Method for improving two-way bearing capacity of transformer area based on Gaussian mixture model

Abstract

The invention provides a method for improving the two-way bearing capacity of a transformer area based on a Gaussian mixture model. The method comprises the steps that S1, transformer area power distribution network operation data are collected, and all kinds of data are preprocessed; s2, Gaussian mixture model construction and training: constructing a feature vector representing the operation state of the transformer area for each time section based on the collected historical operation data of the transformer area power distribution network, matching the operation state of the transformer area for each time point, and training a Gaussian mixture model by using an obtained data set; and S3, based on the historical operation data and the source load development trend of the transformer area power distribution network, predicting the load demand and the distributed power supply output of the transformer area, obtaining the feature vector of the transformer area operation state at the future T moment, and inputting the trained Gaussian mixture model to determine the transformer area operation state at the future T moment. Obtaining risks confronted by transformer area operation in a future period of time and risk duration; and S4, on the basis of risks faced by transformer area operation in a period of time in the future, matching an optimal strategy from the strategy library and executing the optimal strategy.

Description

Technical Field

[0001] This invention belongs to the field of power system and automation technology, specifically relating to a method for improving the bidirectional carrying capacity of transformer substations based on a Gaussian mixture model. Background Technology

[0002] With the deepening implementation of the "dual-carbon" strategic goals, distributed energy sources, represented by photovoltaics and wind power, are experiencing explosive growth in the scale of grid connection, especially on the user side. Traditional low-voltage distribution substations are gradually evolving from a radial network with unidirectional passive power supply to a bidirectional active power supply network. This fundamental transformation has brought unprecedented challenges to the safe, stable, and efficient operation of these substations, exposing many limitations of traditional technologies and management models, specifically in the following aspects:

[0003] (1) The operating characteristics of the transformer substation have undergone fundamental changes, and traditional design standards are facing challenges.

[0004] Traditional distribution transformer planning and design, equipment selection (such as transformer capacity and line diameter), and protection configuration are all based on a "top-down" unidirectional power flow model. When a large number of distributed photovoltaic systems send power back "from the bottom up," it can lead to a series of unprecedented technical problems, such as reverse power flow, overvoltage, and heavy overload of transformers and lines.

[0005] (2) Existing monitoring and identification methods are outdated and cannot accurately detect bidirectional carrying problems.

[0006] Relying on traditional SCADA / AMI data: Existing systems mainly rely on voltage, current, and power data from distribution automation and advanced measurement systems. These data often suffer from coarse sampling granularity (usually more than 15 minutes) and high data latency, making it difficult to capture instantaneous voltage fluctuations and impacts caused by photovoltaic power output fluctuations.

[0007] "Blind commissioning" and "post-event analysis" models: Current operation and maintenance are mostly based on threshold over-limit alarms, which is a passive response. There is a lack of forward-looking assessment and refined diagnosis of the bidirectional carrying capacity of the transformer area. Operation and maintenance personnel cannot accurately answer key questions such as "How much photovoltaic power can the transformer area currently accept?" and "Where exactly are the risk points of voltage over-limit?", and can only take rigid "one-size-fits-all" control (such as forced curtailment), resulting in the waste of green energy.

[0008] (3) Existing governance and improvement methods are simplistic, uneconomical, and lack coordination.

[0009] Traditional capacity expansion methods: Faced with insufficient carrying capacity, the most direct solution is to replace transformers with larger ones and thicken transmission lines. This method involves huge investment costs, complex engineering implementation, and widespread power outage impacts. Moreover, it is a crude capacity expansion method that relies on the load and fails to fully utilize the flexible adjustment potential of the existing network, resulting in poor economic efficiency.

[0010] Single-device management: Some solutions use the addition of reactive power compensation devices or voltage regulators to improve voltage quality. However, these measures are often isolated and localized, failing to coordinate with the flexible resources within the transformer area, resulting in low overall optimization efficiency.

[0011] Lack of Flexible Resource Aggregation and Control: With the widespread adoption of massive numbers of smart terminals, distribution areas possess enormous controllable potential. However, existing technologies lack effective aggregation models and collaborative control strategies, failing to transform these distributed flexible resources into effective means to enhance the bidirectional carrying capacity of distribution areas.

