Distributed photovoltaic energy efficiency dynamic monitoring method based on time sequence analysis

Abstract

The invention provides a distributed photovoltaic energy efficiency dynamic monitoring method based on time sequence analysis, and relates to the technical field of photovoltaic power generation, and the method comprises the steps: deploying an edge calculation node, regularly receiving the electrical operation parameters of a photovoltaic inverter, and recording a monitoring node; predicting photovoltaic output power according to the electrical operation parameters, and obtaining predicted output power; analyzing the power supply log, obtaining actual output power under the monitoring node, forming energy efficiency analysis data based on the monitoring node, the predicted output power and the actual output power, and storing the energy efficiency analysis data in an energy efficiency analysis database; performing power prediction deviation analysis by using the energy efficiency analysis database, and constructing a time period-deviation comparison table; and when grid-connected scheduling is carried out on the distributed photovoltaic equipment, power output compensation is carried out based on the time period-deviation comparison table. The technical problem that the power prediction precision is insufficient due to the fact that a traditional time sequence analysis method lacks accurate analysis for different time periods is solved, and the power prediction accuracy and the grid-connected dispatching quality are improved.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic power generation technology, and specifically to a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis. Background Art

[0002] Distributed photovoltaic systems, as an important component of clean energy, are increasingly becoming a key driving force for global energy transformation. Distributed photovoltaic systems can generate electricity directly at the user end, reduce transmission losses, and support grid-connected power generation, optimizing the energy structure of the power system. However, photovoltaic power generation is easily affected by factors such as weather and time, and the power generation fluctuates greatly. Accurately predicting the power output of distributed photovoltaics is not only crucial for optimizing the operating efficiency of the photovoltaic system itself, but also a key step in ensuring the stable and efficient operation of the power system.

[0003] Time series analysis methods are often used in dynamic monitoring of distributed photovoltaic energy efficiency. The trend of photovoltaic output power changes over time is analyzed through historical data, and future power forecasts and grid-connected scheduling are performed based on these trends. However, the current time series analysis methods mainly focus on the prediction of macro trends and lack refined analysis of power fluctuations in short time periods. This rough analysis of time periods makes it difficult to capture the immediate impact of rapidly changing weather conditions on photovoltaic power, which in turn affects the accuracy of power forecasts. In grid-connected scheduling, relying on such rough predictions to adjust power output not only fails to respond to grid demand in a timely manner, but may also cause grid fluctuations due to large deviations between predictions and actual outputs, affecting the overall operation quality and stability of the grid. Summary of the invention

[0004] The present application provides a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis, which solves the technical problem that traditional time series analysis methods usually perform power forecasting based on historical data and macro trends, lack precise analysis for different time periods, resulting in insufficient power forecasting accuracy, which in turn affects the accuracy of grid-connected scheduling. The application achieves the technical effect of improving the accuracy of photovoltaic power forecasting to improve the quality of grid-connected scheduling and enhance the stability of grid operation.

[0005] In view of the above problems, the present application provides a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis, the method comprising: deploying edge computing nodes at distributed photovoltaic devices, regularly receiving electrical operating parameters of photovoltaic inverters, and recording monitoring nodes; predicting photovoltaic output power according to the electrical operating parameters to obtain predicted output power; analyzing power supply logs to obtain the actual output power under the monitoring node, and forming energy efficiency analysis data based on the monitoring node, predicted output power and actual output power, and storing them in an energy efficiency analysis database, wherein the energy efficiency analysis database is embedded in the edge computing node; using the energy efficiency analysis database, performing power prediction deviation analysis based on multiple energy efficiency analysis data, and constructing a time period-deviation comparison table; when the distributed photovoltaic device is connected to the grid for scheduling, performing power output compensation based on the time period-deviation comparison table.

[0006] Preferably, the electrical operating parameters include at least a DC input voltage, a DC input current, a power factor and an inverter temperature, wherein the inverter temperature is monitored by a temperature sensor deployed in the photovoltaic inverter.

[0007] Preferably, photovoltaic output power is predicted according to the electrical operating parameters to obtain the predicted output power, including: taking the equipment attribute characteristics and equipment service life of the photovoltaic inverter as equipment constraints, taking the electrical operating parameters as keyword constraints, and taking the output power prediction as a guide, obtaining a sample data set based on power grid big data retrieval; performing differential screening on the sample data set according to a predetermined strategy to obtain a sample training set, and using the sample training set to supervise the BP neural network to obtain a power predictor; using the power predictor to predict the photovoltaic output power of the electrical operating parameters to obtain the predicted output power.

[0008] Preferably, the sample data set is differentially screened according to a predetermined strategy to obtain a sample training set, including: randomly selecting first sample data from the sample data set, wherein the first sample data includes a first voltage, a first current, a first power factor, a first temperature and a first power; performing deviation analysis on the first voltage and other voltages in the sample data set respectively, and statistically analyzing the amount of data with a deviation greater than a predetermined voltage threshold, which is recorded as a first random coefficient; if the first random coefficient is greater than a coefficient scalar, performing deviation analysis on the first current, the first power factor and the first temperature in turn to obtain a second random coefficient, a third random coefficient and a fourth random coefficient; if the second random coefficient, the third random coefficient and the fourth random coefficient are all greater than the coefficient scalar, adding the first sample data to the sample training set.

