Low-voltage distributed photovoltaic power prediction method and device

Abstract

According to the low-voltage distributed photovoltaic power prediction method and device provided by the invention, on one hand, the low-voltage distributed photovoltaic characteristic parameters with insufficient measurement information are extracted, and the feeder line low-voltage distributed photovoltaic installation capacity identification model is constructed through the improved particle swarm optimization, so that the distributed photovoltaic capacity information installed in the feeder line is effectively identified; and on the other hand, the improved bidirectional long-short-term memory neural network model is adopted, so that the forward and backward dependency relationship of the load sequence can be learned at the same time, and a complex time sequence mode can be captured more effectively. And on the other hand, an optimal set of key parameters of the bidirectional long-short-term memory neural network is automatically searched by using an improved particle swarm optimization algorithm, so that the model performance is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power prediction technology, and in particular to a method and apparatus for predicting low-voltage distributed photovoltaic power. Background Technology

[0002] With the rapid development of distributed energy resources, especially distributed photovoltaic (PV), their output is intermittent, fluctuating, and exhibits anti-peak-shaving characteristics, leading to significant changes in the power curve shape of power lines and posing a severe challenge to the safe and stable operation of the power system. Due to insufficient measurement information on low-voltage distributed PV in feeders—such as whether distributed PV is installed on the feeder, its installed capacity, and its power output—power prediction inaccurately affects the safe and stable operation of the power grid and optimal dispatch decisions.

[0003] Existing photovoltaic (PV) power prediction methods mainly include traditional statistical methods (such as regression analysis and time series analysis) and artificial intelligence methods (such as support vector machines (SVM), random forests (RF), and long short-term memory networks (LSTM). Traditional statistical methods are ineffective in handling nonlinear features. While artificial intelligence methods can handle nonlinear problems, they still have limitations when dealing with low-voltage distributed PV: 1) they fail to effectively identify the power output of low-voltage distributed PV in feeders; 2) they lack the ability to extract the characteristics of low-voltage distributed PV that lacks measurement information; 3) the parameters of the power prediction model have a significant impact on the prediction accuracy, but manual parameter tuning is inefficient and it is difficult to find the optimal solution. Summary of the Invention

[0004] In view of this, the present invention provides a method for predicting low-voltage distributed photovoltaic power. One or more embodiments of this specification also relate to a low-voltage distributed photovoltaic power prediction device, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the present invention, a low-voltage distributed photovoltaic power prediction method is provided, comprising: Acquire historical power data of low-voltage distributed photovoltaic nodes on the target feeder; Based on historical power data, a matrix of characteristic parameters for photovoltaic nodes is generated for different weather types during preset time periods each day. The weather types include sunny, partly cloudy, cloudy, and rainy. The feature parameter matrix is ​​imported into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty. If the predicted rated capacity is not empty, the predicted rated capacity of the photovoltaic node is determined based on historical power data. Historical power data, weather type at the target time, and predicted rated capacity are imported into a trained power prediction model to predict the target output power of photovoltaic nodes at the target time.

[0006] In some implementations, based on historical power data, characteristic parameter matrices are generated for photovoltaic nodes corresponding to different weather types during preset time periods each day, including: Based on historical power data, multiple power curves for photovoltaic nodes are generated for various weather types during a preset time period each day. The preset time period is from 08:00 to 17:00 each day, and the data collection interval between data collection points is 15 minutes each day. The data for each data collection point is the average of all data under the same weather type. Based on each power curve, corresponding feature parameters are generated, and a feature parameter matrix for the photovoltaic node is constructed based on the generated feature parameters. The feature parameters include: It represents the power ratio of the line in sunny weather to that in cloudy or rainy weather, the concavity value reflecting the degree of concavity in the power curve, the concavity index, a characteristic parameter quantifying the sharpness of the concavity in the power curve at noon, and the power change rate reflecting the modulation intensity of the line power by the photovoltaic power generation during the day and night.

[0007] In some implementations, the formula for calculating the power ratio includes:

[0008] In the formula: C1 represents the power ratio, || represents the absolute value, and PC S (t) represents the power curve value for a clear day at time t, PC R (t) represents the power curve value for a clear day at time t. s Let t be the time corresponding to the start time S of the preset time period. e This refers to the time corresponding to the end time E of the preset time period.

[0009] In some implementations, the step of calculating the concavity value includes: Draw a straight line L connecting the start time point S and the end time point E, and calculate the slope k of L. L The calculation formulas include:

[0010]

[0011] Among them, y L PC(t) is a function of line L with respect to time t. e ) for t e The power curve value at time t, PC(t) e ) for t e The power curve value at time t; Construct a set A based on all points at time t along line L, i.e., A = {y L (t) | t∈[t s , t e ]}; Construct a set B based on all points at time t that lie below line L, where B = {PC} S (t) | PC S (t) <y L (t), t∈[t s , t e ]}; The concavity value is calculated based on set A and set B, and the formula includes:

[0012] Where C2 represents the concave value, and || represents the number of elements in set A or set B.

[0013] In some implementations, the step of calculating the concavity index includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon )); Using 08:00 time point t respectively A1 and 18:00 time point t B1 Connect this point to form straight lines L1 and L2. Calculate the slope k of L1 and L2 using the following formula. L1 and k L2 :

[0014]

[0015] Among them, PC S (t noon1 ) represents t noon1 Power curve value on a sunny day at any given time, PC S (t A1 ) represents t A1 Power curve value on a sunny day at any given time, PC S (t B1 ) represents t B1 The power curve value for a sunny day at any given time; k L1 and k L2 Adding the absolute values ​​gives the slope and k of a sunny day. S ; Using 08:00 time point t respectively A2 and 18:00 time point tB2 Connect this point to form lines L3 and L4. Calculate the slope k of L3 and L4 using the following formula. L3 and k L4 :

[0016]

[0017] Among them, PC S (t noon2 ) represents t noon2 Power curve value on a sunny day at any given time, PC S (t A2 ) represents t A2 Power curve value on a sunny day at any given time, PC S (t B2 ) represents t B2 The power curve value for a sunny day at any given time; k L3 and k L4 Adding the absolute values ​​gives the slope and k of a sunny day. R ; The formula for calculating the concavity index includes:

[0018] C3 is the concave index.

[0019] In some implementations, the step of calculating the rate of change of power includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon The coordinate of the point at 21:00 at night is t. night The coordinates of this point on the line power curve are (t night , PC(t night Let λ be the slope of the straight line connecting this point to the end time point E. Calculate the slope for sunny days and rainy days using the following formulas:

[0020]

[0021] Where, λ S Let λ be the slope of a sunny day. R The slope under rainy weather conditions; The formulas for calculating the rate of change of power include:

[0022] Where C4 represents the power change rate.

[0023] In some implementations, the feature parameter matrix is ​​imported into a preset identification model to determine whether the predicted rated capacity of the photovoltaic node is empty, including: The power ratio, concavity value, concavity index, and power change rate are imported into the identification model to determine whether the predicted rated capacity of the photovoltaic node is empty. The identification model is either a random forest model or a support vector machine model.

