visualization charts of each power curve, a summary of key indicators, and a historical trend comparison area. After the data is populated, the report will be output as a document in the specified format and stored in the preset report archive directory. At the same time, a notification that the report generation is complete will be sent to the associated management system.

Images