High-proportion distributed photovoltaic access distribution transformer openable capacity prediction method and system

Abstract

The invention discloses a high-proportion distributed photovoltaic access distribution transformer openable capacity prediction method and system, and relates to the technical field of electric power metering, and the method comprises the steps: collecting data, carrying out the data preprocessing, and calculating indexes affecting the distribution transformer openable capacity; performing stabilization processing on the maximum load curve of the historical distribution transformer day, and constructing a maximum load data feature set of the distribution transformer day; and designing a distribution transformer daily maximum load curve prediction model, a distributed photovoltaic output prediction model and a distribution transformer openable capacity measurement and calculation model, and measuring and calculating the distribution transformer openable capacity. According to the high-proportion distributed photovoltaic access distribution transformer openable capacity prediction method, noise and abnormity in data are eliminated through data preprocessing, the accuracy and reliability of model training and prediction are improved, the stability and predictability of historical data are improved through stationary processing of the historical data, and the prediction accuracy of the distribution transformer openable capacity is improved. The method has better effects in the aspects of prediction accuracy, robustness and utilization rate.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power metering, and in particular to a method for predicting the open capacity of a distribution transformer with high-proportion distributed photovoltaic access. Background Art

[0002] In recent years, with the gradual construction of new power systems, massive distributed renewable energy has poured in disorderly in the form of high proportions and multiple access points. The grid topology and operation mode have become increasingly complex. Phenomena such as low voltage, three-phase imbalance, and heavy overload of distribution transformers have become increasingly prominent. The operating pressure and operating costs of distribution network equipment have increased accordingly, and traditional power grids are facing huge challenges.

[0003] The open capacity of distribution transformers is an important assessment indicator for measuring the load capacity of distribution networks and the expansion and packaging business of State Grid Corporation. In order to ensure the reliability of users' electricity use and the safe and economical operation of distribution equipment, it is necessary to reasonably analyze and accurately calculate the open capacity of distribution transformers. Traditional evaluation of the open capacity of distribution transformers often adopts subjective judgment based on manual experience, or uses the highest load in the past year to make long-term planning of the open capacity of distribution transformers based on empirical formulas. There is no detailed and effective data support, and the calculation accuracy is not high, resulting in uneven distribution of load capacity and serious problems of light no-load or heavy overload of distribution transformer equipment. This method also has the following problems: First, it is difficult to accurately grasp the evolution law of the output characteristics of distributed photovoltaic users, and distributed photovoltaic users in different seasons have different output characteristics. The large difference in volt output load means that the traditional evaluation method based on historical maximum load and empirical formula cannot accurately reflect the actual operation changes of the distribution network and it is difficult to achieve the expected results; second, with the large-scale influx of distributed photovoltaic users, the topology of the distribution network has changed, and it is difficult for the empirical formula to continue to effectively measure the actual load capacity of the distribution transformer; third, the access of temporary emergency power loads requires a fast and accurate response, and manual experience and subjective judgment or measurement methods based on empirical formulas are unable to cope with the needs of temporary emergency power load access. In recent years, researchers have used artificial intelligence algorithms to make many attempts in the fields of open capacity analysis, but there is no research on the open capacity of distribution transformers suitable for large-scale distributed photovoltaic access. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing distribution transformer open capacity prediction method has the problems of slow decision-making speed, lack of data support, poor evaluation accuracy, and optimization of how to capture the dynamic changes of output characteristics of distributed photovoltaic users over time and seasons.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access, comprising collecting data and performing data preprocessing, calculating indicators that affect the open capacity of the distribution transformer; stabilizing the historical daily maximum load curve of the distribution transformer, and constructing a feature set of the daily maximum load data of the distribution transformer; designing a distribution transformer daily maximum load curve prediction model, a distributed photovoltaic output prediction model and a distribution transformer open capacity calculation model to calculate the open capacity of the distribution transformer.

[0007] As a preferred solution of the method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access described in the present invention, wherein: the data collection and data preprocessing include detection and replacement of abnormal values, collection of distribution transformer load curve data, distributed photovoltaic output data, distribution transformer rated capacity, distribution transformer cancellation user capacity, distribution transformer in-transit registration user capacity, distribution transformer and distributed photovoltaic user relationship, distribution transformer and electricity user relationship, irradiance, temperature, humidity, wind direction and pressure, for abnormal values, based on the normal distribution through the criterion, the load value beyond the positive and negative three standard deviations is regarded as an abnormal value, and the normal distribution is expressed as:

[0008]

[0009] The abnormal data is judged and eliminated by the three sigma criterion, which is expressed as:

[0010] p{(μ-3σ) <X<(μ+3σ)}

[0011] Among them, σ is the standard deviation and μ is the mean.