[0024] In some implementations, the power prediction model is a long short-term memory neural network model; Before training the power prediction model, the parameters were optimized by improving the particle swarm optimization algorithm; Historical power data, target time weather type, and predicted rated capacity are imported into the optimized power prediction model for training, resulting in a trained power prediction model.

[0025] According to a second aspect of the present invention, a low-voltage distributed photovoltaic power prediction device is provided, comprising: The acquisition module is configured to acquire historical power data of low-voltage distributed photovoltaic nodes on the target feeder; The generation module is configured to generate characteristic parameter matrices for photovoltaic nodes under different weather types during preset time periods each day, based on historical power data. The weather types include sunny, partly cloudy, cloudy, and rainy. The first determining module is configured to import the feature parameter matrix into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty. The second determination module is configured to determine the predicted rated capacity of the photovoltaic node based on historical power data when the predicted rated capacity is not empty. The prediction module is configured to import historical power data, the weather type at the target time, and the predicted rated capacity into a trained power prediction model to predict the target output power of the photovoltaic nodes at the target time.

[0026] In some implementations, based on historical power data, characteristic parameter matrices are generated for photovoltaic nodes corresponding to different weather types during preset time periods each day, including: Based on historical power data, multiple power curves for photovoltaic nodes are generated for various weather types during a preset time period each day. The preset time period is from 08:00 to 17:00 each day, and the data collection interval between data collection points is 15 minutes each day. The data for each data collection point is the average of all data under the same weather type. Based on each power curve, corresponding feature parameters are generated, and a feature parameter matrix for the photovoltaic node is constructed based on the generated feature parameters. The feature parameters include: It represents the power ratio of the line in sunny weather to that in cloudy or rainy weather, the concavity value reflecting the degree of concavity in the power curve, the concavity index, a characteristic parameter quantifying the sharpness of the concavity in the power curve at noon, and the power change rate reflecting the modulation intensity of the line power by the photovoltaic power generation during the day and night.

[0027] In some implementations, the formula for calculating the power ratio includes:

[0028] In the formula: C1 represents the power ratio, || represents the absolute value, and PC S (t) represents the power curve value for a clear day at time t, PC R (t) represents the power curve value for a clear day at time t. s Let t be the time corresponding to the start time S of the preset time period. e This refers to the time corresponding to the end time E of the preset time period.

[0029] In some implementations, the step of calculating the concavity value includes: Draw a straight line L connecting the start time point S and the end time point E, and calculate the slope k of L. L The calculation formulas include:

[0030]

[0031] Among them, y L PC(t) is a function of line L with respect to time t. e ) for t e The power curve value at time t, PC(t) e ) for t e The power curve value at time t; Construct a set A based on all points at time t along line L, i.e., A = {y L (t) | t∈[t s , t e ]}; Construct a set B based on all points at time t that lie below line L, where B = {PC} S (t) | PC S (t) <y L (t), t∈[t s , t e ]}; The concavity value is calculated based on set A and set B, and the formula includes:

[0032] Where C2 represents the concave value, and || represents the number of elements in set A or set B.

[0033] In some implementations, the step of calculating the concavity index includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon )); Using 08:00 time point t respectively A1 and 18:00 time point t B1 Connect this point to form straight lines L1 and L2. Calculate the slope k of L1 and L2 using the following formula. L1 and k L2 :

[0034]

[0035] Among them, PC S (t noon1 ) represents t noon1 Power curve value on a sunny day at any given time, PC S (t A1 ) represents t A1 Power curve value on a sunny day at any given time, PC S (t B1 ) represents t B1 The power curve value for a sunny day at any given time; k L1 and k L2 Adding the absolute values ​​gives the slope and k of a sunny day. S ; Using 08:00 time point t respectively A2 and 18:00 time point t B2 Connect this point to form lines L3 and L4. Calculate the slope k of L3 and L4 using the following formula. L3 and k L4 :

[0036]

[0037] Among them, PC S (t noon2 ) represents t noon2 Power curve value on a sunny day at any given time, PC S (t A2 ) represents t A2 Power curve value on a sunny day at any given time, PC S (t B2) represents t B2 The power curve value for a sunny day at any given time; k L3 and k L4 Adding the absolute values ​​gives the slope and k of a sunny day. R ; The formula for calculating the concavity index includes:

[0038] C3 is the concave index.

[0039] In some implementations, the step of calculating the rate of change of power includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon The coordinate of the point at 21:00 at night is t. night The coordinates of this point on the line power curve are (t night , PC(t night Let λ be the slope of the straight line connecting this point to the end time point E. Calculate the slope for sunny days and rainy days using the following formulas:

[0040]

[0041] Where, λ S Let λ be the slope of a sunny day. R The slope under rainy weather conditions; The formulas for calculating the rate of change of power include:

[0042] Where C4 represents the power change rate.

[0043] In some implementations, the feature parameter matrix is ​​imported into a preset identification model to determine whether the predicted rated capacity of the photovoltaic node is empty, including: The power ratio, concavity value, concavity index, and power change rate are imported into the identification model to determine whether the predicted rated capacity of the photovoltaic node is empty. The identification model is either a random forest model or a support vector machine model.

[0044] In some implementations, the power prediction model is a long short-term memory neural network model; Before training the power prediction model, the parameters were optimized by improving the particle swarm optimization algorithm; Historical power data, target time weather type, and predicted rated capacity are imported into the optimized power prediction model for training, resulting in a trained power prediction model.

[0045] According to a third aspect of the present invention, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the low-voltage distributed photovoltaic power prediction method described above.

[0046] According to a fourth aspect of the present invention, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the low-voltage distributed photovoltaic power prediction method described above.

[0047] According to a fifth aspect of the present invention, a computer program is provided, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the low-voltage distributed photovoltaic power prediction method described above.

[0048] At least one embodiment of this invention, on the one hand, extracts characteristic parameters of low-voltage distributed photovoltaic (PV) systems with insufficient measurement information and constructs a feeder-based low-voltage distributed PV installation capacity identification model using an improved particle swarm optimization algorithm, effectively identifying the distributed PV capacity information installed in the feeder. On the other hand, it employs an improved bidirectional long short-term memory (LSTM) neural network model, which can simultaneously learn the forward and backward dependencies of load sequences, more effectively capturing complex time-series patterns. Furthermore, it utilizes an improved particle swarm optimization algorithm to automatically search for the optimal set of key parameters of the bidirectional LSTM neural network, significantly improving model performance. Attached Figure Description

