Transformer heavy overload prediction method and device, equipment and storage medium

Abstract

This invention discloses a method, apparatus, device, and storage medium for predicting transformer overload. By calculating the multi-dimensional dynamically coupled adjacency matrix of the transformer, this adjacency matrix can be used in conjunction with a spatiotemporal attention bidirectional feedback network to effectively improve the accuracy of overload prediction and reduce the false judgment rate under extreme conditions. Furthermore, both the multi-dimensional dynamically coupled adjacency matrix and the spatiotemporal attention bidirectional feedback network are lightweight parallel computing architectures, resulting in a short prediction time for the target transformer's load rate and low resource consumption, which can meet the real-time dispatching requirements of the power grid. The invention also obtains overload and overload benchmark thresholds based on the historical number of overloads in the target transformer's static equipment dataset. These thresholds are used to predict whether the target transformer is overloaded or overloaded, allowing for threshold settings based on the specific conditions of the target transformer to reflect individual differences and further improve prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of transformer prediction technology, and in particular to a transformer overload prediction scheme, device, equipment and storage medium. Background Technology

[0002] As the core equipment connecting users and the power grid in the distribution network, the operating status of the distribution transformer directly affects the reliability of power supply and the lifespan of the equipment.

[0003] However, existing overload prediction technologies for distribution transformers have several shortcomings, including: First, the adjacency matrix has a single dimension and the dynamic association representation is not comprehensive. Specifically, it only constructs a binary dynamic adjacency matrix based on geographical distance and traffic correlation for traffic prediction; it cannot reflect the dynamic characteristics of the interaction of multiple factors; and it ignores the dynamic changes of time-series interaction, resulting in a large deviation between the node association representation and the actual operation of the distribution network.

[0004] Second, the spatiotemporal feature processing is serial and lacks interactive collaboration. Specifically, the existing technology adopts a unidirectional serial structure, and the spatiotemporal features are only transmitted in one direction. It cannot capture the closed-loop interaction law of spatiotemporal fluctuations triggering spatial correlation and spatial correlation amplifying time features. Moreover, it lacks dynamic coupling, which leads to a sharp drop in prediction accuracy in scenarios such as abrupt changes in coincidence and feeder radiation collaborative rise.

[0005] Third, the overload judgment threshold of the equipment is fixed and ignores individual differences. Specifically, the existing technology adopts a uniform fixed threshold without considering the individual characteristics of the equipment and the dynamic changes in the working conditions. This leads to a high misjudgment rate of old equipment and is prone to over-maintenance or missed faults.

[0006] Therefore, there is an urgent need for a solution that can address the aforementioned technical problems. Summary of the Invention

[0007] The main objective of this invention is to provide a method, apparatus, device, and storage medium for predicting transformer overload.

[0008] To achieve the above objectives, the first aspect of the present invention provides a method for predicting transformer overload, the method comprising: Collect multi-source data from multiple transformers, including: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range; The geographic correlation parameter is obtained by using the geographic topology data of the multiple transformers, the load coordination parameter is obtained by using the load data of the multiple transformers, the meteorological response coefficient is obtained by using the meteorological data of the multiple transformers, and the equipment health index is obtained by using the equipment static data and equipment status data of the multiple transformers. Based on the aforementioned geographical correlation parameters, load synergy parameters, meteorological response coefficients, and equipment health index, a weighted fusion and normalization operation is performed to obtain the adjacency matrix of each transformer with multi-dimensional dynamic coupling. The load data and meteorological data are preprocessed and normalized to obtain a sequence of continuous characteristics of each transformer based on load and meteorological conditions, which is then used as sample data. Based on the sample data, device static data, and the adjacency matrix, the initial spatiotemporal attention bidirectional feedback network is trained to obtain the trained spatiotemporal attention bidirectional feedback network. The target sequence obtained from real-time load data and meteorological data based on the target transformer, and the target adjacency matrix obtained from multi-source data based on the target transformer are input into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for a future preset time period. The static data of the target transformer and the number of historical heavy overloads are input into the trained GBDT model to obtain the heavy load benchmark threshold and overload benchmark threshold corresponding to the target transformer. The target transformer is predicted to be overloaded or overloaded based on the heavy load reference threshold, the overload reference threshold, and the predicted load rate.