[0012] As a preferred solution of the method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access described in the present invention, the calculation of the index affecting the open capacity of the distribution transformer includes calculating the maximum daily load of the distribution transformer according to the preprocessed distribution transformer load curve, which is expressed as:

[0013] S max_p =maxp t∈[0,96)

[0014] Among them, S max_p is the maximum load of the distribution transformer on a daily basis, p t is the instantaneous load of the t-th point of the distribution transformer. Every 15 minutes in a day is a point, and there are 96 points in a day. According to the cancellation status of users under the distribution transformer, the total cancellation capacity under the distribution transformer is calculated, which is expressed as:

[0015]

[0016] Among them, S c is the total capacity of users canceled under the distribution transformer, s jcis the cancellation capacity of the jth user under the distribution transformer. According to the in-transit registration status of the user under the distribution transformer, the total in-transit registration capacity under the distribution transformer is calculated, which is expressed as:

[0017]

[0018] Among them, S rc is the total capacity of users in transit under the distribution transformer, s jrc It is the in-transit capacity reported by the j-th user under the distribution transformer.

[0019] As a preferred solution of the method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access described in the present invention, the stabilization of the historical distribution transformer daily maximum load curve includes checking whether the historical load data is a stable sequence through ADF. If the stable sequence is not satisfied, the difference method is used to calculate the load difference between adjacent points, which is expressed as:

[0020] f(i)=d(i+1)-d(i)

[0021] Among them, f(i) is the difference series, d(i) is the original series, and i is the i-th series. The difference series is used to eliminate the influence of trend and periodic factors on the maximum load curve of the distribution transformer, and the disordered load data is transformed into regular load data.

[0022] As a preferred solution of the method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access described in the present invention, the construction of the distribution transformer daily maximum load data feature set includes extracting features of the distribution transformer monthly maximum load through three deep learning models, extracting effective nonlinear local features of the distribution transformer load data through the convolution layer, and iteratively extracting global features from local features by the bidirectional LSTM. There are two LSTM layers in the bidirectional LSTM structure, one for processing the sequence from front to back and the other for processing the sequence from back to front. At each time step, the input is respectively transmitted to the two LSTM layers, and the output is merged. The calculation formula of the unidirectional LSTM network is expressed as:

[0023] i t =σ 1 (W ix x t +W ih h t-1 +b i )

[0024] f t =σ 1 (W fx x t +W fh h t-1 +b f )

[0025]

[0026] o t =σ 1 (W ox x t +W oh h t-1 +b o )

[0027] h t =ot⊙tanh(C t )

[0028] Among them, i t is the input gate, f t For the Gate of Forgetfulness, is the cell renewal state, C t is the cell state, o t is the output gate, h t is the hidden state, σ 1 is the sigmoid function, W i is the linear relationship coefficient parameter, b i is the bias parameter, h t-1 The state of the two output gates is combined to iteratively extract complex global features for the output of the previous layer, and the feature set of the daily maximum load data of the distribution transformer is constructed through feature extraction.

[0029] As a preferred solution of the method for predicting the open capacity of a distribution transformer with high proportion of distributed photovoltaic access described in the present invention, the distribution transformer daily maximum load curve prediction model includes combining the dense connection DenseNet model framework idea, a convolutional bidirectional long short-term memory prediction model based on the attention mechanism, and constructing a distribution transformer daily maximum load curve prediction model. The prediction model of the distribution transformer daily maximum load is constructed by combining a convolutional neural network and a long short-term memory network, extracting local features, and processing the input features through a convolutional neural network to capture the time correlation and long-term dependence in the data. The prediction model of the distribution transformer daily maximum load is expressed as:

[0030]

[0031] Among them, S pre_p is the maximum predicted load of the distribution transformer on the day, w j is the weight parameter of the convolutional neural network and the long short-term memory network, x iAs input features, combined with weather characteristics and constructed statistical and polynomial features, a multi-layer structure based on the Stacking framework is designed. The extreme gradient boosting XGBoost, long short-term memory network LSTM and independent recurrent neural network INDRNN are used as basic models to build a model suitable for distributed photovoltaic output prediction. By analyzing the relationship between multiple weather factors and historical load characteristics, a regression equation is constructed to predict the output load of distributed users under specific weather conditions, which is expressed as:

[0032]

[0033] Among them, S pre_pv is the predicted output load of distributed users at the time of the maximum predicted load of the distribution transformer on the day, α 1 , α 2 , α 3 is the weight coefficient of each model output, XGB(W,P) is the output of the extreme gradient boosting model, LSTM(M,P) is the output of the long short-term memory network model, INDRNN(W,P) is the output of the independent recurrent neural network model, W k is the kth weather feature, P j is the jth historical load characteristic, β kj are the polynomial regression coefficients between weather characteristics and historical load characteristics, λ and δ i is the weight parameter introduced, f(x i ) is the normal distribution function applied to specific input features, and ∈ is the random error term.

[0034] As a preferred solution of the method for predicting the open capacity of the distribution transformer with high proportion of distributed photovoltaic access described in the present invention, wherein: the distribution transformer open capacity calculation model includes calculating the remaining capacity of the distribution transformer according to the quota capacity and utilization coefficient of the distribution transformer, combined with the total capacity of the current cancelled users under the distribution transformer, the total capacity of the users in transit and the equivalent load coefficient, and calculating the open capacity of the distribution transformer according to the difference between the remaining capacity and the maximum predicted load of the distribution transformer on the day and the predicted output load of the photovoltaic users at the time of the maximum predicted load of the distribution transformer on the day. The distribution transformer open capacity calculation model is expressed as:

[0035] S o =S r ×α-S pre_p -S pre_pv +S c ×β-S rc ×β

[0036] Among them, S o S is the available capacity of the distribution transformer. r is the rated capacity of the distribution transformer, α is the utilization factor of the distribution transformer capacity, S pre_p is the maximum predicted load of the distribution transformer on the day, S pre_pvis the predicted output load of distributed users at the time of the maximum predicted load of the distribution transformer on the day, S c is the total capacity of users canceling their accounts under the distribution transformer, S rc is the total capacity of users in transit under the distribution transformer, and β is the equivalent load factor.

[0037] Another object of the present invention is to provide a distribution transformer open capacity prediction system with a high proportion of distributed photovoltaic access, which can stabilize the historical distribution transformer daily maximum load curve through a curve stabilization module, solving the current problem of difficulty in capturing complex relationships in the data.

[0038] As a preferred solution of the distribution transformer open capacity prediction system with high proportion of distributed photovoltaic access described in the present invention, it includes: an indicator calculation module, a curve stabilization module, and a capacity measurement module; the indicator calculation module is used to collect data and perform data preprocessing, and calculate the indicators that affect the distribution transformer open capacity; the curve stabilization module is used to stabilize the historical distribution transformer daily maximum load curve and construct a distribution transformer daily maximum load data feature set; the capacity measurement module is used to design a distribution transformer daily maximum load curve prediction model, a distributed photovoltaic output prediction model and a distribution transformer open capacity measurement model, and measure the distribution transformer open capacity.

[0039] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that the processor executes the computer program to implement a step of a method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access.

[0040] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access are implemented.

[0041] Beneficial effects of the present invention: The method for predicting the open capacity of distribution transformers with high-proportion distributed photovoltaic access provided by the present invention eliminates noise and anomalies in the data through data preprocessing, thereby improving the accuracy and reliability of model training and prediction, and improves the stability and predictability of historical data by stabilizing the historical data. The feature expression ability of the model is improved through feature extraction, and the accuracy and robustness of the prediction results are improved through the distribution transformer daily maximum load curve prediction model and the distributed photovoltaic output prediction algorithm. The utilization rate of the distribution equipment is improved through the distribution transformer open capacity calculation model. The present invention achieves better results in prediction accuracy, robustness and utilization rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0043] Figure 1 This is an overall flow chart of a method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access provided by the first embodiment of the present invention.

[0044] Figure 2 This is an overall module diagram of a distribution transformer open capacity prediction system for high-proportion distributed photovoltaic access provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0046] Example 1

[0047] Reference Figure 1 , which is an embodiment of the present invention, provides a method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access, including:

[0048] S1: Collect data and perform data preprocessing to calculate the indicators that affect the open capacity of distribution transformers.

[0049] Furthermore, data is collected and preprocessed including outlier detection and replacement.