[0012] In summary, existing technologies lack a comprehensive diagnostic method capable of accurately and in real-time identifying weaknesses in distribution transformer areas (such as voltage overruns and equipment overloads) under bidirectional power flow conditions. Furthermore, there is a lack of a method to economically, efficiently, and collaboratively utilize various flexible resources to dynamically enhance the bidirectional carrying capacity of distribution transformer areas based on the diagnostic results. Therefore, an innovative technical solution is urgently needed to address these issues and ensure the safe, efficient, and intelligent operation of distribution transformer areas with a high proportion of distributed energy access. Summary of the Invention

[0013] The purpose of this invention is to propose a method for improving the bidirectional load-bearing capacity of photovoltaic power distribution areas based on Gaussian mixture models. This method addresses the bidirectional load-bearing problem faced by photovoltaic power distribution areas by automatically identifying the operating modes of the power distribution area using Gaussian mixture models (GMM), accurately identifying abnormal states that exceed the normal mode, and automatically matching the optimal, multi-resource coordinated control strategy through problem-strategy matching to flexibly improve the load-bearing capacity and avoid the high cost of traditional "hard expansion".

[0014] To achieve the above objectives, the technical solution of the present invention is: a method for improving the bidirectional bearing capacity of a transformer substation based on a Gaussian mixture model, specifically including the following steps:

[0015] S1. Collect the operation data of the distribution network in the transformer area, including transformer voltage, current, active power, distributed power output, and time-series data of user electricity load in the transformer area; and preprocess the various types of data.

[0016] S2. Construction and training of Gaussian mixture model: Based on the collected historical operation data of the distribution network of the transformer substation, a feature vector representing the operation status of the transformer substation is constructed for each time section, and the operation status of the transformer substation is matched for each time point. The obtained dataset is used to train the Gaussian mixture model, which includes multiple single Gaussian models corresponding to various transformer substation operation states.

[0017] S3. Based on the historical operation data of the distribution network in the transformer area and the development trend of the source and load, predict the load demand and distributed power output of the transformer area, obtain the feature vector of the transformer area operation status at time T in the future, and input it into the trained Gaussian mixture model to determine the transformer area operation status at time T in the future, and obtain the risks faced by the transformer area operation in the future and the duration of the risks.

[0018] S4. Based on the risks that the transformer area will face in the near future, match the optimal strategy from the strategy library and execute it.

[0019] Preferably, the preprocessing of various types of data specifically includes:

[0020] Data cleaning: Use the 3σ criterion to remove outliers and use linear interpolation to fill in missing values;

[0021] Time series data are standardized by using a unified timestamp.

[0022] Preferably, feature vectors are constructed for each time segment. The details are as follows:

[0023]

[0024] In the formula, P loadi (t) represents the total electrical load of the i-th medium-voltage line at time t. The output of the distributed power source at time t.

[0025] Preferably, if the transformer area is connected to distributed photovoltaic power, then the feature vector of each time segment is... Represented as:

[0026]

[0027] In the formula, P PVj X(t) represents the output of the j-th photovoltaic unit at time t, and the dimension of the feature vector X(t) is i+j.

[0028] Preferably, if the transformer area is also connected to energy storage and distributed wind power, then the feature vector of each time segment... Represented as:

[0029]

[0030] In the formula, PWindk (t) represents the output of the k-th wind turbine at time t, P Storagel X(t) represents the output of the l-th energy storage unit at time t, and the dimension of the feature vector X(t) is i+j+k+l.

[0031] Preferably, the operating status of the transformer substation includes seven operating states: transformer forward heavy overload but voltage does not exceed the limit, transformer forward heavy overload and voltage exceeds the lower limit, transformer voltage exceeds the lower limit but not forward heavy load, transformer reverse heavy overload but voltage does not exceed the limit, transformer reverse heavy overload and voltage exceeds the limit, transformer voltage exceeds the upper limit but not reverse heavy load, and normal operation.

[0032] The preferred method for training a Gaussian mixture model is as follows:

[0033] S2.1. For each single Gaussian model in the Gaussian mixture model, initialize the mean μ. i' Standard deviation σ i' Component weights α i' μ i' Initialize σ randomly between (0, 1). i' Initialize as a positive unit definite matrix, α i' Initialize to 1 / I, where i' is the index of a single Gaussian model, i'=1,2,...,I, and I is the number of single Gaussian models;

[0034] S2.2 Calculation using the expected maximum algorithm, including E-step and M-step:

[0035] Step E: For sample point X n Calculate the operating states of the transformer substations S1, S2, ..., S3. I The i'th state S i' probability :