[0049] Figure 1 This is a flowchart of a low-voltage distributed photovoltaic power prediction method provided by the present invention; Figure 2a This is a schematic diagram of the RMSE optimization result in the particle swarm optimization algorithm process of the low-voltage distributed photovoltaic power prediction method provided by the present invention; Figure 2b This is a schematic diagram of the parameter optimization trajectory in the particle swarm optimization algorithm process of a low-voltage distributed photovoltaic power prediction method provided by the present invention; Figure 2c This is a schematic diagram of the learning rate vs. RMSE in the particle swarm optimization algorithm process of a low-voltage distributed photovoltaic power prediction method provided by the present invention; Figure 2dThis is a schematic diagram of the grid size vs. RMSE in the particle swarm optimization algorithm process of a low-voltage distributed photovoltaic power prediction method provided by the present invention; Figure 3 This is a schematic diagram considering all prediction results of distributed photovoltaic power in some embodiments of a low-voltage distributed photovoltaic power prediction method provided by the present invention; Figure 4 This is a schematic diagram showing the time-segmented prediction results of distributed photovoltaic power, based on some embodiments of the low-voltage distributed photovoltaic power prediction method provided by the present invention. Figure 5 This is a simplified structural diagram of a low-voltage distributed photovoltaic power prediction device provided by the present invention; Figure 6 This is a structural block diagram of a computing device provided by the present invention. Detailed Implementation

[0050] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0051] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the one or more embodiments of this specification. The singular forms “a” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The modifications “a” and “a plurality” as used in this disclosure are illustrative and not restrictive, and those skilled in the art will understand that they should be understood as “one or more” unless the context clearly indicates otherwise.

[0052] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0053] See Figure 1 , Figure 1A flowchart of a low-voltage distributed photovoltaic power prediction method according to some embodiments of this specification is shown, which specifically includes the following steps.

[0054] Step 101: Obtain historical power data of the low-voltage distributed photovoltaic nodes on the target feeder.

[0055] Historical power data refers to a sequence of actual output power values ​​of photovoltaic (PV) power generation units recorded at past points in time. For example, this data is typically arranged chronologically, including date, timestamp, and corresponding power values, reflecting the patterns of PV power generation variation with sunshine and weather. This helps reveal the operational characteristics of PV nodes and weather-related influence patterns.

[0056] Step 102: Based on historical power data, generate characteristic parameter matrices for photovoltaic nodes for different weather types during preset time periods each day. Weather types include sunny, partly cloudy, cloudy, and rainy.

[0057] The feature parameter matrices corresponding to different weather types can refer to a data structure organized in matrix form for each photovoltaic power generation unit, representing multiple feature parameters calculated for each unit within a fixed daily period (e.g., daytime) based on different weather conditions (sunny, partly cloudy, etc.). For example, for "sunny" weather, all dates meeting the criteria are extracted from historical data, the average power curve for that period is calculated, and a set (e.g., four) of quantified feature parameters are generated based on this curve, forming a row vector. Different weather types correspond to different row vectors, collectively constituting the feature matrix of that node. This matrix compactly represents the power output characteristics of the node under different weather conditions, used for subsequent capacity identification. Weather type can refer to the classification of meteorological conditions based on cloud cover, solar radiation intensity, and precipitation. For example, in this scheme, it is mainly divided into four categories: sunny, partly cloudy, cloudy, and overcast / rainy. This classification directly affects the power output level of photovoltaic power generation. It serves as one of the key input variables for the power prediction model, distinguishing different power generation scenarios.

[0058] Step 103: Import the feature parameter matrix into the preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty.

[0059] The pre-set identification model can refer to a classification model pre-trained using machine learning algorithms to determine the state of a specific object. For example, in this scheme, the model is trained to determine whether the corresponding photovoltaic node is installed and connected to the grid (i.e., its capacity is not empty) or not installed (its capacity is empty) based on the input feature parameter matrix. It can automatically filter out photovoltaic nodes that are actually in operation, filtering out invalid data and ensuring the accuracy of subsequent predictions. Predicting rated capacity refers to estimating the possible rated installed power of a photovoltaic node when its actual installed capacity is unknown, using algorithms. For example, using the node's known historical power data and its power generation patterns, its approximate maximum power generation capacity (rated capacity) can be deduced through methods such as peak identification, curve fitting, or regression analysis. This provides a key equipment parameter input for the power prediction model, improving prediction accuracy.

[0060] Step 104: If the predicted rated capacity is not empty, determine the predicted rated capacity of the photovoltaic node based on historical power data.

[0061] Step 105: Import historical power data, weather type at the target time, and predicted rated capacity into the trained power prediction model to predict the target output power of the photovoltaic nodes at the target time.

[0062] A well-trained power prediction model can refer to an artificial intelligence or statistical model that has completed parameter learning using a large amount of historical data and can output future power prediction values ​​based on input conditions. For example, taking historical power sequences, weather type at the target time, and (predicted) rated capacity as inputs, the model, after training, can output the predicted power value of that node at the target time. This is used to ultimately achieve an accurate estimate of photovoltaic power generation at a specific future time. The target time can refer to a specific future time or period where photovoltaic power generation prediction is needed. For example, it could be 2:00 PM tomorrow, or every 15-minute interval within the next 24 hours. This time condition input to the power prediction model specifies the target point for prediction. The target output power can refer to the actual active power that the photovoltaic power generation unit is expected to generate at the target time. For example, a model prediction of 5.2 kW means that the photovoltaic node is expected to output 5.2 kW of electrical energy at the target time. This is the key information ultimately generated by the power prediction method and can be used by the grid dispatch.

[0063] After calculating the target output power, the actual power of the line in operation can be calculated by combining it with the actual detected line power. That is, Pline(t) = Preal(t) - Ppv(t); Where Pline(t) is the actual detected line power at time t, Preal(t) is the true line power at time t, and Ppv(t) is the target output power of the line at time t.

[0064] As a concrete example: A municipal power supply company needs to predict the next day's power output for numerous low-voltage distributed photovoltaic (PV) systems connected to a 10 kV feeder. First, historical power data for the past year from all PV access points along the feeder is obtained from the electricity consumption information collection system, with a collection frequency of 15 minutes. Then, for each PV node, its historical data is categorized into four weather types: sunny, partly cloudy, cloudy, and overcast / rainy. Power data for the period from 08:00 to 17:00 each day is extracted to generate an average daily power curve for that weather type. Based on these curves, four characteristic parameters—power ratio, concavity value, concavity index, and power change rate—are calculated to form a feature parameter matrix for that node. Next, this matrix is ​​input into a pre-trained Random Forest recognition model with a large number of samples. The model outputs that the node has "non-empty capacity." Subsequently, based on the peak power output of the node's historical sunny-day data, the system uses a fitting method to estimate its predicted rated capacity as 8 kW. Finally, the historical power data sequence of the node, the weather forecast for the next day (such as "cloudy"), and the estimated rated capacity of 8 kilowatts are input into a Long Short-Term Memory (LSTM) neural network with parameters optimized by an improved Particle Swarm Optimization algorithm. The trained LSTM model outputs a predicted output power of 4.7 kilowatts for the photovoltaic node at 14:00 the next day.