[0050] It should be noted that the distribution transformer load curve data, distributed photovoltaic output data, distribution transformer rated capacity, distribution transformer cancellation user capacity, distribution transformer in-transit registration user capacity, distribution transformer and distributed photovoltaic user relationship, distribution transformer and electricity user relationship, irradiance, temperature, humidity, wind direction and pressure are collected. For abnormal values, based on the normal distribution criterion, the data with load values ​​exceeding the positive and negative three standard deviations are regarded as abnormal values. The normal distribution is expressed as:

[0051]

[0052] The abnormal data is judged and eliminated by the three sigma criterion, which is expressed as:

[0053] p{(μ-3σ) <X<(μ+3σ)}

[0054] Among them, σ is the standard deviation and μ is the mean.

[0055] It should also be noted that the abnormal data in the load can be identified based on the quartile values ​​of the load data through a box plot, and replaced with the mean, median, or mode. For missing values, the mean, median, or mode can be used to fill in.

[0056] Furthermore, calculating the index affecting the openable capacity of the distribution transformer includes calculating the preprocessed distribution transformer load curve.

[0057] It should be noted that the maximum daily load of the distribution transformer is calculated as:

[0058] S max_p =maxp t∈[0,96)

[0059] Among them, S max_p is the maximum load of the distribution transformer on a daily basis, p t is the instantaneous load of the t-th point of the distribution transformer. Every 15 minutes in a day is a point, and there are 96 points in a day. According to the cancellation status of users under the distribution transformer, the total cancellation capacity under the distribution transformer is calculated, which is expressed as:

[0060]

[0061] Among them, S c is the total capacity of users canceled under the distribution transformer, s jc is the cancellation capacity of the jth user under the distribution transformer. According to the in-transit registration status of the user under the distribution transformer, the total in-transit registration capacity under the distribution transformer is calculated, which is expressed as:

[0062]

[0063] Among them, S rc is the total capacity of users in transit under the distribution transformer, s jrc It is the in-transit capacity reported by the j-th user under the distribution transformer.

[0064] S2: Stabilize the historical distribution transformer daily maximum load curve and construct the distribution transformer daily maximum load data feature set.

[0065] Furthermore, the stabilization process of the historical daily maximum load curve of the distribution transformer includes testing whether the historical load data is a stationary sequence through ADF.

[0066] It should be noted that if the stationary sequence is not satisfied, the difference method is used to calculate the load difference between adjacent points, which is expressed as:

[0067] f(i)=d(i+1)-d(i)

[0068] Among them, f(i) is the difference series, d(i) is the original series, and i is the i-th series. The difference series is used to eliminate the influence of trend and periodic factors on the maximum load curve of the distribution transformer, and the disordered load data is transformed into regular load data.

[0069] It should also be noted that distributed photovoltaic users are classified according to the characteristics of weather data and historical load data. For photovoltaic users who are insensitive to meteorological changes, predictions are made based on the historical maximum similar day data. For photovoltaic users who are sensitive to meteorological changes, it is determined whether the characteristic data obeys the normal distribution. If the characteristic data obeys the normal distribution, the Pearson algorithm is used to calculate the correlation coefficient. If the characteristic data does not obey the overall distribution, the Spearman algorithm is used to calculate the correlation coefficient.

[0070] Furthermore, constructing the feature set of the daily maximum load data of the distribution transformer includes extracting features of the monthly maximum load of the distribution transformer through three deep learning models.

[0071] It should be noted that the convolution layer extracts effective nonlinear local features of the distribution transformer load data, and the bidirectional LSTM iteratively extracts global features from local features. There are two LSTM layers in the bidirectional LSTM structure, one for processing sequences from front to back, and one for processing sequences from back to front. At each time step, the input is passed to the two LSTM layers respectively, and the output is merged. The calculation formula of the unidirectional LSTM network is expressed as:

[0072] i t =σ 1 (W ix x t +W ih h t-1 +b i )

[0073] f t =σ 1 (W fx x t +W fh h t-1 +b f )

[0074]

[0075] o t =σ 1 (W ox x t +W oh h t-1 +b o )

[0076] h t =o t ⊙tanh(Ct )

[0077] Among them, i t is the input gate, f t For the Gate of Forgetfulness, is the cell renewal state, C t is the cell state, o t is the output gate, h t is the hidden state, σ 1 is the sigmoid function, W i is the linear relationship coefficient parameter, b i is the bias parameter, ht -1 The state of the two output gates is combined to iteratively extract complex global features for the output of the previous layer, and the feature set of the daily maximum load data of the distribution transformer is constructed through feature extraction.