[0009] Preferably, power prediction deviation analysis is performed based on multiple energy efficiency analysis data to construct a time period-deviation comparison table, including: dividing the multiple energy efficiency analysis data according to a predetermined period, time-aligning and extracting data on the division results, and obtaining multiple energy efficiency data sets under multiple identical nodes, wherein the predetermined period is 24 hours; based on the multiple energy efficiency data sets, power prediction deviation analysis is performed on multiple monitoring nodes respectively to determine multiple prediction deviation coefficients; and the time period-deviation comparison table is constructed based on the multiple monitoring nodes and the multiple prediction deviation coefficients.

[0010] Preferably, multiple prediction deviation coefficients are determined, including: randomly selecting a first monitoring node and a first energy efficiency data set of the first monitoring node; performing power prediction deviation calculation based on the first energy efficiency data set to obtain multiple first prediction deviations, wherein the first prediction deviation is the ratio of the difference between the actual output power and the predicted output power to the actual output power; performing mean calculation on the multiple first prediction deviations to obtain a first prediction deviation coefficient, and adding it to the multiple prediction deviation coefficients.

[0011] Preferably, the time period-deviation comparison table is constructed according to the multiple monitoring nodes and multiple prediction deviation coefficients, including: arranging the multiple monitoring nodes in node order to construct a monitoring node sequence; selecting the first deviation coefficient of the first node and the second deviation coefficient of the second node, and calculating the first coefficient difference between the first deviation coefficient and the second deviation coefficient, wherein the first node is the initial node of the monitoring node sequence, and the second node is the adjacent node of the first node; if the first coefficient difference is less than a predetermined coefficient threshold, the first node and the second node are merged to construct a first time period, wherein the deviation coefficient of the first time period is the average of the first deviation coefficient and the second deviation coefficient; if the first coefficient difference is greater than or equal to the predetermined coefficient threshold, a first node dividing line is set, and iterative analysis is performed until all nodes of the monitoring node sequence are analyzed and multiple node dividing lines are obtained; based on the multiple node dividing lines, multiple monitoring time periods and multiple deviation coefficients corresponding to the multiple monitoring time periods are determined, and the time period-deviation comparison table is constructed according to the multiple monitoring time periods and multiple deviation coefficients.

[0012] Preferably, power output compensation is performed based on the time period-deviation comparison table, including: receiving real-time electrical operating parameters of the current node, using the power predictor to predict the photovoltaic output power of the real-time electrical operating parameters to obtain real-time predicted output power; inputting the current node into the time period-deviation comparison table for matching, and outputting an adaptation deviation coefficient; compensating the real-time predicted output power according to the adaptation deviation coefficient to obtain the real-time output power, and executing grid-connected scheduling of the distributed photovoltaic equipment.

[0013] One or more technical solutions provided in this application have at least the following technical effects or advantages: By deploying edge computing nodes at distributed photovoltaic devices, regularly receiving the electrical operating parameters of photovoltaic inverters, and recording monitoring nodes, localized data processing, reducing data transmission delays and costs, improving the real-time and response speed of data processing, and providing fast and accurate basic data for subsequent power forecasting. Photovoltaic output power is predicted according to the electrical operating parameters, the predicted output power is obtained, and the power demand and power generation conditions are predicted, providing data reference for subsequent deviation analysis and compensation. The power supply log is analyzed to obtain the actual output power under the monitoring node, and energy efficiency analysis data is composed based on the monitoring node, predicted output power and actual output power. The data is stored in the energy efficiency analysis database to monitor the operating status of distributed photovoltaic devices, providing complete and continuous data for subsequent deviation analysis. Using the energy efficiency analysis database, power prediction deviation analysis is performed based on multiple energy efficiency analysis data, a time period-deviation comparison table is constructed, prediction error characteristics of different time periods are captured, power error analysis is refined, and power fluctuations in a short period of time are identified, providing a key basis for subsequent power output compensation. When the distributed photovoltaic equipment is connected to the grid, power output compensation is performed based on the time period-deviation comparison table to eliminate the impact of prediction errors on grid-connected power generation, so as to ensure the stability of the power system, reduce power fluctuations during the grid connection process, and improve the quality and stability of grid-connected scheduling.

[0014] In summary, this application combines edge computing, power prediction and actual power monitoring to accurately capture power prediction errors in different time periods, and through dynamic output power compensation, it reduces grid fluctuations caused by power prediction errors. This solution significantly improves the accuracy and real-time performance of dynamic monitoring of energy efficiency of distributed photovoltaic systems, improves the flexibility and stability of grid-connected scheduling, enhances the ability of distributed photovoltaic power generation systems to respond quickly to grid demand, and ensures the efficient operation of distributed photovoltaic systems.