[0065] The beneficial effects of one of the embodiments in this specification include at least the following: Firstly, by extracting characteristic parameters of low-voltage distributed photovoltaic (PV) systems with insufficient measurement information and constructing a feeder-based low-voltage distributed PV installation capacity identification model using an improved particle swarm optimization algorithm, the capacity information of distributed PV systems installed in feeders can be effectively identified. Secondly, by employing an improved bidirectional long short-term memory (LSTM) neural network model, the forward and backward dependencies of load sequences can be learned simultaneously, more effectively capturing complex time-series patterns. Thirdly, by utilizing an improved particle swarm optimization algorithm to automatically search for the optimal set of key parameters of the bidirectional LSTM neural network, the model performance is significantly improved.

[0066] In some implementations, based on historical power data, a matrix of characteristic parameters corresponding to different weather types for different time periods each day is generated for the photovoltaic nodes. This includes: generating multiple power curves for different weather types for different time periods each day based on historical power data, wherein the preset time period is 08:00-17:00 each day, the data collection interval between data collection points is 15 minutes each day, and the data at each data collection point is the average of all data under the same weather type; generating corresponding characteristic parameters based on each power curve, and constructing a characteristic parameter matrix for the photovoltaic nodes based on the generated multiple characteristic parameters, wherein the characteristic parameters include: the power ratio of the line in sunny weather to the line in cloudy and rainy weather, the concavity value reflecting the degree of concavity of the power curve, the concavity index of the characteristic parameter quantifying the sharpness of the concavity of the power curve at noon, and the power change rate reflecting the modulation intensity of the line power diurnal variation of photovoltaic power generation.

[0067] A power curve can refer to a graphical or data-driven representation of how photovoltaic (PV) power generation changes over time under specific conditions. For example, in this scheme, for a certain weather type, the power values ​​at each collection time (e.g., 08:00, 08:15…) between 08:00 and 17:00 each day of all historical days conforming to that type are averaged to obtain a power-time series representing the typical daily power generation under that weather type. This serves as a benchmark curve for calculating various characteristic parameters, reflecting the power generation pattern under that type of weather. A preset time period can refer to a pre-defined daily time window used for feature analysis and power prediction. For example, in this scheme, it is defined as 08:00 to 17:00 each day, covering the period of strong sunshine and the main time of PV power generation. This focuses on the active power generation period, improving the targeting and efficiency of feature extraction and prediction. Data collection points can refer to the specific moments within the preset time period where power data is recorded at fixed time intervals. For example, starting from 08:00, data is recorded every 15 minutes. 08:00, 08:15, 08:30, etc., are data collection points. These constitute the discrete data points on the power curve, forming the basis for generating the continuous curve and performing calculations. Feature parameters refer to numerical indicators extracted from the power curve that can quantify certain specific shapes or attributes of the curve. For example, in this scheme, the power ratio, concavity value, concavity index, and power change rate each characterize the shape features of the power curve from different perspectives. These are used to transform the intuitive shape of the power curve into numerical features that can be recognized and processed by the model, forming the feature parameter matrix.

[0068] The power ratio refers to the quantitative ratio of the difference between the power curve under sunny weather and the power curve under cloudy / rainy weather within a preset time period. The larger the value, the more significant the difference in power generation under the two weather conditions, reflecting the sensitivity of photovoltaic power generation to weather (especially sunshine). The concavity value is a numerical indicator that quantifies the degree of concavity (or convexity) of the sunny power curve relative to the line connecting its start and end times. It characterizes the degree of the "hump" or "concave" shape formed by the midday output drop of the sunny power curve due to factors such as temperature. The concavity index is a numerical indicator that further refines the measurement of the sharpness of the concave shape of the sunny power curve at noon. The larger the value, the faster the power drop at noon and the "sharper" the curve, used to capture the steep local changes in the power curve near its peak. The power change rate can be an indicator that reflects the intensity of the modulation of the overall power of the line by photovoltaic power generation during the day and night. It is obtained by comparing the slope of power decrease during a specific period (such as from noon to night) on sunny and cloudy days. It is used to characterize the weakening or strengthening effect of photovoltaic access on the typical characteristic of high load during the day and low load at night of the distribution line.

[0069] As a concrete example: In the step of generating the feature parameter matrix, the system processes a photovoltaic node. First, from the node's historical data over the past three years, all dates marked as "sunny" are selected, and the power sequence for each day from 08:00 to 17:00 is extracted. This sequence consists of 37 points, one point every 15 minutes. Then, the power values ​​at the same time each day (e.g., 09:30) are averaged to obtain a "typical daily power curve for sunny days" composed of 37 average data points. Similarly, a "typical daily power curve for cloudy / rainy days" is generated. Based on these two curves, the system begins calculating characteristic parameters: The power ratio C1, representing the difference between the two curves, is calculated using an integral formula; a straight line L is drawn connecting the points 08:00 and 17:00 on the sunny curve, and the proportion of points on the curve below L is counted to obtain the concavity value C2; ​​the sum of the absolute values ​​of the slopes of the lines connecting the sunny curve at 12:00 with the points 08:00 and 17:00 is calculated to obtain the concavity index C3; finally, the ratio of the slopes of the lines connecting the sunny and cloudy / rainy curves from 12:00 to 21:00 is calculated to obtain the power change rate C4. This set of parameters [C1, C2, C3, C4] constitutes the characteristic row vector for the "sunny" weather type at this node. The same process is performed for the partly cloudy and partly cloudy types, ultimately resulting in a 4-row (weather type) and 4-column (feature) characteristic parameter matrix.

[0070] By clearly defining the preset time period and data collection specifications, the uniformity and comparability of the feature extraction data foundation were ensured. By defining and calculating a set of multi-dimensional feature parameters such as power ratio, concavity value, concavity index, and power change rate, the core shape characteristics of the photovoltaic power generation curve under different weather conditions can be comprehensively and quantitatively characterized from multiple technical dimensions such as global comparison, overall shape, local sharpness, and diurnal modulation. This provides rich and accurate input information for subsequent capacity identification and power prediction.

[0071] In some implementations, the formula for calculating the power ratio includes:

[0072] In the formula: C1 represents the power ratio, || represents the absolute value, and PC S (t) represents the power curve value for a clear day at time t, PC R (t) represents the power curve value for a clear day at time t. s Let t be the time corresponding to the start time S of the preset time period. e This refers to the time corresponding to the end time E of the preset time period.

[0073] A calculation formula can refer to a formula or expression used for mathematical operations. For example, this claim provides a specific integral form formula for calculating the power ratio C1. This is used to clarify the specific mathematical definitions and calculation paths of key parameters in the technical solution, ensuring consistency and accuracy in implementation.

[0074] As a concrete example: When calculating the power ratio C1, the system has obtained typical power curves for the photovoltaic node under sunny and cloudy / rainy weather conditions. Both curves define 37 data points at 15-minute intervals within the time interval [08:00, 17:00]. The system discretizes the time variable t, corresponding to these 37 time points. For each time point, the values ​​of the sunny and cloudy / rainy curves are read separately, and the absolute value of their difference is calculated. Since the acquisition interval is 15 minutes (i.e., 0.25 hours), the system uses the rectangular integration method, multiplying the absolute value of each time point by the time interval of 0.25 hours, and then summing the 37 products. The final sum is an approximation of the power ratio C1. The larger this value, the more significant the difference in power generation of the photovoltaic node between sunny and cloudy / rainy days.