[0078] It should also be noted that the backward LSTM network is represented as:

[0079] i' t =σ 1 (W' ixxt +W' ih h' t+1 +b' i )

[0080] f' t =σ 1 (W' fx x t +W' fh h' t+1 +b' f )

[0081]

[0082] o' t =σ 1 (W ox x t +W oh h t-1 +b o )

[0083] h' t =o' t ⊙tanh(C' t )

[0084] Attention mines deep temporal correlations, effectively reduces the loss of historical information and highlights the information of key historical time points. Through feature extraction, it effectively analyzes the peaks, troughs, cycles and other related characteristic information of distribution transformer load data to construct the feature set of distribution transformer daily maximum load data.

[0085] It should also be noted that the mechanism of Attention is that in the sequence X: [X 1 ,X 2 ,……,X n ] to find information related to a feature, first introduce a representation information related to the task, denoted as Q, and calculate the input sequence X through the scoring function: [X 1 ,X 2 ,……,X n ] and Q, the common scoring functions are dot product operation, scaled dot product operation and bilinear operation, which are expressed as:

[0086] Similarity(Q,X i )=Q·X i =X i T Q

[0087]

[0088] Similarity(Q,X i )=X i T WQ

[0089] Among them, W is the network parameter, d is the dimension of the input information, and its value range varies according to the specific generation method. A calculation method similar to Softmax is introduced to convert the calculated score into a numerical value, which is expressed as:

[0090]

[0091] Finally, the attention distribution a i , and the input sequence X: [X 1 ,X 2 ,……,X n ] to summarize, the formula is:

[0092]

[0093] Finally, based on the results of correlation analysis, prediction features such as statistical features and polynomial features are constructed for strongly correlated features such as irradiance and temperature. Statistical features are the mean and standard deviation of constructed features, and polynomial features are combinations of features such as irradiance and temperature to construct a distributed photovoltaic data feature set.

[0094] S3: Design the distribution transformer daily maximum load curve prediction model, distributed photovoltaic output prediction model and distribution transformer open capacity calculation model to calculate the distribution transformer open capacity.

[0095] Furthermore, the distribution transformer daily maximum load curve prediction model includes the idea of ​​combining the densely connected DenseNet model framework.

[0096] It should be noted that the convolutional bidirectional long short-term memory prediction model based on the attention mechanism is used to construct a distribution transformer daily maximum load curve prediction model. The prediction model of the distribution transformer daily maximum load is constructed by combining the convolutional neural network and the long short-term memory network. The local features are extracted, and the input features are processed by the convolutional neural network to capture the time correlation and long-term dependence in the data. The prediction model of the distribution transformer daily maximum load is expressed as:

[0097]

[0098] Among them, S pre_p is the maximum predicted load of the distribution transformer on the day, w j is the weight parameter of the convolutional neural network and the long short-term memory network, x i As input features, combined with weather characteristics and constructed statistical and polynomial features, a multi-layer structure based on the Stacking framework is designed. The extreme gradient boosting XGBoost, long short-term memory network LSTM and independent recurrent neural network INDRNN are used as basic models to build a model suitable for distributed photovoltaic output prediction. By analyzing the relationship between multiple weather factors and historical load characteristics, a regression equation is constructed to predict the output load of distributed users under specific weather conditions, which is expressed as:

[0099]

[0100] Among them, S pre_pv is the predicted output load of distributed users at the time of the maximum predicted load of the distribution transformer on the day, α 1 , α 2 , α 3 is the weight coefficient of each model output, XGB(W,P) is the output of the extreme gradient boosting model, LSTM(W,P) is the output of the long short-term memory network model, INDRNN(W,P) is the output of the independent recurrent neural network model, W k is the kth weather feature, P j is the jth historical load characteristic, β kj are the polynomial regression coefficients between weather characteristics and historical load characteristics, λ and δ i is the weight parameter introduced, f(x i ) is the normal distribution function applied to specific input features, and ∈ is the random error term.

[0101] Furthermore, the distribution transformer open capacity calculation model includes the distribution transformer quota capacity and utilization coefficient.