[0036]

[0037] In the formula, j' is used to iterate through each single Gaussian model. Here is the probability density function:

[0038]

[0039] M-step: Use the probabilities estimated in the E-step to update the Gaussian mixture model parameters μ in the iteration. i' σ i' α i' Then, the updated parameters are reused in step E; the parameter update formula is as follows:

[0040]

[0041]

[0042]

[0043] In the formula, N is the total number of samples in the training set;

[0044] S2.3 Repeat the E-step and M-step calculations until convergence to a local optimum to obtain the parameters of the Gaussian mixture model, where the mean value μ of the single Gaussian model is... i' It is regarded as the distribution center, that is, the i'th type of station area operation state.

[0045] Preferably, in step S3, the feature vector X(t+T) of the operating state of the transformer area at time T is input into the trained Gaussian mixture model, and the feature vector X(t+T) is expressed as:

[0046]

[0047] Calculate the posterior probability of the feature vector X(t+T) belonging to the operating state of each transformer area and sort them by size. Determine the operating state of the transformer area at time T based on the maximum posterior probability.

[0048] Preferably, the strategy library is constructed as follows:

[0049] (1) No measures will be taken when the duration of heavy load or voltage over-limit in the transformer area is less than 5% of the predicted period.

[0050] (2) When the duration of heavy load or voltage over-limit in the distribution area exceeds 50% of the predicted period, the distribution transformer capacity of the distribution area shall be expanded to solve the bidirectional load problem.

[0051] (2) When the duration of overload or voltage exceeding the limit in the transformer area exceeds 5% of the predicted period but is less than 50% of the predicted period:

[0052] To address the issue of positive heavy overload, measures should be taken according to the following priority: first, load regulation should be implemented, that is, the load during peak hours should be transferred to the off-peak hours; second, load transfer should be implemented; and finally, the capacity of distribution transformers should be expanded.

[0053] To address the voltage exceeding the limit issue: SVG reactive power compensation combined with distributed photovoltaic four-way adjustable voltage is used to solve the problem of voltage exceeding the upper limit; voltage regulator regulation combined with load transfer is used to solve the problem of voltage exceeding the lower limit.

[0054] To address the impact of reverse power flow, measures should be taken according to the following priority: first, adjust the grid-connected control strategy of distributed photovoltaic power generation and reduce photovoltaic output through the four-way adjustable distributed photovoltaic system; second, adjust photovoltaic output through distribution and storage; and finally, expand the capacity of distribution transformers.

[0055] Preferably, after the strategy is executed, the operation data of the distribution network in the transformer area is collected again to evaluate the effect of the improvement of the bidirectional carrying capacity of the transformer area; at the same time, the new operation data of the distribution network in the transformer area is added to the dataset, and the Gaussian mixture model is retrained regularly to realize the self-evolution of the model and adapt to the changes in the structure of the transformer area.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] This invention fully integrates the bidirectional load-bearing requirements of photovoltaic power distribution areas and applies Gaussian Mixture Model (GMM) to identify load-bearing capacity issues. It proposes a method for improving the bidirectional load-bearing capacity of power distribution areas based on GMM, following a "data acquisition-model construction-problem identification-strategy formulation" approach. First, feature vectors are constructed based on historical source-load operation data, and seven operating states are defined according to the actual operation of the power grid in the distribution area, corresponding to a risk set. Second, utilizing the historical data clustering and pattern learning capabilities of the GMM, model parameters are optimized through a training set to construct an analytical model adapted to the operating characteristics of the distribution area. The accuracy of the model is verified through a test set. Then, using the source-load prediction results of the distribution area as input, the GMM model accurately identifies the operational risks and probabilities of the distribution area at different time points. Finally, based on the duration of the risks, a load-bearing capacity improvement strategy library is constructed. Optimal capacity improvement strategies are generated through intelligent problem-strategy matching, effectively improving the bidirectional load-bearing capacity of the distribution area and achieving safe, green, and economical operation. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the power grid structure in the photovoltaic distribution area;

[0059] Figure 2 A flowchart for the problem-policy automatic policy process;

[0060] Figure 3 Flowchart of a method for identifying bidirectional load-bearing problems and improving the capacity of transformer substations based on Gaussian mixture models;

[0061] Figure 4 This is a flowchart of a method for identifying and enhancing the bidirectional load-bearing capacity of transformer substations based on a Gaussian mixture model. Detailed Implementation

[0062] The following is in conjunction with the appendix Figure 1-4 The technical solution of the present invention will be described in detail below.