[0015] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a process flow of a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis provided in an embodiment of the present application; Figure 2 A schematic diagram of a process for obtaining predicted output power in a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis provided in an embodiment of the present application; Figure 3A schematic diagram of a process for constructing a time period-deviation comparison table in a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application provide a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis, which solves the technical problem that traditional time series analysis methods usually perform power forecasts based on historical data and macro trends, lack precise analysis for different time periods, resulting in insufficient power forecast accuracy, which in turn affects the accuracy of grid-connected scheduling. The technical effect of improving the accuracy of photovoltaic power prediction to improve the quality of grid-connected scheduling and enhance the stability of grid operation is achieved.

[0018] like Figure 1 As shown, the embodiment of the present application provides a distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis, the method comprising: Step S1: deploy edge computing nodes at distributed photovoltaic equipment, regularly receive electrical operating parameters of photovoltaic inverters, and record monitoring nodes.

[0019] Specifically, an edge computing node is a computing infrastructure deployed near a data source, responsible for real-time processing, analysis and storage of data, thereby reducing data transmission delays. Compared with cloud computing, edge computing can process data nearby and reduce response time. In an embodiment of the present application, an edge computing device may be a small computing unit installed in a photovoltaic system, such as a single-board computer or a customized industrial computing device. First, multiple edge computing nodes are installed and deployed near distributed photovoltaic equipment. These edge computing nodes regularly receive the electrical operating parameters provided by the photovoltaic inverter and mark the monitoring time points. Among them, the photovoltaic inverter is responsible for converting the direct current generated by the photovoltaic module into alternating current, and its electrical parameters include voltage, current, power factor, inverter temperature, etc.

[0020] Through step S1, data can be quickly collected and analyzed locally to reduce dependence on remote cloud platforms, reduce data transmission delays, and provide real-time and accurate data support for subsequent photovoltaic power prediction and energy efficiency analysis.

[0021] Step S2: predicting the photovoltaic output power according to the electrical operating parameters to obtain the predicted output power.

[0022] Specifically, the prediction of photovoltaic output power is mainly based on historical data training prediction models, such as linear regression, random forest, neural network, etc. The previously acquired electrical operation parameters are input into the pre-trained prediction model to obtain the predicted output power for a period of time in the future, and the power generation capacity of photovoltaic equipment is estimated in advance to provide reference data for subsequent grid-connected scheduling.

[0023] Step S3: Analyze the power supply log, obtain the actual output power of the monitoring node, form energy efficiency analysis data based on the monitoring node, predicted output power and actual output power, and store them in an energy efficiency analysis database, wherein the energy efficiency analysis database is embedded in the edge computing node.

[0024] Specifically, the power supply log records the power supply data during the operation of the photovoltaic power generation system, including information such as actual output power and timestamp. By analyzing the power supply log, the actual output power data under the same monitoring node as the output power data predicted in step S2 is extracted. Each monitoring node is combined with its corresponding predicted output power and actual output power to form a set of energy efficiency analysis data. The energy efficiency analysis data not only contains the predicted and actual output power, but also includes auxiliary information such as timestamp and device status, providing comprehensive data support for subsequent energy efficiency analysis. These data reflect the performance of the photovoltaic system and the accuracy of the prediction model. These energy efficiency analysis data are stored in the energy efficiency analysis database, which is embedded in the edge computing node to facilitate data processing and storage near the device, and timely response, analysis and adjustment of power generation strategies. The energy efficiency analysis database can not only store data for a single period, but also accumulate long-term historical records to help with more accurate power prediction and deviation analysis in the future.

[0025] Step S4: using the energy efficiency analysis database, performing power prediction deviation analysis based on a plurality of energy efficiency analysis data, and constructing a time period-deviation comparison table.

[0026] Specifically, energy efficiency analysis data of multiple monitoring nodes are extracted from the energy efficiency analysis database, which includes predicted output power and actual output power. The extracted data are grouped according to time information (such as hours, days, months, etc.) to analyze the prediction deviations in different time periods. For each time period, the deviation between the predicted output power and the actual output power is calculated. The calculated deviations are sorted according to time periods to construct a comparison table, namely the time period-deviation comparison table, which records the power prediction deviations in different time periods and provides a reference for power generation scheduling and compensation decisions. The data in the comparison table is analyzed to identify the regularity and periodicity of the deviations. Through systematic deviation analysis, the power prediction deviation can be more accurately identified and quantified, providing an important basis for optimizing the grid-connected scheduling of photovoltaic systems, reducing power fluctuations caused by prediction errors, and improving overall power generation efficiency and grid stability.

[0027] Step S5: When the distributed photovoltaic equipment is connected to the grid, power output compensation is performed based on the time period-deviation comparison table.

[0028] Specifically, when distributed photovoltaic equipment is connected to the grid, power output compensation refers to adjusting the output power of photovoltaic equipment according to the real-time demand of the power grid and the characteristics of distributed photovoltaic power generation to ensure the stability of the power grid and the quality of power. This compensation mechanism can reduce the volatility and uncertainty of photovoltaic power generation and avoid impact on the power grid.

[0029] When conducting grid-connected dispatching, the load demand, frequency, voltage and other key parameters of the grid are monitored in real time to timely understand the operation status of the grid. The time period-deviation comparison table constructed in step S4 is used to query the corresponding prediction deviation coefficient according to the time period of the current time. The output power of the photovoltaic equipment is adjusted according to the deviation coefficient to compensate for the prediction deviation and ensure that photovoltaic power generation is more matched with grid demand. Through this compensation mechanism, the accuracy and reliability of grid-connected dispatching can be greatly improved, and stable power output can be achieved.