[0009] Optionally, the geographic topology data includes: the topology of the power grid where the transformer is located and the line path distance between transformers, then the geographic correlation parameter is calculated according to the following formula:

[0010] in, The parameter representing the geographical correlation between transformer i and transformer j at time t. This represents the distance of the line path between transformer i and transformer j. α represents the maximum line path distance between two transformers in an area containing multiple transformers. 1、 α2 and σ G Both represent preset constants; The load data includes the transformer load, and the load coordination parameter is calculated using the following formula:

[0011] in, The load coordination parameters of transformer i and transformer j at time t are represented, where β1 and β2 are preset constants. This represents the load Pearson correlation coefficient between transformer i and transformer j. This represents the difference between the load on transformer i at time t and the load at time t-1. This represents the difference between the load on transformer j at time t and the load at time t-1. The meteorological data includes: ambient temperature, relative humidity, and precipitation. The meteorological response coefficient is calculated using the following formula:

[0012] This represents the meteorological response coefficient at time t. This represents the ambient temperature at time t. Indicates the preset reference temperature. Indicates the relative humidity of the environment. Indicates precipitation. , and This represents a preset constant; The static data of the equipment includes: years of operation; the status data of the equipment includes: number of failures in the past year; and the health index of the equipment is calculated using the following formula:

[0013]

[0014]

[0015] Indicates the equipment health index. This indicates the number of years that transformer i has been in operation. This indicates the number of failures of transformer i in the past year. This indicates the number of years that transformer j has been in operation. This indicates the number of failures of transformer j in the past year.

[0016] Optionally, the multi-dimensional dynamic coupling adjacency matrix can be calculated using the following formula:

[0017] in, This represents the multi-dimensional dynamic coupling adjacency matrix between transformer i and transformer j. , , , These represent the geographical correlation parameter, load coordination parameter, meteorological response coefficient, and equipment health index of transformer i and transformer j at time t, respectively. Let k represent N transformers, where k is any one of the N transformers except for i. , , , These represent the geographical correlation parameter, load coordination parameter, meteorological response coefficient, and equipment health index of transformer i and transformer k at time t, respectively.

[0018] Optionally, if the equipment static data includes rated capacity, then predicting whether the target transformer is under heavy load or overload based on the heavy load reference threshold, the overload reference threshold, and the predicted load rate includes: Based on the rated capacity, the heavy load reference threshold and the overload reference threshold, safety boundary constraints are applied to determine the heavy load boundary value and the overload boundary value. Predict whether the target transformer is under heavy load or overload based on the heavy load boundary value, the overload boundary value, and the predicted load rate.

[0019] Optionally, the step of determining the overload boundary value and the overload boundary value based on the rated capacity, the heavy-load reference threshold, and the overload reference threshold for safety boundary constraints includes: The maximum value among the product of the heavy-load reference threshold and the rated capacity and the preset first ratio value is selected as the heavy-load boundary value. The maximum value among the products of the overload reference threshold and the rated capacity and the preset second ratio value is selected as the overload boundary value.

[0020] Optionally, predicting whether the target transformer is under heavy load or overload based on the heavy load boundary value, the overload boundary value, and the predicted load rate includes: If the predicted load rate is greater than or equal to the heavy load boundary value, less than the overload boundary value, and the duration is greater than a preset first duration, then the target transformer is predicted to be under heavy load. If the predicted load rate is greater than or equal to the overload boundary value, and the duration is greater than a preset second duration, then the target transformer is predicted to be overloaded.

[0021] Optionally, the method further includes: The KL divergence index of the target transformer is calculated using the load data and meteorological data of the target transformer, and the PSI index of the target transformer is calculated using the meteorological data and load data of the target transformer. The KL divergence index and the PSI index are used to detect drift, determine the degree of drift, and update the spatiotemporal attention bidirectional feedback network based on the degree of drift.