[0075] By providing a specific integral formula for the power ratio, the definition of this characteristic parameter has a rigorous mathematical foundation and is operable, avoiding ambiguity. By employing absolute value integration, the overall shape difference of the power curve under two extreme weather conditions throughout the entire power generation period can be effectively captured, thus stably reflecting the degree of dependence of photovoltaic power generation on sunny weather.

[0076] In some implementations, the step of calculating the concavity value includes: Draw a straight line L connecting the start time point S and the end time point E, and calculate the slope k of L. L The calculation formulas include:

[0077]

[0078] Among them, y L PC(t) is a function of line L with respect to time t. e ) for t e The power curve value at time t, PC(t) e ) for t e The power curve value at time t; construct a set A based on all points at time t along line L, i.e., A = {y L (t) | t∈[t s , t e ]}; Construct a set B based on all points at time t that lie below line L, B = {PC S (t) | PC S (t) <y L (t), t∈[t s ,t e The concavity value is calculated based on set A and set B, and the formula includes:

[0079] Where C2 represents the concave value, and || represents the number of elements in set A or set B.

[0080] As a concrete example: when calculating the concavity value C2, the system already has a typical power curve for a sunny day. First, the starting point (08:00, P_start) and ending point (17:00, P_end) of the curve are taken, and the slope of the line connecting the two points is calculated. Then, for each data acquisition point at a 15-minute interval (37 in total), the value of line L at that time is calculated. Next, the system iterates through all t, determining whether the actual power curve value is less than the corresponding y. L (t). Statistically, PC is satisfied. S (t) <y L The number of points in condition (t) is denoted as N. below The concavity value C2 is ultimately calculated to be N. below / 37. If the C2 value is close to 0.5, it indicates that the curve is roughly symmetrical about the straight line; if C2 is significantly greater than 0.5, it indicates that the curve is located below the straight line and has a distinct "concave" shape.

[0081] By constructing a reference straight line connecting the start and end points and calculating the proportion of curve points located below it, an intuitive and robust method is provided to quantify the concave characteristics of the overall shape of the power curve. This method is sensitive to the overall trend of the curve, effectively capturing the degree of "midday dip" in the power curve on sunny days caused by temperature effects, etc., and is simple to calculate with good noise resistance.

[0082] In some implementations, the step of calculating the concavity index includes: labeling the abscissa of the point at 12:00 noon as t. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon )); respectively using 08:00 time point t A1 and 18:00 time point t B1 Connect this point to form straight lines L1 and L2. Calculate the slope k of L1 and L2 using the following formula. L1 and k L2 :

[0083]

[0084] Among them, PC S (t noon1 ) represents t noon1 Power curve value on a sunny day at any given time, PC S (t A1 ) represents t A1 Power curve value on a sunny day at any given time, PC S (t B1 ) represents t B1 The power curve value of a sunny day at a given time; k L1 and k L2 Adding the absolute values ​​gives the slope and k of a sunny day. S ; using 08:00 time point t respectively A2 and 18:00 time point t B2 Connect this point to form lines L3 and L4. Calculate the slope k of L3 and L4 using the following formula. L3 and k L4 :

[0085]

[0086] Among them, PC S (t noon2 ) represents t noon2 Power curve value on a sunny day at any given time, PC S (t A2 ) represents t A2Power curve value on a sunny day at any given time, PC S (t B2 ) represents t B2 The power curve value of a sunny day at a given time; k L3 and k L4 Adding the absolute values ​​gives the slope and k of a sunny day. R The formula for calculating the concavity index includes:

[0087] C3 is the concave index.

[0088] The concave index was defined by calculating and comparing the sum of the absolute values ​​of the slope changes around noon on sunny and cloudy days. This index effectively isolates the influence of overall power generation levels, focusing on characterizing the "sharpness" or "steepness" of the local shape of the power curve near its peak. On sunny days, due to the temperature effect and strong sunlight, the power curve rises and falls faster near noon, and this index can keenly capture this characteristic difference, further enriching the dimensions of the power curve's shape description.

[0089] In some implementations, the step of calculating the rate of change of power includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon The coordinate of the point at 21:00 at night is t. night The coordinates of this point on the line power curve are (t night , PC(t night Let λ be the slope of the straight line connecting this point to the end time point E. Calculate the slope for sunny days and rainy days using the following formulas:

[0090]

[0091] Where, λ S Let λ be the slope of a sunny day. R The slope under rainy weather conditions; The formulas for calculating the rate of change of power include:

[0092] Where C4 represents the power change rate.

[0093] By calculating and comparing the power decline slope from noon to night on sunny and cloudy days, a power change rate index was defined. This index cleverly embeds the impact of photovoltaic power generation into the diurnal variation pattern of the overall power of the line for evaluation. It can effectively characterize the intensity of the transformation of the original load curve shape of the distribution network by distributed photovoltaic access, that is, the quantitative manifestation of its "peak shaving" capability, providing another important characteristic perspective for the grid analysis of the grid-connected impact of photovoltaics.

[0094] In some implementations, the feature parameter matrix is ​​imported into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty. This includes: importing the power ratio, concavity value, concavity index and power change rate into the recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty. The recognition model is a random forest model or a support vector machine model.

[0095] As a concrete example: In the capacity identification stage, the system has generated feature parameter matrices for tens of thousands of historical photovoltaic nodes, and the true capacity status (known or unknown) of these nodes is pre-labeled. The system uses this data to train a random forest identification model. This model uses four feature parameters (power ratio C1, concavity value C2, concavity index C3, and power change rate C4) as input features. When it is necessary to determine the capacity status of a new photovoltaic node, the system first calculates these four feature values ​​for the node and then inputs them into the trained random forest model. Multiple decision trees within the model make judgments and votes, ultimately outputting a classification result of "capacity not empty," thereby triggering the subsequent capacity prediction process.

[0096] Construct a neural network model with a bidirectional long short-term memory neural network as its core. The input layer dimension is equal to the multi-dimensional time-series feature dimension, the distributed photovoltaic installation capacity feature is 1-dimensional, and the weather factor feature is 1-dimensional.

[0097] The main calculation formula for Long Short-Term Memory (LSTM) neural networks is: (twenty four) in, f t Let t be the vector of the forget gate at time step t; σ It is the sigmoid activation function; W f 、b f Here are the weight matrix and bias vector for the forget gate; h t-1 、h t These are the output values ​​of the state units at time step t-1 and time step t; x t The input value for time step t; i t The vector of the input gate at time step t;W i 、b i Here are the weight matrix and bias vector of the input gate; c t-1 、c t Let be the cell state vectors at time step t-1 and time step t; W o 、b o The weight matrix and bias vector represent the candidate cell states. o t The output gate vector for time step t; W o 、b o These are the weight matrix and bias vector of the output gate.