[0063] Gaussian Mixture Models (GMMs) are an effective analytical representation of uncertainty factors, enabling accurate modeling of non-Gaussian random variables. They utilize the Expectation-Maximization (EM) algorithm to estimate uncertainty parameters and are suitable for characterizing the uncertainty of non-Gaussian variables in scenarios with small sample sizes or missing data. GMMs offer significant advantages in handling the uncertainty and correlation of non-Gaussian variables. Compared to traditional Weiber and beta distribution models, GMMs possess superior mathematical properties such as linear invariance, superposition, and conditional probability invariance.

[0064] This invention proposes a method for improving the bidirectional load-bearing capacity of photovoltaic (PV) distribution areas based on Gaussian mixture models (GMMs). Addressing the bidirectional load-bearing problem faced by PV distribution areas, it utilizes the historical data clustering and pattern learning capabilities of GMMs to accurately identify bidirectional load-bearing risks, predict the probability of each risk occurring, and combine the risk probability with available resources in the distribution area. Through problem-strategy automatic matching, it generates the optimal capacity improvement strategy to enhance the bidirectional load-bearing capacity of the distribution area, achieving safe, green, and economical operation. The working principle and steps of this invention's method for identifying and improving the bidirectional load-bearing capacity of PV distribution areas based on Gaussian mixture models are described below:

[0065] (1) Data acquisition and preprocessing

[0066] 1) Data Collection

[0067] The data collected includes the following: transformer voltage, current, active power, distributed photovoltaic power output, and time-series data of user electricity load in the distribution area. The data collection period is one year (the collection period can be flexibly adjusted according to actual conditions; one year is used as an example here), and the collection frequency is once per hour.

[0068] 2) Data processing

[0069] Preprocessing of various types of data: First, data cleaning is performed, outlier data is removed using the 3σ criterion, and missing values ​​are filled in using linear interpolation. Second, time series data are standardized with unified timestamps.

[0070] Assume the collected data sequence is {x1, x2, ..., x} m ...,x M The mean of the data is μ. x The standard deviation is σ x .

[0071]

[0072]

[0073] Set the data validity range to [μ] x -3σ x ,μx +3σ x After removing outliers, linear interpolation is used to fill the gaps.

[0074]

[0075] (2) Construction and training of Gaussian mixture model

[0076] 1) Model Building

[0077] ① Constructing feature vectors

[0078] By combining historical distribution network operation data, a feature vector is constructed: a feature vector X(t) is constructed for each time segment that can comprehensively characterize the operating status of the distribution area.

[0079]

[0080] In the formula, P loadi (t) represents the total electrical load of the i-th medium-voltage line at time t, P PVj X(t) represents the output of the j-th photovoltaic unit at time t, and the dimension of the feature vector X(t) is i+j.

[0081] It should be noted that if the transformer area has energy storage and distributed wind power access, the feature vector X(t) needs to be supplemented with corresponding elements, as shown below:

[0082]

[0083] In the formula, P Windk (t) represents the output of the k-th wind turbine at time t, P Storagel X(t) represents the output of the l-th energy storage unit at time t, and the dimension of the feature vector X(t) is i+j+k+l.

[0084] ② Construct an operational risk set

[0085] Based on the actual operation of the photovoltaic power grid in the distribution area, and addressing the bidirectional load-bearing problem, the operating status of the distribution area is divided into seven states: distribution transformer forward heavy overload but voltage does not exceed the limit; distribution transformer forward heavy overload and voltage exceeds the lower limit; distribution transformer voltage exceeds the lower limit but not forward heavy load; distribution transformer reverse heavy overload but voltage does not exceed the limit; distribution transformer reverse heavy overload and voltage exceeds the limit; distribution transformer voltage exceeds the upper limit but not reverse heavy load; and normal operation. The explanations for each state are as follows:

[0086] Status 1: Distribution transformer forward overload but voltage does not exceed limits: Due to excessive load demand in the distribution area, the forward load rate of the distribution transformer exceeds 80%, but the voltage is within the normal operating range;

[0087] State 2: Distribution transformer forward overload and voltage below the lower limit: Due to excessive load demand in the distribution area, the forward load rate of the distribution transformer exceeds 80%, and the voltage is below the lower limit (i.e., 10×(1-7%)kV).