[0022] To achieve the above objectives, a second aspect of the present invention provides a transformer overload prediction device, the device comprising: The data acquisition module is used to collect multi-source data from multiple transformers. The multi-source data includes: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range. The parameter calculation module is used to calculate the geographic correlation parameter using the geographic topology data of the multiple transformers, to calculate the load coordination parameter using the load data of the multiple transformers, to calculate the meteorological response coefficient using the meteorological data of the multiple transformers, and to calculate the equipment health index using the equipment static data and equipment status data of the multiple transformers. The matrix calculation module is used to perform weighted fusion and normalization operations based on the geographical correlation parameters, load coordination parameters, meteorological response coefficients and equipment health index to obtain the adjacency matrix of each transformer in a multi-dimensional dynamic coupling. The sample module is used to preprocess and normalize the load data and meteorological data to obtain a sequence of continuous characteristics of each transformer based on load and meteorology, which is then used as sample data. The training module is used to train the initial spatiotemporal attention bidirectional feedback network based on the sample data, device static data and the adjacency matrix, so as to obtain the trained spatiotemporal attention bidirectional feedback network. The prediction module is used to input the target sequence obtained from the real-time load data and meteorological data based on the target transformer and the target adjacency matrix obtained from the multi-source data of the target transformer into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for a future preset time period. The threshold module is used to input the static data of the target transformer and the number of historical heavy overloads into the trained GBDT model to obtain the heavy load benchmark threshold and overload benchmark threshold corresponding to the target transformer. The heavy overload prediction module is used to predict whether the target transformer is under heavy load or overload based on the heavy load reference threshold, the overload reference threshold and the predicted load rate.

[0023] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method mentioned in the first aspect.

[0024] To achieve the above objectives, a fourth aspect of the present invention provides a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the method mentioned in the first aspect.

[0025] The embodiments of the present invention have the following beneficial effects: This invention provides a method for predicting transformer overload. The method includes: collecting multi-source data from multiple transformers, including: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range; calculating geographic correlation parameters using the geographic topology data of the multiple transformers, calculating load synergy parameters using the load data of the multiple transformers, calculating meteorological response coefficients using the meteorological data of the multiple transformers, and calculating equipment health indexes using the equipment static data and equipment status data of the multiple transformers; performing weighted fusion and normalization operations based on the geographic correlation parameters, load synergy parameters, meteorological response coefficients, and equipment health indexes to obtain a multi-dimensional dynamically coupled adjacency matrix for each transformer; and processing the load data and meteorological data... Preprocessing and normalization are performed to obtain a sequence of continuous characteristics of each transformer based on load and weather, which is used as sample data. Based on the sample data, equipment static data, and adjacency matrix, the initial spatiotemporal attention bidirectional feedback network is trained to obtain the trained spatiotemporal attention bidirectional feedback network. The target sequence obtained from real-time load and weather data of the target transformer, and the target adjacency matrix obtained from multi-source data of the target transformer, are input into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for a future preset time period. The equipment static data and historical heavy overload counts of the target transformer are input into the trained GBDT model to obtain the heavy load benchmark threshold and overload benchmark threshold corresponding to the target transformer. Based on the heavy load benchmark threshold, overload benchmark threshold, and predicted load rate, it is predicted whether the target transformer is under heavy load or overload.

[0026] In the aforementioned transformer overload prediction method, geographical correlation parameters, load coordination parameters, meteorological response coefficients, and equipment monitoring indexes are calculated, and then weighted, fused, and normalized to obtain a multi-dimensional dynamically coupled adjacency matrix of the transformer. This adjacency matrix can be used in conjunction with a spatiotemporal attention bidirectional feedback network to effectively improve the accuracy of overload prediction, with a low misjudgment rate under extreme conditions, making it suitable for load change scenarios. Furthermore, both the multi-dimensional dynamically coupled adjacency matrix and the spatiotemporal attention bidirectional feedback network are lightweight parallel computing architectures, resulting in short prediction time for the target transformer's load rate and low resource consumption, which can meet the real-time dispatching requirements of the power grid. In addition, overload and overload benchmark thresholds are obtained based on the historical number of overloads in the target transformer's equipment static dataset. This allows for the prediction of whether the target transformer is overloaded or overloaded based on the overload and overload benchmark thresholds, enabling the setting of thresholds based on the target transformer's condition, reflecting individual differences, and further improving the accuracy of prediction. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] in: Figure 1 This is a flowchart of the transformer overload prediction method in an embodiment of the present invention; Figure 2 This is a structural diagram of the transformer overload prediction device in an embodiment of the present invention; Figure 3 This is a structural block diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figure 1 The diagram below illustrates the process of predicting transformer overload in an embodiment of the present invention, including: Step 101: Collect multi-source data from multiple transformers. The multi-source data includes: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range. In this embodiment of the invention, in order to achieve heavy overload prediction of transformers, it is necessary to collect multi-source data from multiple transformers. This multi-source data includes geographical topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range.