[0098] The main calculation formula for a bidirectional long short-term memory neural network is:

[0099] in, The hidden layer state of the forward long short-term memory neural network at time step t; For positive long short-term memory neural network units; x t For time step t; The hidden layer state of the forward long short-term memory neural network at time step t-1; The hidden layer state of the reverse long short-term memory neural network at time step t; It is a reverse long short-term memory neural network unit; H represents the hidden layer state of the inverse long short-term memory neural network at time step t+1. t Let t represent the hidden layer state of the bidirectional long short-term memory neural network at time step t.

[0100] By employing mature machine learning models such as random forests or support vector machines as recognition models, the four carefully designed features of power ratio, concavity value, concavity index, and power change rate can be fully utilized to achieve rapid, automatic, and accurate binary classification judgment of photovoltaic node capacity status, effectively improving the intelligence level of the front end of the entire prediction process and the reliability of data screening.

[0101] In some implementations, the power prediction model is a long short-term memory neural network model; before training the power prediction model, the parameters are optimized by improving the particle swarm optimization algorithm; historical power data, weather type at the target time, and predicted rated capacity are imported into the parameter-optimized power prediction model for training to obtain a trained power prediction model.

[0102] Improved Particle Swarm Optimization (PSO) algorithms can refer to enhancements or variations of the standard PSO algorithm, aiming to improve its optimization performance, such as avoiding premature convergence and improving accuracy. For example, improvements might involve dynamic adjustment of inertia weights, adaptive changes in learning factors, or the integration of other optimization strategies. Parameter optimization is used to automatically and efficiently search for and determine the optimal combination of key hyperparameters (such as the number of hidden layer neurons and the learning rate) of an LSTM model before training it. It involves automatically adjusting the configuration parameters (hyperparameters) of a machine learning model to achieve better performance on a specific task. For example, using an improved PSO algorithm, the prediction error on the validation set is used as the fitness function to iteratively evaluate the performance of an LSTM model under different hyperparameter combinations, ultimately finding the set of hyperparameters that minimizes the error. It can improve the final prediction accuracy and generalization ability of power prediction models, reducing the tediousness and uncertainty of manual parameter tuning.

[0103] As a concrete example, an improved particle swarm optimization algorithm is used to optimize the parameters of a bidirectional long short-term memory (BSSM) neural network distributed photovoltaic power prediction model, including the number of BSSM units, learning rate, dropout rate, and batch size. The specific steps for optimization using the improved particle swarm optimization algorithm are as follows: 1) Define the parameters of the particle swarm optimization algorithm: 10 particles, 50 iterations, and 4 optimization dimensions, corresponding to the number of units, learning rate, Dropout rate, and batch size of the bidirectional long short-term memory neural network. 2) Define the hyperparameter search space, that is, determine the hyperparameters to be optimized and their ranges (number of units in the bidirectional long short-term memory neural network, learning rate, dropout rate, and batch size). The number of units in the bidirectional long short-term memory neural network is 50-500, the learning rate is 0.0001-0.1, the dropout rate is 0.05-0.5, and the batch size is 16-256. 3) Train the model based on the given hyperparameters and return the evaluation metrics for the validation set; 4) Initialize the position and velocity of the particle swarm, with each particle representing a set of parameters; 5) Iteratively update the particle swarm optimization, record the results of each iteration, and update the global optimal solution in real time; 6) Train the final distributed photovoltaic power prediction model using the optimal parameters.

[0104] The optimal parameter set was finally obtained: 88 units, a learning rate of 0.00057, a dropout rate of 0.012, and a batch size of 223. The optimization process is as follows: Figure 2a , Figure 2b , Figure 2c and Figure 2d As shown.

[0105] The final distributed photovoltaic power prediction model is configured using the optimal parameter set, with 88 units, a learning rate of 0.00057, a dropout rate of 0.012, a batch size of 223, a maximum training epoch of 70, a learning rate reduction period of 12 epochs, and a learning rate reduction ratio of 60%.

[0106] By employing an LSTM neural network, which can effectively capture long-term dependencies in time series, as the power prediction model, the model's ability to model the time-series patterns of photovoltaic power generation is enhanced. By introducing an improved particle swarm optimization algorithm for automatic hyperparameter optimization before training, the optimal configuration of the model can be determined scientifically and efficiently, avoiding the blindness of manual parameter tuning. This ensures the optimal performance and prediction accuracy of the final power prediction model from both the model structure and parameter perspectives.

[0107] To further illustrate this disclosure, data is compared with other existing models: By conducting comparative experiments on relevant models, the evaluation indicators for all models are shown in Table 1.

[0108] Table 1 Evaluation metrics for all models

[0109] Among them, the weighted average absolute error (MAE) is a metric that measures the average absolute error between predicted and actual values. It is typically used to emphasize the importance of prediction errors at different time points (such as peak photovoltaic power generation periods). The smaller the value, the higher the model's prediction accuracy. The root mean square error (RMSE) is a metric that measures the square root of the average of the squared prediction errors. It is more sensitive to larger errors; the smaller the value, the better the model's prediction accuracy and stability. The mean absolute error (MAPE) is a metric that expresses the relative magnitude of prediction errors as a percentage; the smaller the value, the smaller the model's relative error. 2 The coefficient of determination (or goodness of fit) is an indicator with a value between 0 and 1. The closer the value is to 1, the better the model fits the data.

[0110] By comparing the evaluation metrics of each model group, it can be found that the predictive performance of the model also improves slightly with the increase of training epochs; when the number of training epochs is the same, the two-layer model shows a slight improvement over the single-layer model. Furthermore, because BiLSTM is better at capturing features of complex temporal sequences, its performance is generally superior to LSTM. After optimizing the hyperparameters of the algorithm using IPSO, the PSO-LSTM model achieves a MAPE of 3.2729% and an R² of 0.92165, representing improvements of 9.42% and 3.12% respectively compared to the single-layer LSTM's 3.6132% and 0.89375; and improvements of 7.64% and 1.78% respectively compared to the two-layer LSTM's 3.5437% and 0.90553. The IPSO-BiLSTM model achieved a MAPE of 3.2322% and an R² of 0.92877, representing improvements of 8.93% and 2.84% respectively compared to the single-layer BiLSTM (3.549% and 0.90311); and improvements of 7.74% and 2.42% respectively compared to the two-layer BiLSTM (3.5035% and 0.90684). This demonstrates that the improved PSO-optimized algorithm models exhibit superior predictive performance.

[0111] After adding the characteristic parameter of distributed photovoltaic installation capacity, the prediction results of the bidirectional long short-term memory neural network prediction model optimized by the improved particle swarm optimization algorithm are as follows: Figure 3 and Figure 4 As shown in Table 2, the evaluation indicators for the results are as follows.