[0102] It should be noted that the remaining capacity of the distribution transformer is calculated by combining the total capacity of the current users who have cancelled their accounts, the total capacity of the users who are in transit and the equivalent load factor. The available capacity of the distribution transformer is calculated based on the difference between the remaining capacity and the maximum predicted load of the distribution transformer on the day and the predicted output load of the photovoltaic users at the time of the maximum predicted load of the distribution transformer on the day. The calculation model of the available capacity of the distribution transformer is expressed as follows:

[0103] S o =S r ×α-S pre_p -S pre_pv +S c ×β-S rc ×β

[0104] Among them, S o S is the available capacity of the distribution transformer. r is the rated capacity of the distribution transformer, α is the utilization factor of the distribution transformer capacity, S pre_p is the maximum predicted load of the distribution transformer on the day, S pre_pv is the predicted output load of distributed users at the time of the maximum predicted load of the distribution transformer on the day, S c is the total capacity of users canceling their accounts under the distribution transformer, S rc is the total capacity of users in transit under the distribution transformer, and β is the equivalent load factor.

[0105] Example 2

[0106] An embodiment of the present invention provides a method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0107] Twenty distribution transformers in a certain area were selected, and their archival data such as distribution transformer load data, distributed photovoltaic output data, distribution transformer rated capacity, user relations, and weather data such as irradiance, temperature, and humidity were recorded. Smart meters, environmental monitoring instruments and other equipment were used to collect data every 15 minutes for one month to ensure the continuity and integrity of the data. The collected data was cleaned, and the normal distribution and box plot method were used to identify and process outliers. The mean and median were used to fill in missing values ​​to ensure the accuracy and consistency of the data.

[0108] According to the preprocessed data, key indicators such as the maximum daily load of the distribution transformer, the in-transit capacity under the distribution transformer, and the capacity of the distribution transformer under the distribution transformer are calculated respectively. The daily maximum load value of each distribution transformer during the test period is counted. According to the user profile data, the user capacity of each distribution transformer added and cancelled during the test period is counted. The historical distribution transformer daily maximum load curve is stabilized, and the ADF test is used to determine the stability of the load data. If the data is not stable, the difference method is used to calculate the load difference between adjacent points to eliminate trends and periodic fluctuations. The convolutional layer, bidirectional LSTM and attention mechanism are used to extract the load data features and construct The daily maximum load data feature set of the distribution transformer is used. The convolution layer extracts nonlinear local features, the bidirectional LSTM extracts global features from local features, and the attention mechanism captures time correlation. The distribution transformer daily maximum load curve prediction model and distributed photovoltaic output prediction algorithm are designed. Using the aforementioned data feature set, based on the DenseNet framework and Stacking structure, combined with XGBoost, LSTM and INDRNN for training, the trained model is used to predict the daily maximum load of the distribution transformer and the distributed photovoltaic output, and the correction is made based on the latest weather data, and finally the open capacity of the distribution transformer is calculated.

[0109] As shown in Table 1, in terms of data preprocessing and key indicator calculation, our invention eliminates noise and anomalies in the data through a variety of methods, thereby improving the accuracy of load forecasting. Compared with the traditional method that relies on empirical formulas, the method of the present invention uses a deep learning model, which can better capture the complex relationships in the data. Therefore, the load forecasting results of our invention are closer to the actual load, and the error is significantly reduced.

[0110] The existing technology often has problems of information lag and inaccurate calculation when processing the daily maximum load forecast of distribution transformers and the distributed photovoltaic output forecast. Our invention can timely reflect the daily maximum load of distribution transformers and the distributed photovoltaic output through dynamic collection and real-time update of user archive data, thereby improving the real-time and accuracy of capacity measurement. For example, in the data of the daily maximum load of distribution transformers and the distributed photovoltaic output, the method of the present invention can accurately reflect the changes of each distribution transformer and ensure the real-time nature of the measurement results. The traditional method is relatively simple to stabilize the historical data and is easily affected by data fluctuations. The method of the present invention adopts ADF detection The verification and difference method is combined with the deep learning model to extract features from the data, which effectively eliminates trends and periodic fluctuations in the data, making the data more stable and regular. This processing method improves the stability of model training and the accuracy of prediction. The capacity measurement models in the existing technology often lack scientific basis and are easily affected by subjective factors. The method of the present invention designs a complex deep learning prediction model and comprehensively considers multiple factors such as load, weather and user capacity to make the measurement results more scientific and accurate. For example, in the photovoltaic output prediction, our invention combines the latest weather data and historical characteristics to significantly improve the accuracy of the prediction.