[0031] The data includes geographical topology data, which shows the topology of the power grid where multiple transformers are located and the distances between the line paths between the transformers. Equipment static data includes rated capacity, years of operation, insulation class, etc. Equipment status data includes the number of faults in the past year and the number of historical heavy overloads. Time-series data includes transformer load, specifically historical and real-time load data. Meteorological data includes ambient temperature, relative humidity, precipitation, etc. It is understood that time-series and meteorological data can be collected according to a preset sampling frequency to obtain multiple data sets arranged in chronological order.

[0032] Step 102: Calculate the geographic correlation parameter using the geographic topology data of the multiple transformers, calculate the load coordination parameter using the load data of the multiple transformers, calculate the meteorological response coefficient using the meteorological data of the multiple transformers, and calculate the equipment health index using the equipment static data and equipment status data of the multiple transformers. In this embodiment of the invention, after obtaining the aforementioned multi-source data, the geographic topology data of the multiple transformers will be used to calculate the geographic correlation parameter. Specifically, the geographic correlation parameter can be calculated according to the following formula:

[0033] in, The parameter representing the geographical correlation between transformer i and transformer j at time t. This represents the distance of the line path between transformer i and transformer j. α represents the maximum line path distance between two transformers in an area containing multiple transformers. 1、 α2 and σ G All of these represent preset constants.

[0034] In addition, load data from multiple transformers can be used to calculate the load coordination parameter, which is calculated using the following formula:

[0035] in, The load coordination parameters of transformer i and transformer j at time t are represented, where β1 and β2 are preset constants. This represents the load Pearson correlation coefficient between transformer i and transformer j. This represents the difference between the load on transformer i at time t and the load at time t-1. This represents the difference between the load on transformer j at time t and the load at time t-1.

[0036] In addition, meteorological data from multiple transformers can be used to calculate the meteorological response coefficient. The data required for calculating the meteorological response coefficient includes temperature, relative humidity, and precipitation. The specific formula for calculating the meteorological response coefficient is as follows:

[0037] This represents the meteorological response coefficient at time t. This represents the ambient temperature at time t. Indicates the preset reference temperature. Indicates the relative humidity of the environment. Indicates precipitation. , and This represents a preset constant.

[0038] In addition, the equipment health index can be calculated using static and status data from multiple transformers. The required static data includes the number of years the transformer has been in operation, and the status data includes the number of failures in the past year. Specifically, the following formula can be used for calculation:

[0039]

[0040]

[0041] Indicates the equipment health index. This indicates the number of years that transformer i has been in operation. This indicates the number of failures of transformer i in the past year. This indicates the number of years that transformer j has been in operation. This indicates the number of failures of transformer j in the past year.

[0042] Step 103: Perform weighted fusion and normalization operations based on the geographical correlation parameters, load coordination parameters, meteorological response coefficients and equipment health indexes to obtain the adjacency matrix of each transformer with multi-dimensional dynamic coupling. In this embodiment of the invention, after obtaining the geographical correlation parameter, load coordination parameter, meteorological response coefficient and equipment health index, these parameters are used to perform weighted fusion and normalization operations to obtain the adjacency matrix of multi-dimensional dynamic coupling of each transformer.

[0043] Specifically, the formula for the adjacency matrix is ​​as follows:

[0044] in, This indicates the number of transformers in the power grid other than transformer i.

[0045] It is understandable that the aforementioned multi-dimensional approach refers to the four dimensions of "geographical association, load coordination, meteorological response, and equipment health". This adjacency matrix is ​​a square matrix, where the number of rows is equal to the number of columns and the number of transformers. By converting the operating status of adjacent transformers into numerical values, the correlation strength between transformers can be better identified.

[0046] Step 104: Preprocess and normalize the load data and meteorological data to obtain a sequence of continuous characteristics of each transformer based on load and meteorology, and use it as sample data.

[0047] In this embodiment of the invention, after obtaining the adjacency matrix of the multi-dimensional dynamic coupling of each transformer, the load data and meteorological data will be further preprocessed and normalized to obtain a sequence of continuous characteristics of each transformer based on load and meteorology, which will be used as sample data.