[0030] Furthermore, the electrical operating parameters in step S1 of the embodiment of the present application at least include a DC input voltage, a DC input current, a power factor and an inverter temperature, wherein the inverter temperature is monitored by a temperature sensor deployed in the photovoltaic inverter.

[0031] Specifically, in step S1, the electrical operating parameters to be monitored include at least DC input voltage, DC input current, power factor and inverter temperature. DC input voltage and DC input current are core indicators of the photovoltaic system, which directly reflect the power generated by the solar panels. The fluctuations in voltage and current can indicate the working status of the photovoltaic panels. These parameters may change dramatically when the weather changes or the equipment fails. For example, on a sunny day, the photovoltaic panels may generate higher voltage and current. On the contrary, the voltage and current will be significantly reduced on cloudy days. The power factor is used for the power utilization efficiency of distributed photovoltaic power generation systems. If the power factor is reduced, the inverter may have reactive power consumption, which needs to be adjusted to improve the overall power generation efficiency.

[0032] The inverter temperature is an important indicator of stable operation of the equipment. Excessive temperature may cause the inverter performance to degrade or even damage. Therefore, the temperature sensor deployed on the inverter is used to obtain the inverter temperature in real time to detect the operating status of the photovoltaic equipment. The temperature sensor can measure the temperature of the inverter in real time through the thermistor integrated in the inverter or the temperature probe installed externally. When the operating temperature of the inverter exceeds a certain threshold, it may be necessary to take cooling measures or reduce the power output to avoid equipment damage.

[0033] For example, a distributed photovoltaic system installed on the roof of a factory has an inverter with a DC input voltage of 400V, a DC current of 12A, a power factor of 0.98, and an inverter temperature of 60°C. In this state, the photovoltaic system is operating normally and efficiently. However, if the inverter temperature suddenly rises to 80°C, the photovoltaic system may activate an automatic protection mechanism to reduce power output to avoid overheating.

[0034] Further, such as Figure 2 As shown, step S2 of the embodiment of the present application also includes: Taking the equipment attribute characteristics and equipment service life of the photovoltaic inverter as equipment constraints, taking the electrical operating parameters as keyword constraints, and taking the output power prediction as a guide, a sample data set is obtained based on the power grid big data retrieval; the sample data set is differentially screened according to a predetermined strategy to obtain a sample training set, and the sample training set is used to perform supervised training on the BP neural network to obtain a power predictor; the power predictor is used to predict the photovoltaic output power of the electrical operating parameters to obtain the predicted output power.

[0035] Specifically, equipment attribute characteristics refer to the basic technical parameters and performance indicators of photovoltaic inverters, such as rated power, efficiency, rated input voltage and current, etc. These characteristics help identify the type and scope of application of the inverter. Equipment age refers to the use time of the photovoltaic inverter since its installation. This factor affects the performance and efficiency of the inverter, and older equipment is generally less efficient. First, the photovoltaic inverter is screened in the power grid big data according to its equipment attribute characteristics and age. Then, the electrical operating parameters obtained above are used as keyword constraints to retrieve sample data sets under similar equipment and similar operating parameters from the power grid big data. The sample data set contains similar equipment attribute characteristics and age, and historical output power data under different electrical operating parameters, which is used for power predictor training and testing.

[0036] Then, according to the predetermined strategy, the sample data set is differentially screened to determine the data set for the power predictor, namely the sample training set. The BP neural network is supervised and trained using these sample training sets. The BP neural network is a machine learning model that can learn complex nonlinear relationships. Through the back propagation algorithm, the network continuously adjusts its internal weights to improve the ability to predict photovoltaic power generation. During the training process, the structure of the BP neural network (input layer, hidden layer, output layer) is set, the weights and biases are randomly initialized, the historical electrical operating parameters are input, and the output power is predicted through neural network calculation. The predicted value is compared with the actual value, and the loss function is calculated. According to the loss function, the weights and biases are adjusted in reverse to reduce the prediction error. Repeat the forward and backward propagation process until the model converges.

[0037] The trained BP neural network is a power predictor, which is used to predict the photovoltaic output power based on the input electrical operating parameters. The predictor can be applied to new electrical operating parameters in real time to predict the photovoltaic output power. When new electrical parameters are input, the power predictor can quickly calculate the expected photovoltaic output power under the current conditions.

[0038] Through the above steps, based on the sample data set of equipment constraints and keyword constraints, combined with the supervised training of BP neural network, a high-precision power predictor can be obtained to provide accurate photovoltaic output efficiency prediction data.

[0039] Furthermore, in step S2 of the embodiment of the present application, the sample data set is differentially screened according to a predetermined strategy to obtain a sample training set, including: A first sample data is randomly selected from the sample data set, wherein the first sample data includes a first voltage, a first current, a first power factor, a first temperature and a first power; a deviation analysis is performed on the first voltage and other voltages in the sample data set respectively, and the amount of data with a statistical deviation greater than a predetermined voltage threshold is recorded as a first random coefficient; if the first random coefficient is greater than a coefficient scalar, deviation analysis is performed on the first current, the first power factor and the first temperature in turn to obtain a second random coefficient, a third random coefficient and a fourth random coefficient; if the second random coefficient, the third random coefficient and the fourth random coefficient are all greater than the coefficient scalar, the first sample data is added to the sample training set.