[0112] Table 2 Model Evaluation Indicators

[0113] According to the evaluation metrics, the MAPE and R² of IPSO-LSTM-pv are 5.1321% and 0.9558, respectively, while those of IPSO-BiLSTM-pv are 4.3905% and 0.9621, respectively. IPSO-BiLSTM-pv represents an improvement of 14.45% and 6.59% compared to IPSO-LSTM-pv. Because the model incorporates photovoltaic-related features, the accuracy of the model predictions is also increased to some extent.

[0114] This method identifies the installed capacity of low-voltage distributed photovoltaic (PV) systems by using tags and optimizes the parameters of a bidirectional long short-term memory (LSTM) neural network using an improved particle swarm optimization algorithm. This effectively improves the power prediction accuracy of low-voltage distributed PV systems with insufficient measurement information, providing more reliable technical support for the safe and stable operation of new power systems.

[0115] Corresponding to the above method embodiments, this specification also provides embodiments of a low-voltage distributed photovoltaic power prediction device. Figure 5This specification illustrates a schematic diagram of the structure of a low-voltage distributed photovoltaic power prediction device according to some embodiments. For example... Figure 5 As shown, the device includes: The acquisition module is configured to acquire historical power data of low-voltage distributed photovoltaic nodes on the target feeder; The generation module is configured to generate characteristic parameter matrices for photovoltaic nodes under different weather types during preset time periods each day, based on historical power data. The weather types include sunny, partly cloudy, cloudy, and rainy. The first determining module is configured to import the feature parameter matrix into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty. The second determination module is configured to determine the predicted rated capacity of the photovoltaic node based on historical power data when the predicted rated capacity is not empty. The prediction module is configured to import historical power data, the weather type at the target time, and the predicted rated capacity into a trained power prediction model to predict the target output power of the photovoltaic nodes at the target time.

[0116] In some implementations, based on historical power data, characteristic parameter matrices are generated for photovoltaic nodes corresponding to different weather types during preset time periods each day, including: Based on historical power data, multiple power curves for photovoltaic nodes are generated for various weather types during a preset time period each day. The preset time period is from 08:00 to 17:00 each day, and the data collection interval between data collection points is 15 minutes each day. The data for each data collection point is the average of all data under the same weather type. Based on each power curve, corresponding feature parameters are generated, and a feature parameter matrix for the photovoltaic node is constructed based on the generated feature parameters. The feature parameters include: It represents the power ratio of the line in sunny weather to that in cloudy or rainy weather, the concavity value reflecting the degree of concavity in the power curve, the concavity index, a characteristic parameter quantifying the sharpness of the concavity in the power curve at noon, and the power change rate reflecting the modulation intensity of the line power by the photovoltaic power generation during the day and night.

[0117] In some implementations, the formula for calculating the power ratio includes:

[0118] In the formula: C1 represents the power ratio, || represents the absolute value, and PC S (t) represents the power curve value for a clear day at time t, PC R (t) represents the power curve value for a clear day at time t. s Let t be the time corresponding to the start time S of the preset time period. eThis refers to the time corresponding to the end time E of the preset time period.

[0119] In some implementations, the step of calculating the concavity value includes: Draw a straight line L connecting the start time point S and the end time point E, and calculate the slope k of L. L The calculation formulas include:

[0120]

[0121] Among them, y L PC(t) is a function of line L with respect to time t. e ) for t e The power curve value at time t, PC(t) e ) for t e The power curve value at time t; Construct a set A based on all points at time t along line L, i.e., A = {y L (t) | t∈[t s , t e ]}; Construct a set B based on all points at time t that lie below line L, where B = {PC} S (t) | PC S (t) <y L (t), t∈[t s , t e ]}; The concavity value is calculated based on set A and set B, and the formula includes:

[0122] Where C2 represents the concave value, and || represents the number of elements in set A or set B.

[0123] In some implementations, the step of calculating the concavity index includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon )); Using 08:00 time point t respectively A1 and 18:00 time point t B1 Connect this point to form straight lines L1 and L2. Calculate the slope k of L1 and L2 using the following formula. L1 and k L2 :

[0124]

[0125] Among them, PC S (t noon1 ) represents t noon1 Power curve value on a sunny day at any given time, PC S (t A1 ) represents t A1 Power curve value on a sunny day at any given time, PC S (t B1 ) represents t B1 The power curve value for a sunny day at any given time; k L1 and k L2 Adding the absolute values ​​gives the slope and k of a sunny day. S ; Using 08:00 time point t respectively A2 and 18:00 time point t B2 Connect this point to form lines L3 and L4. Calculate the slope k of L3 and L4 using the following formula. L3 and k L4 :

[0126]

[0127] Among them, PC S (t noon2 ) represents t noon2 Power curve value on a sunny day at any given time, PC S (t A2 ) represents t A2 Power curve value on a sunny day at any given time, PC S (t B2 ) represents t B2 The power curve value for a sunny day at any given time; k L3 and k L4 Adding the absolute values ​​gives the slope and k of a sunny day. R ; The formula for calculating the concavity index includes:

[0128] C3 is the concave index.

[0129] In some implementations, the step of calculating the rate of change of power includes: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noonThe coordinate of the point at 21:00 at night is t. night The coordinates of this point on the line power curve are (t night , PC(t night Let λ be the slope of the straight line connecting this point to the end time point E. Calculate the slope for sunny days and rainy days using the following formulas:

[0130]

[0131] Where, λ S Let λ be the slope of a sunny day. R The slope under rainy weather conditions; The formulas for calculating the rate of change of power include:

[0132] Where C4 represents the power change rate.

[0133] In some implementations, the feature parameter matrix is ​​imported into a preset identification model to determine whether the predicted rated capacity of the photovoltaic node is empty, including: The power ratio, concavity value, concavity index, and power change rate are imported into the identification model to determine whether the predicted rated capacity of the photovoltaic node is empty. The identification model is either a random forest model or a support vector machine model.

[0134] In some implementations, the power prediction model is a long short-term memory neural network model; Before training the power prediction model, the parameters were optimized by improving the particle swarm optimization algorithm; Historical power data, target time weather type, and predicted rated capacity are imported into the optimized power prediction model for training, resulting in a trained power prediction model.

[0135] The above is a schematic scheme of a low-voltage distributed photovoltaic power prediction device according to this embodiment. It should be noted that the technical solution of this low-voltage distributed photovoltaic power prediction device and the technical solution of the low-voltage distributed photovoltaic power prediction method described above belong to the same concept. For details not described in detail in the technical solution of the low-voltage distributed photovoltaic power prediction device, please refer to the description of the technical solution of the low-voltage distributed photovoltaic power prediction method described above.

[0136] Figure 6 A structural block diagram of a computing device 600 according to some embodiments of this specification is shown. The components of the computing device 600 include, but are not limited to, a memory 601 and a processor 602. The processor 602 is connected to the memory 601 via a bus 603, and a database 605 is used to store data.

[0137] The computing device 600 also includes an access device 604 that enables the computing device 600 to communicate via one or more networks 606. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 604 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0138] In one embodiment of this specification, the above-described components of the computing device 600 and Figure 6 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 6 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0139] The computing device 600 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 600 can also be a mobile or stationary server.