[0048] The preprocessing process includes: using the three-standard-deviation principle (3σ principle) to remove outliers from the load and meteorological data, and then filling in the missing values ​​using the linear interpolation method to complete the preprocessing process. After the preprocessing process is completed, the preprocessed meteorological and load data will be normalized, and the normalized data will be used as sample data. Specifically, the Min-Max normalization (also known as deviation standardization) method can be used. It can be understood that one transformer corresponds to one sample data.

[0049] Among them, the sequence of continuous features can also be called time-series features. It is a continuous feature sequence of data that changes dynamically over time, collected at a preset fixed frequency, and then preprocessed and normalized.

[0050] Step 105: Based on the sample data, device static data and the adjacency matrix, train the initial spatiotemporal attention bidirectional feedback network to obtain the trained spatiotemporal attention bidirectional feedback network. In this embodiment of the invention, the initial spatiotemporal attention bidirectional feedback network includes a temporal attention layer, a spatial attention layer, a fusion layer, and a fully connected layer, and the specific process includes: 1. Forward Propagation: Sample data is input into the temporal attention layer, mapped to a query matrix and a key matrix. Initial temporal attention weights are then calculated using these matrices. These weights measure the importance of the association between the target time *t* and historical time *τ*. A higher weight indicates a stronger contribution of the features at historical time *τ* to feature extraction at target time *t*. *t* represents the target time for feature calculation, and *τ* represents a historical time (e.g., the past 1 hour, 3 hours, etc.) involved in the association calculation.

[0051] Among them, the query matrix The transpose of is:

[0052] Among them, the key matrix The transpose of is:

[0053] Initial attention weights:

[0054] Where T represents transpose. Indicates the initial attention weights. Let d' represent the sample data, and W represent the dimensions of the query matrix and the key matrix.Q W K These represent the query weight value and the key weight value, respectively. Indicates the query bias term. Indicates key bias term, This represents a dynamically adjusted matrix.

[0055] The initial time attention weight is used to measure the importance of the correlation between the current time t and the historical time τ. The larger the weight, the stronger the contribution of the features of the corresponding historical time to the feature extraction of the current time.

[0056] The query matrix is ​​generated by mapping the sample data of the transformer. It focuses on the feature requirements at the current moment and is used to actively query the matching relationship between the features and historical moments. The matrix dimension is usually related to the feature dimension.

[0057] The key matrix is ​​also generated by mapping the sample data of the transformer, but it focuses on the features of historical moments and is used to match the query matrix at the current moment to calculate the degree of correlation between the two.

[0058] The dynamic adjustment matrix is ​​used to dynamically correct the correlation weights between different times to adapt to the time-series fluctuation characteristics of transformer load.

[0059] 2. Reverse Feedback: The sample data and adjacency matrix are also input into the spatial attention layer. The spatial attention layer calculates the initial spatial attention weights based on the sample data and adjacency matrix, generates the temporal weight correction coefficient, and optimizes the temporal weight correction coefficient to obtain the final temporal weights.

[0060] The formulas involved in the initial spatial attention weights are as follows:

[0061]

[0062]

[0063] in, Represents the transpose of the spatial query matrix. The transpose of the space bond matrix. Indicates the initial spatial attention weights. Represents the weights of the spatial query matrix. Represents the weights of the space bond matrix. Represents sample data, Represents the adjacency matrix. Represents the matrix dimensions of the spatial query matrix and the spatial key matrix, and ⊙ is the Hadamard product.

[0064] The formula for obtaining the final time weight based on the time weight correction coefficient is as follows:

[0065]

[0066]

[0067] in, This represents the time weighting correction coefficient, where λ is the dynamic balance factor. Indicates the final time weight. This indicates the number of transformers in the power grid other than transformer i. This represents the load of transformer j, a neighbor of transformer i, at time t. This indicates that transformer j is at time j. The load, This indicates the rated capacity of the transformer. This represents the difference between the maximum load at time t and the minimum load at time t.