[0040] Specifically, before training the model, the sample data set needs to be differentially screened to improve the diversity of the training data, thereby saving training time, enabling the BP neural network to converge quickly, and improving the training quality. First, a set of first sample data is randomly selected from the sample data set, and the data includes electrical operating parameters such as a first voltage, a first current, a first power factor, a first temperature, and a first power.

[0041] Next, the first voltage is subjected to deviation analysis: the first voltage of the first sample data is compared with all other voltage values ​​in the sample data set, and the amount of data with a deviation greater than a predetermined voltage threshold is counted, and this amount of data is called the first random coefficient. The predetermined voltage threshold is a pre-set acceptable voltage deviation range, usually a specific value, and data exceeding this range will be considered to have a large deviation. For example, if the current first voltage is 350V and the predetermined voltage threshold is 10V, all voltage values ​​in the sample data set are searched, and the number of voltage data with a difference of more than 10V from 350V is counted, and this number is used as the first random coefficient.

[0042] If the first random coefficient is greater than the preset coefficient scalar, the same deviation analysis is performed on other electrical parameters. First, the first current is subjected to deviation analysis to obtain the second random coefficient. If the second random coefficient is greater than the preset coefficient scalar, the power deviation analysis is continued. The first power factor is subjected to deviation analysis to obtain the third random coefficient; if the third random coefficient is greater than the preset coefficient scalar, the temperature deviation analysis is continued. The first temperature is subjected to deviation analysis to obtain the fourth random coefficient. If the fourth random coefficient is greater than the preset coefficient scalar, the first sample data is added to the sample training set. The above deviation analysis process is similar to the voltage deviation analysis, and can be referred to as the voltage deviation analysis. Among them, the second random coefficient, the third random coefficient, and the fourth random coefficient correspond to the deviation analysis results of the first current, the first power factor, and the first temperature, respectively, and are used to comprehensively evaluate the overall stability of the current sample. The coefficient scalar is a preset value used to compare with the random coefficient to determine whether the sample meets the conditions for inclusion in the training set.

[0043] If any of the first random coefficient, the second random coefficient, the third random coefficient, and the fourth random coefficient is less than or equal to the system scalar, the deviation analysis is terminated and the first sample data is removed from the data set. A new set of sample data is selected and the above steps are repeated until all the sample data in the sample data set are traversed to finally obtain the sample training set.

[0044] Through the above-mentioned differentiated screening strategy, the diversity and representativeness of the sample training set can be ensured, the overfitting problem during model training can be avoided, the generalization ability and prediction accuracy of the model can be improved, and high-quality training data can be provided for the subsequent power predictor training based on BP neural network.

[0045] Further, such as Figure 3 As shown, step S4 of the embodiment of the present application also includes: The multiple energy efficiency analysis data are divided according to a predetermined period, and the division results are time-aligned and data extracted to obtain multiple energy efficiency data sets under multiple identical nodes, wherein the predetermined period is 24 hours; based on the multiple energy efficiency data sets, power prediction deviation analysis of multiple monitoring nodes is performed respectively to determine multiple prediction deviation coefficients; and the time period-deviation comparison table is constructed based on the multiple monitoring nodes and the multiple prediction deviation coefficients.

[0046] Specifically, firstly, multiple energy efficiency analysis data are divided according to a predetermined cycle, i.e., 24 hours, and the daily photovoltaic power generation data is sorted and processed as an independent cycle. The divided energy efficiency analysis data are time-aligned according to the detection nodes to avoid analysis errors caused by time deviations and ensure that data from different time periods are consistent and comparable in subsequent analysis. Subsequently, energy efficiency data of specific nodes are extracted from each time cycle, including predicted output power and actual output power, to form multiple energy efficiency data sets. For example, energy efficiency data at 9:00 in a consecutive week are extracted to obtain 7 energy efficiency data sets.

[0047] For each monitoring node, multiple energy efficiency data sets extracted from the database are used to perform power prediction deviation analysis. The deviation between the predicted output power and the actual output power is calculated. Deviation calculation can be performed in a variety of ways, such as absolute deviation, relative deviation, or root mean square error. Based on the deviation statistics, the prediction deviation coefficient of each monitoring node is determined. The prediction deviation coefficient reflects the consistency level between the predicted output and the actual output of the node, and can be used as a quantitative indicator to evaluate the prediction performance of the model.

[0048] The prediction deviation coefficients of all monitoring nodes are summarized and associated with the corresponding time period information. Based on the summary information, a time period-deviation comparison table is constructed. The table clearly shows the deviation between the predicted output power and the actual output power at different monitoring nodes and different time periods. By analyzing the comparison table, the monitoring nodes and time periods with large deviations can be identified, and then targeted scheduling compensation can be performed.