[0088] State 3: Distribution transformer voltage is below the lower limit but not under forward overload: The distribution transformer voltage is below the lower limit (i.e., 10×(1-7%)kV), but the distribution transformer is not under overload;

[0089] Status 4: Distribution transformer reverse overload but voltage does not exceed limit: The distributed power supply in the distribution area is too powerful, causing the reverse load rate of the distribution transformer to exceed 80%, but the voltage is within the normal operating range.

[0090] Status 5: Distribution transformer reverse overload and voltage exceeds limit: The distributed power supply output in the distribution area is too high, causing the reverse load rate of the distribution transformer to exceed 80% and the voltage to exceed the upper limit (i.e., 10×(1+7%)kV).

[0091] Status 6: Distribution transformer voltage exceeds the upper limit but does not reverse overload: The distribution transformer voltage exceeds the upper limit (i.e., 10×(1+7%)kV), but the distribution transformer does not experience overload;

[0092] Status 7: Normal operation: No overload or voltage over-limit issues occur in the transformer area, and all indicators are within the normal operating range.

[0093] It should be noted that forward overload and voltage exceeding the lower limit issues only occur when the transformer area is overloaded; reverse overload and voltage exceeding the upper limit issues only occur when the transformer area's power output is excessive. There is no linear correlation between overload and voltage exceeding the limit.

[0094] Based on the seven operating states of the power grid in the distribution area, the operating state at each time point is matched to construct a full-time operating risk set Y.

[0095]

[0096] In the formula, S(t) represents the operating status of the transformer area at time t, and the risk set Y for the entire operating period contains 8760 data points. It should be noted that the feature vector and the risk set correspond in the time domain.

[0097] 2) Model Training

[0098] ① Construct training and test sets

[0099] The historical data was divided into training set X in an 8:2 ratio. train and test set X test This ratio can be adjusted according to actual needs.

[0100] ② Training process

[0101] The main parameters of the Gaussian mixture model are the mean μ, standard deviation σ, and component weights α. μ reflects the center frequency position of each frequency band, σ reflects the bandwidth, and α represents the proportion of each modal component. Based on the seven operating states of the distribution network, the Gaussian mixture model consists of seven single Gaussian models.

[0102] Step 1: For each single Gaussian model in the Gaussian mixture model, initialize the mean μ of the 7 Gaussian distributions. i' Standard deviation σ i' Component weights α i' μ i' Initialize σ randomly between (0, 1). i' Initialize as a positive unit definite matrix, α i' Take 1 / 7, i'=1,2,...,7.

[0103] Step 2: Calculate using the Expected Maximum Value Algorithm (EM algorithm), which includes E-step and M-step.

[0104] Step E: For sample point X n Calculate the state S that the sample belongs to among states S1, S2, ..., S7, i'-th state S. i' probability :

[0105]

[0106] In the formula, is the probability density function.

[0107]

[0108] M-step: Use the probabilities estimated in the E-step to update the Gaussian mixture model parameters μ in the iteration. i' σ i' α i' Then, the newly obtained parameters are reused in the E-step. The parameter update formula is as follows:

[0109]

[0110]

[0111]

[0112] In the formula, N is the total number of samples in the training set.

[0113] Step 3: Repeat the E-step and M-step calculations until convergence to a local optimum, thus obtaining the parameters of the GMM, where the mean value μ of the single Gaussian model is... i' This can be considered the center of the distribution, i.e., the i'th type of station operation state.

[0114] ③ Training result test

[0115] For the trained GMM model, using the test set X test Perform performance testing to verify the model's accuracy. If the test accuracy is below 90%, increase the number of samples in the test set (e.g., change the ratio of training set to test set data to 9:1) and retrain the model.

[0116] (3) Identification of bidirectional carrying capacity issues in the transformer area

[0117] ①Source and load prediction

[0118] Based on historical operating data and power generation trends of the transformer substation, the load demand and distributed power output of the substation are predicted, resulting in the feature vector X(t+T) of the substation's operating state at time T in the future:

[0119]

[0120] In the formula, the value of T is [1,2,...,7].

[0121] ②State recognition

[0122] Input the feature vector X(t+T) of the future operating state of the transformer area into the trained GMM, calculate the posterior probability of the data belonging to each component, sort the probability values ​​of each group in order of size, and take the state with the largest probability value as the operating state of the transformer area at that time, thereby obtaining the risks and probabilities faced by the transformer area.