[0068] 3. Spatiotemporal Fusion: The initial temporal attention weights output by the temporal attention layer and the final temporal weights output by the spatial attention layer are both input to the fusion layer. In addition, the device static data is also input to the fusion layer. The fusion layer fuses the initial attention weights and the final temporal weights to obtain the spatiotemporal fusion feature. The spatiotemporal fusion feature is then concatenated with the device static data, and the concatenated feature is input to the fully connected layer.

[0069] Specifically, the splicing formula is as follows: X all =Concat(X spatio_time， X(i),a=1)=Concat(N×(D1+D2)) Among them, X all X represents the spliced ​​features. spatio_time Let X(i) represent the spatiotemporal fusion features, X(i) represent the static data of transformer i, a=1 indicates that the spatiotemporal fusion is limited to the column dimension and the rows remain unchanged, D1 and D2 are both two-dimensional matrices, representing the spatiotemporal fusion features and static data of the transformer corresponding to each row, respectively, and N represents the total number of transformers.

[0070] 4. Fully Connected Layer: The fully connected layer consists of two fully connected networks. After the concatenated features pass through these two fully connected networks, the predicted load rate for a future preset time period will be output. This preset time period can be 1-24 hours in the future.

[0071] In this embodiment of the invention, during the training process, the predicted load rate for the next 1-24 hours and the actual load rate of the transformer in the next 1-24 hours can be used to determine the loss function so as to adjust the parameters involved in the initial spatiotemporal attention bidirectional feedback network until the initial spatiotemporal attention bidirectional feedback network converges to obtain the trained spatiotemporal attention bidirectional feedback network.

[0072] It is understandable that the actual load factor for the next 1-24 hours can also be obtained through data acquisition after step 101.

[0073] Step 106: Input the target sequence obtained from the real-time load data and meteorological data based on the target transformer and the target adjacency matrix obtained from the multi-source data based on the target transformer into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for the future preset time period. In this embodiment of the invention, after training the spatiotemporal attention bidirectional feedback network, multi-source data of the transformer to be predicted, also known as the target transformer, will be collected in real time. The multi-dimensional dynamic coupling target adjacency matrix of the target transformer will be obtained in the manner described in steps 102 and 103. The meteorological data and load data of the target transformer will be preprocessed and normalized in the manner described in step 104 to obtain the target sequence of continuous features of the target transformer. The target sequence, the target adjacency matrix and the equipment status input of the target transformer will be input to the trained spatiotemporal attention bidirectional feedback network to obtain the predicted load rate of the target transformer.

[0074] Step 107: Input the static data of the target transformer and the historical number of heavy overloads into the trained GBDT model to obtain the heavy load reference threshold and overload reference threshold corresponding to the target transformer. In this embodiment of the invention, the GBDT model will be trained first, enabling the customization of benchmark thresholds based on transformer parameters to adapt to individual device differences. Specifically: From the multi-source data of the above multiple transformers, the rated capacity, years of operation, insulation class, and historical number of heavy overloads of each transformer are extracted as sample data for each transformer and input into the initial GBDT model for training. The GBDT model is used to output the heavy load reference threshold and overload reference threshold corresponding to the transformer. The GBDT model is an existing model, and its training process is also an existing process, which will not be described in detail here.

[0075] Furthermore, the rated capacity, years of operation, insulation class, and historical number of heavy overloads of the target transformer will be obtained and input into the trained GBDT model to obtain the heavy overload threshold and overload benchmark threshold corresponding to the target transformer.

[0076] Step 108: Predict whether the target transformer is under heavy load or overload based on the heavy load reference threshold, the overload reference threshold, and the predicted load rate.

[0077] After obtaining the heavy load reference threshold and the overload reference threshold, these two parameters, along with the predicted load rate of the target transformer, will be used to predict whether the target transformer is under heavy load or overload.

[0078] Specifically, step 108 may include the following steps: Step A: Based on the rated capacity, the heavy load reference threshold, and the overload reference threshold, perform safety boundary constraints to determine the heavy load boundary value and the overload boundary value; In this embodiment of the invention, the meteorological correction threshold is calculated first, and then the regional load rate correction threshold is calculated based on the meteorological correction threshold. The specific calculation method is as follows:

[0079]

[0080] in, Indicates the weather correction threshold. This indicates the regional load factor correction threshold. This represents the overload reference threshold, and T represents the ambient temperature. Represents humidity data. Indicates the regional load factor. , , This represents a preset constant.