[0111] Table 1 Experimental data table

[0112] Distribution transformer 1 Distribution transformer 2 Distribution 3 Transformer 4 Transformer 5 Maximum daily load of distribution transformer 120kW 115kW 125kW 118kW 122kW In-transit capacity 30kW 28kW 35kW 33kW 31kW Account cancellation capacity 10kW 12kW 8kW 9kW 11kW Weather Temperature 25℃ 26℃ 24℃ 27℃ 25℃ Photovoltaic output forecast 50kW 48kW 55kW 52kW 50kW

[0113] Example 3

[0114] Reference Figure 2 , which is an embodiment of the present invention, provides a distribution transformer open capacity prediction system with a high proportion of distributed photovoltaic access, including: an indicator calculation module, a curve stabilization module, and a capacity measurement module.

[0115] The indicator calculation module is used to collect data and perform data preprocessing to calculate the indicators that affect the open capacity of the distribution transformer; the curve stabilization module is used to stabilize the historical distribution transformer daily maximum load curve and construct the distribution transformer daily maximum load data feature set; the capacity measurement module is used to design the distribution transformer daily maximum load curve prediction model, distributed photovoltaic output prediction model and distribution transformer open capacity measurement model to calculate the distribution transformer open capacity.

[0116] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0117] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0118] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0119] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the open capacity of distribution transformers with high proportion of distributed photovoltaic access, characterized in that: include: Collect data and perform data preprocessing to calculate the indicators that affect the available capacity of distribution transformers; The historical distribution transformer daily maximum load curve is stabilized to construct the distribution transformer daily maximum load data feature set; Design the distribution transformer daily maximum load curve prediction model, distributed photovoltaic output prediction model and distribution transformer open capacity calculation model to calculate the distribution transformer open capacity.

2. The method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in claim 1 is characterized by: The data collection and data preprocessing include detecting and replacing abnormal values, collecting distribution transformer load curve data, distributed photovoltaic output data, distribution transformer rated capacity, distribution transformer cancellation user capacity, distribution transformer in-transit registration user capacity, distribution transformer and distributed photovoltaic user relationship, distribution transformer and electricity user relationship, irradiance, temperature, humidity, wind direction and pressure. For abnormal values, based on the normal distribution criterion, the data with load values ​​exceeding the positive and negative three standard deviations are regarded as abnormal values. The normal distribution is expressed as: The abnormal data is judged and eliminated by the three sigma criterion, which is expressed as: p{(μ-3σ) <X<(μ+3σ)} Among them, σ is the standard deviation and μ is the mean.

3. The method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in claim 2 is characterized by: The calculation of the index affecting the open capacity of the distribution transformer includes calculating the maximum daily load of the distribution transformer according to the pre-processed distribution transformer load curve, which is expressed as: S max_p =maxp t∈[0,96) Among them, S max_p is the maximum load of the distribution transformer on a daily basis, p t is the instantaneous load of the t-th point of the distribution transformer. Every 15 minutes in a day is a point, and there are 96 points in a day. According to the cancellation status of users under the distribution transformer, the total cancellation capacity under the distribution transformer is calculated, which is expressed as: Among them, S c is the total capacity of users canceling accounts under the distribution transformer, s jc is the cancellation capacity of the jth user under the distribution transformer. According to the in-transit registration status of the user under the distribution transformer, the total in-transit registration capacity under the distribution transformer is calculated, which is expressed as: Among them, S rc is the total capacity of users in transit under the distribution transformer, s jrc It is the in-transit capacity reported by the j-th user under the distribution transformer.

4. The method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in claim 3 is characterized by: The stabilization process of the historical distribution transformer daily maximum load curve includes checking whether the historical load data is a stable sequence by ADF. If it does not meet the stable sequence, the difference method is used to calculate the load difference between adjacent points, which is expressed as: f(i)=d(i+1)-d(i) Among them, f(i) is the difference series, d(i) is the original series, and i is the i-th series. The difference series is used to eliminate the influence of trend and periodic factors on the maximum load curve of the distribution transformer, and the disordered load data is transformed into regular load data.