[0123] (4) Strategy for enhancing bidirectional carrying capacity

[0124] By combining the identification results of different operating states of photovoltaic power stations, we can obtain the risks that the power station will face in the future. Considering the duration of the operating risks, we can formulate targeted optimization and improvement strategies.

[0125] 1) Strategy library construction

[0126] ①When the duration of heavy load or voltage over-limit in the transformer area is less than 5% of the predicted period, no measures will be considered for the time being;

[0127] ② When the duration of heavy load or voltage over-limit in a distribution area exceeds 50% of the predicted period, priority should be given to expanding the capacity of the distribution transformer in the distribution area to solve the bidirectional load problem;

[0128] ③ When the duration of overload or voltage exceeding the limit in the transformer area is less than 50% of the predicted period:

[0129] For the problem of positive heavy overload: the priority should be to carry out load regulation (shifting the load during peak hours to the off-peak hours), followed by load transfer (source-load switching between transformers), and finally consider expanding the capacity of distribution transformers;

[0130] To address the voltage exceeding the limit issue: "SVG reactive power compensation + photovoltaic 'four-way' regulation" is used to solve the problem of voltage exceeding the upper limit; "voltage regulator regulation + load transfer" is used to solve the problem of voltage exceeding the lower limit.

[0131] In response to the impact of reverse power flow: firstly, adjust the grid-connected control strategy of distributed photovoltaics and reduce photovoltaic output through the "four-fold" regulation of distributed photovoltaics; secondly, consider adjusting photovoltaic output through distribution and storage; and finally, consider expanding the capacity of distribution transformers.

[0132] 2) Strategy Execution Effectiveness Analysis

[0133] For the identified problems, the system automatically matches the optimal strategy from the strategy library. After the strategy is executed, data is collected again to evaluate the improvement effect. At the same time, new operational data is added to the dataset, and the GMM model is retrained periodically to achieve self-evolution of the model and adapt to changes in the transformer area structure.

[0134] The innovation of this invention lies in:

[0135] 1. From "threshold judgment" to "pattern learning": Utilizing the unsupervised learning capability of Gaussian models, the system automatically identifies the operating modes of transformer substations under various working conditions, accurately identifies abnormal states that exceed normal patterns and their probability of occurrence, and achieves early warning of problems.

[0136] 2. From "single indicator" to "comprehensive profile": Construct multi-dimensional feature vectors to perform probabilistic clustering and comprehensive evaluation of the operating status of the transformer substation, rather than judging the limits of a single indicator, thus identifying more accurate and comprehensive features.

[0137] 3. From “isolated control” to “collaborative optimization”: Based on the identified problem patterns, through automatic matching of the problem-strategy library, optimization and improvement strategies that consider multi-resource collaboration are formed, flexibly improving bidirectional carrying capacity and avoiding the high cost of traditional “hard expansion”.

[0138] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for improving the bidirectional carrying capacity of a transformer area based on a Gaussian mixture model, characterized in that, Specifically comprising the following steps: S1, collecting the operation data of the transformer area distribution network, including the voltage, current and active power of the distribution transformer, the output power of the distributed power supply, and the time series data of the user power load; and preprocessing various types of data; S2, Gaussian mixture model construction and training: based on the collected historical operation data of the transformer area distribution network, a feature vector representing the operation state of the transformer area is constructed for each time section, and the operation state of the transformer area is matched for each time point; a mixed Gaussian model is trained using the obtained data set, and the mixed Gaussian model includes a plurality of single Gaussian models corresponding to a plurality of transformer area operation states; S3, based on the historical operation data of the transformer area distribution network and the source load development trend, the load demand and the distributed power output of the transformer area are predicted, the feature vector of the transformer area operation state at future T time is obtained, and the trained mixed Gaussian model is input to determine the transformer area operation state at future T time, the risk and risk duration faced by the transformer area operation in a future period of time are obtained; S4, matching the optimal strategy from the strategy library and executing.

2. The Gauss mixture model-based two-way load capacity improvement method for a transformer area according to claim 1, characterized in that, The preprocessing of various types of data specifically includes: Cleaning the data: using the 3σ criterion to remove abnormal data, and using linear interpolation to complete the missing values; The time series data is timestamped and standardized. 3.The transformer area two-way load capacity improvement method based on Gaussian mixture model of claim 1, wherein, constructing a feature vector for each time slice as follows: In the formula, P loadi (t) is the total power consumption of the i-th medium voltage line at time t, is the output of the distributed power source at time t.