[0140] The processor 602 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the aforementioned low-voltage distributed photovoltaic power prediction method. The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the aforementioned low-voltage distributed photovoltaic power prediction method belong to the same concept. Details not described in detail in the technical solution of the computing device can be found in the description of the technical solution of the aforementioned low-voltage distributed photovoltaic power prediction method.

[0141] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the low-voltage distributed photovoltaic power prediction method described above.

[0142] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the low-voltage distributed photovoltaic power prediction method described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the low-voltage distributed photovoltaic power prediction method described above.

[0143] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described low-voltage distributed photovoltaic power prediction method.

[0144] The above is an illustrative example of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the technical solution of the aforementioned low-voltage distributed photovoltaic power prediction method. Details not described in detail in the computer program's technical solution can be found in the description of the technical solution of the aforementioned low-voltage distributed photovoltaic power prediction method.

[0145] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0146] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0147] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0148] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0149] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this invention. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting the power output of low-voltage distributed photovoltaic systems, characterized in that, include: Acquire historical power data of low-voltage distributed photovoltaic nodes on the target feeder; Based on the historical power data, a matrix of characteristic parameters corresponding to different weather types for the photovoltaic nodes during a preset time period each day is generated, wherein the weather types include sunny, partly cloudy, cloudy, and rainy. The feature parameter matrix is ​​imported into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty; If the predicted rated capacity is not empty, the predicted rated capacity of the photovoltaic node is determined based on the historical power data. The historical power data, the weather type at the target time, and the predicted rated capacity are imported into the trained power prediction model to predict the target output power of the photovoltaic node at the target time.

2. The method according to claim 1, characterized in that, Based on the historical power data, a feature parameter matrix is ​​generated for each photovoltaic node corresponding to different weather types during a preset time period each day, including: Based on the historical power data, multiple power curves corresponding to various weather types are generated for the photovoltaic nodes during a preset time period each day. The preset time period is from 08:00 to 17:00 each day, the data collection interval between data collection points is 15 minutes each day, and the data at each data collection point is the average value of all data under the same weather type. Based on each power curve, corresponding feature parameters are generated, and a feature parameter matrix of the photovoltaic node is constructed based on the generated feature parameters. The feature parameters include: It represents the power ratio of the line in sunny weather to that in cloudy or rainy weather, the concavity value reflecting the degree of concavity in the power curve, the concavity index, a characteristic parameter quantifying the sharpness of the concavity in the power curve at noon, and the power change rate reflecting the modulation intensity of the line power by the photovoltaic power generation during the day and night.

3. The method according to claim 2, characterized in that, The formula for calculating the power ratio includes: In the formula: C1 represents the power ratio, || represents the absolute value, and PC S (t) represents the power curve value for a clear day at time t, PC R (t) represents the power curve value for a clear day at time t. s The time corresponding to the start time S of the preset time period, t e The time corresponding to the end time point E of the preset time period.

4. The method according to claim 3, characterized in that, The steps for calculating the concavity value include: Draw a straight line L connecting the start time point S and the end time point E, and calculate the slope k of L. L The calculation formulas include: Among them, y L PC(t) is a function of line L with respect to time t. e ) for t e The power curve value at time t, PC(t) e ) for t e The power curve value at time; Construct a set A based on all points at time t along line L, i.e., A = {y L (t) | t∈[t s , t e ]}; Construct a set B based on all points at time t that lie below line L, where B = {PC} S (t) | PC S (t) < y L (t), t∈[t s , t e ]}; The concavity value is calculated based on set A and set B, and the calculation formula includes: Where C2 represents the concave value, and || represents the number of elements in set A or set B.

5. The method according to claim 4, characterized in that, The steps for calculating the concavity index include: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon ,PC(t noon )); Using 08:00 time point t respectively A1 and 18:00 time point t B1 Connect this point to form straight lines L1 and L2. Calculate the slope k of L1 and L2 using the following formula. L1 and k L2 : Among them, PC S (t noon1 ) represents t noon1 Power curve value on a sunny day at any given time, PC S (t A1 ) represents t A1 Power curve value on a sunny day at any given time, PC S (t B1 ) represents t B1 The power curve value for a sunny day at any given time; k L1 and k L2 Adding the absolute values ​​gives the slope and k of a sunny day. S ; Using 08:00 time point t respectively A2 and 18:00 time point t B2 Connect this point to form lines L3 and L4. Calculate the slope k of L3 and L4 using the following formula. L3 and k L4 : Among them, PC S (t noon2 ) represents t noon2 Power curve value on a sunny day at any given time, PC S (t A2 ) represents t A2 Power curve value on a sunny day at any given time, PC S (t B2 ) represents t B2 The power curve value for a sunny day at any given time; k L3 and k L4 Adding the absolute values ​​gives the slope and k of a sunny day. R ; The formula for calculating the concavity index includes: Wherein, C3 is the concave index.

6. The method according to claim 5, characterized in that, The steps for calculating the power change rate include: Let t be the coordinate of the point at 12:00 noon. noon The coordinates of this point on the line power curve are (t noon , PC(t noon The coordinate of the point at 21:00 at night is t. night The coordinates of this point on the line power curve are (t night ,PC(t night Let λ be the slope of the straight line connecting this point to the end time point E. Calculate the slope for sunny days and rainy days using the following formulas: Where, λ S Let λ be the slope of a sunny day. R The slope under rainy weather conditions; The formula for calculating the power change rate includes: Wherein, C4 represents the power change rate.

7. The method according to claim 6, characterized in that, The feature parameter matrix is ​​imported into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty, including: The power ratio, concavity value, concavity index, and power change rate are imported into the identification model to determine whether the predicted rated capacity of the photovoltaic node is empty. The identification model is a random forest model or a support vector machine model.

8. The method according to any one of claims 1 to 7, characterized in that, The power prediction model is a long short-term memory neural network model; Before training the power prediction model, the parameters are optimized by improving the particle swarm optimization algorithm; The historical power data, the weather type at the target time, and the predicted rated capacity are imported into the power prediction model with optimized parameters for training, resulting in the trained power prediction model.

9. A low-voltage distributed photovoltaic power prediction device, characterized in that, include: The acquisition module is configured to acquire historical power data of low-voltage distributed photovoltaic nodes on the target feeder; The generation module is configured to generate feature parameter matrices for different weather types for the photovoltaic nodes during a preset time period each day, based on the historical power data, wherein the weather types include sunny, partly cloudy, cloudy, and rainy. The first determining module is configured to import the feature parameter matrix into a preset recognition model to determine whether the predicted rated capacity of the photovoltaic node is empty; The second determining module is configured to determine the predicted rated capacity of the photovoltaic node based on the historical power data when the predicted rated capacity is not empty. The prediction module is configured to import the historical power data, the weather type at the target time, and the predicted rated capacity into a trained power prediction model to predict the output power of the photovoltaic node at the target time.

10. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the low-voltage distributed photovoltaic power prediction method according to any one of claims 1 to 8.

Images