5. The method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in claim 4 is characterized by: The construction of the daily maximum load data feature set of the distribution transformer includes extracting features of the monthly maximum load of the distribution transformer through three deep learning models, extracting effective nonlinear local features of the distribution transformer load data through the convolution layer, and iteratively extracting global features from local features by the bidirectional LSTM. There are two LSTM layers in the bidirectional LSTM structure, one for processing the sequence from front to back and the other for processing the sequence from back to front. At each time step, the input is passed to the two LSTM layers respectively, and the output is merged. The calculation formula of the unidirectional LSTM network is expressed as: i t =σ1(W ix x t +W ih h t-1 +b i ) f t =σ1(W fx x t +W fh h t-1 +b f ) o t =σ1(W ox x t +W oh h t-1 +b o ) h t =o t ⊙tanh(C t ) Among them, i t is the input gate, f t For the Gate of Forgetfulness, is the cell renewal state, C t is the cell state, o t is the output gate, h t is the hidden state, σ1 is the sigmoid function, W i is the linear relationship coefficient parameter, b i is the bias parameter, h t-1 The state of the two output gates is combined to iteratively extract complex global features for the output of the previous layer, and the feature set of the daily maximum load data of the distribution transformer is constructed through feature extraction.

6. The method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in claim 5, characterized in that: The distribution transformer daily maximum load curve prediction model includes combining the dense connection DenseNet model framework idea, and building a convolution bidirectional long short-term memory prediction model based on the attention mechanism to construct a distribution transformer daily maximum load curve prediction model. The distribution transformer daily maximum load prediction model is constructed by combining a convolutional neural network and a long short-term memory network, extracting local features, and processing the input features through a convolutional neural network to capture the time correlation and long-term dependence in the data. The distribution transformer daily maximum load prediction model is expressed as: Among them, S pre_p is the maximum predicted load of the distribution transformer on a daily basis, W j is the weight parameter of the convolutional neural network and the long short-term memory network, x i As input features, combined with weather characteristics and constructed statistical and polynomial features, a multi-layer structure based on the Stacking framework is designed. The extreme gradient boosting XGBoost, long short-term memory network LSTM and independent recurrent neural network INDRNN are used as basic models to build a model suitable for distributed photovoltaic output prediction. By analyzing the relationship between multiple weather factors and historical load characteristics, a regression equation is constructed to predict the output load of distributed users under specific weather conditions, which is expressed as: Among them, S pre_pv is the predicted output load of distributed users at the maximum predicted load moment of the distribution transformer day, α1, α2, α3 are the weight coefficients of the output of each model, XGB(W,P) is the output of the extreme gradient boosting model, LSTM(W,P) is the output of the long short-term memory network model, INDRNN(W,P) is the output of the independent recurrent neural network model, W k is the kth weather feature, P j is the jth historical load characteristic, β kj are the polynomial regression coefficients between weather characteristics and historical load characteristics, λ and δ i is the weight parameter introduced, f(x i ) is the normal distribution function applied to specific input features, and ∈ is the random error term.

7. The method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in claim 6, characterized in that: The distribution transformer openable capacity calculation model includes calculating the remaining capacity of the distribution transformer according to the distribution transformer quota capacity and utilization coefficient, combined with the total capacity of the current cancelled users under the distribution transformer, the total capacity of the users who are in transit and the equivalent load coefficient, and calculating the open capacity of the distribution transformer according to the difference between the remaining capacity and the maximum predicted load of the distribution transformer on the day and the predicted output load of the photovoltaic users at the time of the maximum predicted load of the distribution transformer on the day. The distribution transformer openable capacity calculation model is expressed as: S o =S r ×α-S pre_p -S pre_pv +S c ×β-S rc ×β Among them, S o S is the available capacity of the distribution transformer. r is the rated capacity of the distribution transformer, α is the utilization factor of the distribution transformer capacity, S pre_p is the maximum predicted load of the distribution transformer on the day, S pre_pv is the predicted output load of distributed users at the time of the maximum predicted load of the distribution transformer on the day, S c is the total capacity of users canceling their accounts under the distribution transformer, S rc is the total capacity of users in transit under the distribution transformer, and β is the equivalent load factor.

8. A system using the method for predicting the available capacity of a distribution transformer with high proportion of distributed photovoltaic access as claimed in any one of claims 1 to 7, characterized in that: It includes index calculation module, curve stabilization module and capacity measurement module; The index calculation module is used to collect data and perform data preprocessing, and calculate the index that affects the open capacity of the distribution transformer; The curve stabilization module is used to stabilize the historical distribution transformer daily maximum load curve and construct a distribution transformer daily maximum load data feature set; The capacity calculation module is used to design a distribution transformer daily maximum load curve prediction model, a distributed photovoltaic output prediction model and a distribution transformer open capacity calculation model to calculate the distribution transformer open capacity.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting the open capacity of distribution transformers with high-proportion distributed photovoltaic access described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the open capacity of a distribution transformer with a high proportion of distributed photovoltaic access as described in any one of claims 1 to 7 are implemented.

Images