4. The method of claim 3, wherein the method is characterized by, When the distribution type of the power supply is the distributed photovoltaic power supply, the characteristic vector of each time section is obtained by the following formula: is represented as: In the formula, P PVj (t) is the output of the jth photovoltaic unit at time t, and the dimension of the feature vector X(t) is i+j.

5. The Gauss mixture model-based two-way load capacity improvement method for a transformer area according to claim 4, characterized in that, The transformer area is also connected to energy storage and distributed wind power, then the feature vector of each time section is is represented as: In the formula, P Windk (t) is the output of the kth wind turbine at time t, P Storagel (t) is the output of the lth energy storage at time t, and the dimension of the feature vector X(t) is i+j+k+l.

6. The Gauss mixture model-based two-way load capacity improvement method for a transformer area according to claim 1, characterized in that, The transformer area operation state includes seven operation states: distribution transformer positive heavy overload but voltage not over limit, distribution transformer positive heavy overload and voltage over limit, distribution transformer voltage over limit but not positive heavy overload, distribution transformer reverse heavy overload but voltage not over limit, distribution transformer reverse heavy overload and voltage over limit, distribution transformer voltage over limit but not reverse heavy overload, and normal operation.

7. The Gauss mixture model-based two-way load capacity improvement method for a transformer area according to claim 1, characterized in that, The training of the mixed Gaussian model is as follows: S2.1, initializing mean value μ i' , standard deviation σ i' , component weight α i' for each single Gaussian model in the Gaussian mixture model i' , randomly initializing between (0, 1), σ i' , initializing as a unit positive definite matrix, α i' , initializing to 1 / I, i' is a single Gaussian model index, i' = 1, 2,..., I, I is the number of single Gaussian models; S2.2, calculation using the expectation maximization algorithm, including E step and M step: Step E: For sample point X n , calculate the probability that the sample belongs to the i'th state S I in the set of states S1, S2,..., S i' : P(Si | X) : where j' is used to traverse each single Gaussian model, is the distribution probability density function: M step: using the probability estimated by E step, update the parameters of Gaussian mixture model μ i' , σ i' , α i' , and then use the updated parameters in E step; the parameter update formula is as follows: In the formula, N is the total number of training set samples; S2.3, repeat the calculation of E step and M step until converging to a local optimal solution, to obtain the parameters of the Gaussian mixture model, wherein the mean value μ of the single Gaussian model i' is regarded as the distribution center, i.e. the i'th operating state of the district.

8. The method of claim 3-5, wherein the method is characterized in that, In step S3, the feature vector X(t+T) of the transformer area operation state at future T time is input into the trained Gaussian mixture model, and the feature vector X(t+T) is represented as: The posterior probability of the feature vector X(t+T) belonging to each transformer area operation state is calculated and sorted by size, and the transformer area operation state at future T time is determined according to the maximum posterior probability.

9. The method of claim 1, wherein the method is characterized by, The construction of the strategy library is as follows: (1) When the duration of transformer overload or voltage over limit is less than 5% of the predicted period, no measures are taken; (2) When the duration of transformer overload or voltage over limit exceeds 50% of the predicted period, transformer distribution transformer expansion is taken to solve the problem of bidirectional load bearing; (2) When the duration of transformer overload or voltage over limit exceeds 5% of the predicted period and is less than 50% of the predicted period: For positive heavy overload, the following priority measures are taken: priority is given to load regulation, i.e. shifting the load in the peak period to the valley period; secondly, load shifting is carried out, and finally, transformer expansion is considered; For voltage over limit problem: SVG reactive power compensation combined with distributed photovoltaic four adjustable adjustment to solve the problem of voltage over limit; voltage regulator adjustment combined with load shifting to solve the problem of voltage over limit; For the impact of reverse power flow, the following measures are taken in the following priority: priority adjustment of distributed photovoltaic grid-connected control strategy, adjustment of photovoltaic output through distributed photovoltaic four; secondly, adjust the photovoltaic output through the distribution storage; finally, take the distribution of variable expansion.

10. The method of claim 1, wherein the method is characterized by: After the strategy is executed, the operation data of the distribution network in the station area are collected again, the improvement effect of the bidirectional bearing capacity of the station area is evaluated, and the new operation data of the distribution network in the station area are added to the data set. The Gaussian mixture model is retrained regularly to realize the self-evolution of the model and adapt to the changes of the station area structure.

Images