[0049] Furthermore, in step S4 of the embodiment of the present application, multiple prediction deviation coefficients are determined, including: Randomly select a first monitoring node and a first energy efficiency data set of the first monitoring node; perform power prediction deviation calculation based on the first energy efficiency data set to obtain multiple first prediction deviations, wherein the first prediction deviation is the ratio of the difference between the actual output power and the predicted output power to the actual output power; perform mean calculation on the multiple first prediction deviations to obtain a first prediction deviation coefficient, and add it to the multiple prediction deviation coefficients.

[0050] Specifically, a node is randomly selected from multiple monitoring nodes as a first monitoring node, and multiple energy efficiency data sets of the node are extracted as first energy efficiency data sets, each data set including predicted output power and actual output power data of the node in different periods.

[0051] For each data set in the first energy efficiency data set, the power prediction deviation is calculated according to the formula: first prediction deviation = (actual output power - predicted output power) / actual output power, and multiple first prediction deviations are obtained. Then, the average values ​​of multiple first prediction deviations are calculated to obtain a value representing the overall prediction accuracy of the node, namely the first prediction deviation coefficient. If the coefficient value is high, it means that the prediction error of the node is large and the prediction accuracy of the model is low; conversely, if the coefficient value is low, it means that the prediction of the node is more accurate. This first prediction deviation coefficient is added to multiple prediction deviation coefficient sets. This set contains the prediction deviation coefficients of multiple monitoring nodes, which are used for subsequent system optimization and scheduling decisions.

[0052] Through the above steps, not only specific quantitative indicators of prediction deviation are provided for the selected nodes, but also data support is provided for comparing the prediction performance of different monitoring nodes in subsequent analysis, which helps to identify potential improvement points of the prediction model and optimize the operation strategy of the PV system.

[0053] Furthermore, in step S4 of the embodiment of the present application, the time period-deviation comparison table is constructed according to the multiple monitoring nodes and the multiple prediction deviation coefficients, including: Arrange the multiple monitoring nodes in node sequence to construct a monitoring node sequence; select a first deviation coefficient of the first node and a second deviation coefficient of the second node, and calculate a first coefficient difference between the first deviation coefficient and the second deviation coefficient, wherein the first node is the initial node of the monitoring node sequence, and the second node is an adjacent node of the first node; if the first coefficient difference is less than a predetermined coefficient threshold, merge the first node and the second node to construct a first time period, wherein the deviation coefficient of the first time period is the average of the first deviation coefficient and the second deviation coefficient; if the first coefficient difference is greater than or equal to the predetermined coefficient threshold, set a first node dividing line, and perform iterative analysis until all nodes of the monitoring node sequence are analyzed, and obtain multiple node dividing lines; determine multiple monitoring time periods based on the multiple node dividing lines, and multiple deviation coefficients corresponding to the multiple monitoring time periods, and construct the time period-deviation comparison table according to the multiple monitoring time periods and the multiple deviation coefficients.

[0054] Specifically, a monitoring node sequence is formed according to a list of monitoring nodes arranged in order of the nodes. The first node (i.e., the initial node) and its adjacent second node in the sequence are selected, and the difference between the first deviation coefficient of the first node and the second deviation coefficient of the second node is calculated, which is recorded as the first coefficient difference. If the first coefficient difference is less than the predetermined coefficient threshold, the first node and the second node are merged into the same time period, which is recorded as the first time period, and the deviation coefficient of the time period is the average value of the first deviation coefficient and the second deviation coefficient. Among them, the predetermined coefficient threshold is a set critical value used to determine whether the deviation coefficient difference between two monitoring nodes is close enough.

[0055] If the first coefficient difference is greater than or equal to the predetermined coefficient threshold, a node dividing line is set between the first node and the second node to mark the end of the analysis of the first node. Subsequently, the second node is used as the new starting node, and the same steps as above are followed to calculate the new coefficient difference and determine whether to merge the time periods, and perform iterative analysis until all nodes in the sequence are analyzed.

[0056] Through the above iterative process, multiple node segmentation lines, multiple monitoring periods and their corresponding deviation coefficients can be determined, thereby dividing the entire monitoring node sequence into multiple periods with similar deviation characteristics. Based on the monitoring periods determined by multiple node segmentation lines and the deviation coefficients corresponding to each period, a period-deviation comparison table is constructed to intuitively display the prediction deviation in different time periods.

[0057] Exemplarily, there are five monitoring nodes numbered 001 to 005, and the corresponding prediction deviation coefficients are 0.02, 0.03, 0.04, 0.06 and 0.07, respectively, and the predetermined coefficient threshold is 0.015.

[0058] Starting from node 001 (deviation coefficient 0.02), it is compared with node 002 (deviation coefficient 0.03), and the first coefficient difference is 0.01, which is less than the threshold. Therefore, nodes 001 and 002 are merged to form the first time period, and the time period deviation coefficient is 0.025.

[0059] Next, node 002 (now the merged node of the first period, with a deviation coefficient of 0.025) is used as the subsequent node of the starting node, that is, node 003 (deviation coefficient 0.04). The calculated coefficient difference is 0.015, which is equal to the threshold, so the first period and node 003 can be merged, and the deviation coefficient of the first period is updated to (0.02+0.03+0.04) / 3=0.03.

[0060] Node 003 (now the node after merging in the first period, with a deviation coefficient of 0.03) is used as the subsequent node of the starting node, that is, node 004 (deviation coefficient 0.06) for comparison. The calculated coefficient difference is 0.03, which is greater than the threshold, so a node dividing line is set between nodes 003 and 004. Node 004 (deviation coefficient 0.06) is used as the new starting node and compared with node 005 (deviation coefficient 0.07). The coefficient difference is 0.01, which is less than the threshold. Nodes 004 and 005 are merged to form the second period. The period deviation coefficient is (0.06+0.07) / 2=0.065. Through the above steps, the five monitoring nodes are finally divided into two periods, forming a period-deviation comparison table: Furthermore, step S5 of the embodiment of the present application further includes: Receive the real-time electrical operating parameters of the current node, use the power predictor to predict the photovoltaic output power of the real-time electrical operating parameters to obtain the real-time predicted output power; input the current node into the time period-deviation comparison table for matching, and output the adaptation deviation coefficient; compensate the real-time predicted output power according to the adaptation deviation coefficient to obtain the real-time output power, and execute the grid-connected scheduling of the distributed photovoltaic equipment.

[0061] Specifically, when distributed photovoltaic equipment is connected to the grid, the real-time electrical operation parameters of the current node, including real-time DC input voltage, real-time DC input current, real-time power factor and real-time inverter temperature, are first received through the edge computing node. The real-time electrical operation parameters are input into the trained power predictor to predict the photovoltaic output power and obtain the real-time predicted output power. According to the current time node, the time period-deviation comparison table is queried to find the adaptation deviation coefficient under the corresponding time period. This coefficient is the deviation coefficient that best matches the current time period selected from the time period-deviation comparison table according to the current time period. It reflects the degree of deviation between the predicted value and the actual value of the current time period and is used to compensate for the prediction error. The real-time predicted output power is adjusted using the adaptation deviation coefficient to compensate for the prediction deviation by increasing or decreasing the output power. The adjusted power is the real-time output power. According to the real-time output power, the output of the photovoltaic equipment is adjusted to ensure that it matches the power supply and demand of the power grid, while optimizing the operation efficiency and stability of the power grid.

[0062] For example, the current time is 12:00 noon, and the real-time monitored electrical parameters show that the voltage is 380V, the current is 15A, the power factor is 0.98, and the temperature is 45°C. Input these data into the power predictor, and predict that the current power generation is 120kW. Consulting the time period-deviation comparison table, it is found that the deviation coefficient for the 12:00-13:00 time period is 3%. The distributed photovoltaic system will increase the predicted power by 3%, and the adjusted real-time output power is 123.6kW. 123.6kW is used as the power value of the current grid-connected scheduling to guide the grid connection of photovoltaic equipment and ensure the stable operation of the power grid.

[0063] Through the above steps, real-time power prediction and compensation of distributed photovoltaic equipment in grid-connected scheduling are realized, which helps to improve the accuracy of power scheduling and grid-connected quality and ensure stable operation of the power grid.

[0064] In summary, the distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis provided in the embodiments of the present application has the following technical effects: By deploying edge computing nodes at distributed photovoltaic devices, regularly receiving the electrical operating parameters of photovoltaic inverters, and recording monitoring nodes, localized data processing, reducing data transmission delays and costs, improving the real-time and response speed of data processing, and providing fast and accurate basic data for subsequent power forecasting. Photovoltaic output power is predicted according to the electrical operating parameters, the predicted output power is obtained, and the power demand and power generation conditions are predicted, providing data reference for subsequent deviation analysis and compensation. The power supply log is analyzed to obtain the actual output power under the monitoring node, and energy efficiency analysis data is composed based on the monitoring node, predicted output power and actual output power. The data is stored in the energy efficiency analysis database to monitor the operating status of distributed photovoltaic devices, providing complete and continuous data for subsequent deviation analysis. Using the energy efficiency analysis database, power prediction deviation analysis is performed based on multiple energy efficiency analysis data, a time period-deviation comparison table is constructed, prediction error characteristics of different time periods are captured, power error analysis is refined, and power fluctuations in a short period of time are identified, providing a key basis for subsequent power output compensation. When the distributed photovoltaic equipment is connected to the grid, power output compensation is performed based on the time period-deviation comparison table to eliminate the impact of prediction errors on grid-connected power generation, so as to ensure the stability of the power system, reduce power fluctuations during the grid connection process, and improve the quality and stability of grid-connected scheduling.

[0065] Overall, the embodiment of the present application combines edge computing, power prediction and actual power monitoring to accurately capture power prediction errors in different time periods, and through dynamic output power compensation, reduce grid fluctuations caused by power prediction errors. This solution significantly improves the accuracy and real-time performance of dynamic monitoring of energy efficiency of distributed photovoltaic systems, improves the flexibility and stability of grid-connected scheduling, enhances the ability of distributed photovoltaic power generation systems to respond quickly to grid demand, and ensures the efficient operation of distributed photovoltaic systems.

[0066] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis, characterized in that: include: Deploy edge computing nodes at distributed photovoltaic equipment to regularly receive electrical operating parameters of photovoltaic inverters and record monitoring nodes; Predicting photovoltaic output power according to the electrical operating parameters to obtain predicted output power; Analyze the power supply log to obtain the actual output power of the monitoring node, form energy efficiency analysis data based on the monitoring node, the predicted output power and the actual output power, and store the data in an energy efficiency analysis database, wherein the energy efficiency analysis database is embedded in the edge computing node; Utilizing the energy efficiency analysis database, performing power prediction deviation analysis based on multiple energy efficiency analysis data, and constructing a time period-deviation comparison table; When the distributed photovoltaic equipment is connected to the grid, power output compensation is performed based on the time period-deviation comparison table.

2. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 1 is characterized in that: The electrical operating parameters include at least a DC input voltage, a DC input current, a power factor and an inverter temperature, wherein the inverter temperature is monitored by a temperature sensor deployed in the photovoltaic inverter.

3. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 2 is characterized in that: Predicting photovoltaic output power according to the electrical operating parameters to obtain predicted output power includes: Taking the equipment attribute characteristics and equipment service life of the photovoltaic inverter as equipment constraints, taking the electrical operation parameters as keyword constraints, and taking the output power prediction as a guide, a sample data set is obtained based on the power grid big data retrieval; Performing differential screening on the sample data set according to a predetermined strategy to obtain a sample training set, and using the sample training set to perform supervised training on a BP neural network to obtain a power predictor; The power predictor is used to predict the photovoltaic output power of the electrical operating parameters to obtain the predicted output power.

4. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 3 is characterized in that: The sample data set is differentially screened according to a predetermined strategy to obtain a sample training set, including: Randomly selecting first sample data from the sample data set, wherein the first sample data includes a first voltage, a first current, a first power factor, a first temperature, and a first power; Performing deviation analysis on the first voltage and other voltages in the sample data set respectively, and counting the amount of data with deviations greater than a predetermined voltage threshold, which is recorded as a first random coefficient; If the first random coefficient is greater than the coefficient scalar, performing deviation analysis on the first current, the first power factor and the first temperature in sequence to obtain a second random coefficient, a third random coefficient and a fourth random coefficient; If the second random coefficient, the third random coefficient and the fourth random coefficient are all greater than the coefficient scalar, the first sample data is added to the sample training set.

5. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 1 is characterized in that: Power forecast deviation analysis is performed based on multiple energy efficiency analysis data, and a time period-deviation comparison table is constructed, including: Dividing the multiple energy efficiency analysis data according to a predetermined period, performing time alignment and data extraction on the division results, and obtaining multiple energy efficiency data sets under multiple same nodes, wherein the predetermined period is 24 hours; According to the multiple energy efficiency data sets, respectively perform power prediction deviation analysis of multiple monitoring nodes to determine multiple prediction deviation coefficients; The time period-deviation comparison table is constructed according to the multiple monitoring nodes and the multiple prediction deviation coefficients.

6. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 5 is characterized in that: Determines several forecast deviation coefficients, including: Randomly selecting a first monitoring node and a first energy efficiency data set of the first monitoring node; Calculating the power prediction deviation according to the first energy efficiency data set to obtain a plurality of first prediction deviations, wherein the first prediction deviation is a ratio of a difference between the actual output power and the predicted output power to the actual output power; The first prediction deviation coefficient is obtained by performing mean calculation on the multiple first prediction deviations, and is added to the multiple prediction deviation coefficients.

7. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 5 is characterized in that: Constructing the time period-deviation comparison table according to the plurality of monitoring nodes and the plurality of prediction deviation coefficients includes: Arranging the multiple monitoring nodes in order of node sequence to construct a monitoring node sequence; Selecting a first deviation coefficient of a first node and a second deviation coefficient of a second node, and calculating a first coefficient difference between the first deviation coefficient and the second deviation coefficient, wherein the first node is an initial node of the monitoring node sequence, and the second node is an adjacent node of the first node; If the first coefficient difference is less than a predetermined coefficient threshold, merging the first node and the second node to construct a first time period, wherein the deviation coefficient of the first time period is the average of the first deviation coefficient and the second deviation coefficient; If the first coefficient difference is greater than or equal to the predetermined coefficient threshold, a first node segmentation line is set, and an iterative analysis is performed until all nodes of the monitoring node sequence are analyzed to obtain a plurality of node segmentation lines; Based on the multiple node dividing lines, multiple monitoring time periods and multiple deviation coefficients corresponding to the multiple monitoring time periods are determined, and the time period-deviation comparison table is constructed according to the multiple monitoring time periods and the multiple deviation coefficients.

8. The distributed photovoltaic energy efficiency dynamic monitoring method based on time series analysis according to claim 3 is characterized in that: The power output compensation is performed based on the period-deviation comparison table, including: Receiving real-time electrical operation parameters of the current node, and using the power predictor to predict photovoltaic output power based on the real-time electrical operation parameters to obtain real-time predicted output power; Input the current node into the time period-deviation comparison table for matching, and output an adaptation deviation coefficient; The real-time predicted output power is compensated according to the adaptation deviation coefficient to obtain the real-time output power, and the grid-connected scheduling of the distributed photovoltaic equipment is performed.

Images