[0081] The corrected heavy load reference threshold is obtained by multiplying the aforementioned regional load rate correction threshold with the heavy load reference threshold, and the corrected overload reference threshold is obtained by multiplying the aforementioned regional load rate correction threshold with the overload reference threshold.

[0082] The maximum value among the products of the modified heavy-load reference threshold, the rated capacity, and a preset first ratio value is selected as the heavy-load boundary value to achieve heavy-load safety boundary constraints. Similarly, the maximum value among the products of the modified overload reference threshold, the rated capacity, and a preset second ratio value is selected as the overload boundary value to achieve overload safety boundary constraints. For example, the first ratio value could be 0.65, and the second ratio value could be 0.9.

[0083] Step B: Predict whether the target transformer is under heavy load or overload based on the heavy load boundary value, the overload boundary value, and the predicted load rate.

[0084] In this embodiment of the invention, if the predicted load rate is greater than or equal to the heavy load boundary value, less than the overload boundary value, and the duration is greater than a preset first duration, then the target transformer is predicted to be overloaded; if the predicted load rate is greater than or equal to the overload boundary value, and the duration is greater than a preset second duration, then the target transformer is predicted to be overloaded. It can be understood that other than the two situations described, other situations are considered to be neither overloaded nor overloaded.

[0085] In this embodiment of the invention, in order to match the target transformer and achieve more accurate heavy overload judgment, the following method will be used to calculate a first proportional value and a second proportional value that match the target transformer by utilizing multi-source data of the target transformer: Step c1: Determine the thermal aging damage factor of the target transformer based on the equipment temperature data contained in the equipment status data;

[0086] in, Here, E represents the thermal aging damage factor, E represents the preset activation energy of the transformer's insulation material, and k represents the preset Boltzmann constant. This indicates the preset insulation reference temperature. This indicates the target temperature determined based on the equipment temperature data. , This represents a preset constant. Indicates the preset design life. This indicates the operating time of the target transformer.

[0087] Step c2: Calculate the aging damage factor of the target transformer based on the years of operation;

[0088] in, The aging damage factor represents the aging over time, the preset design life represents the aging damage factor, and Y represents the number of years of operation.

[0089] Step c3: Calculate the cumulative damage factor of the target transformer based on the historical number of heavy overloads, as follows:

[0090] in, Indicates the cumulative damage factor. This represents the number of historical heavy overloads, where z represents the z-th heavy overload. This indicates the preset allowed cumulative overload count.

[0091] Step c4: Calculate the humidity degradation factor of the target transformer based on the humidity data in the meteorological data, as follows:

[0092] in, Indicates the humidity degradation factor. Represents humidity data. This indicates the preset upper limit of humidity.

[0093] Step c5: Calculate the first coupling factor based on the thermal aging damage factor and the age-related aging damage factor; calculate the second coupling factor based on the cumulative damage factor and the thermal aging damage factor; and calculate the third coupling factor based on the humidity degradation factor and the thermal aging damage factor. The specific formulas used are as follows:

[0094]

[0095]

[0096] in, Indicates the first coupling factor. Indicates the second coupling factor. represents the third coupling factor, and q and u represent preset constants.

[0097] Step c6: Multiply the first coupling factor, the second coupling factor, and the third coupling factor to obtain the total coupling factor; and calculate the weights of the thermal aging damage factor, the annual aging damage factor, the cumulative damage factor, and the humidity degradation factor respectively; the specific weight formulas are as follows:

[0098] Where i represents thermal aging damage factor, age-related aging damage factor, cumulative damage factor, and humidity degradation factor. This represents the rate of change of damage, reflecting the speed of deterioration. The faster the rate, the higher the weight, and the sum of the weights of all factors is 1.

[0099] Step c7: Calculate the second ratio value based on the thermal aging damage factor, the annual aging damage factor, the cumulative damage factor, the humidity degradation factor and the corresponding weights, and the total coupling factor. Select the maximum value between the difference between the second ratio value and the preset minimum ratio value as the first ratio value.

[0100] The specific calculation formula is as follows:

[0101] S represents the second proportional value, C represents the total coupling factor, and F represents a preset constant.

[0102] In this embodiment of the invention, by calculating the thermal aging damage factor, the annual aging damage factor, the cumulative damage factor, and the humidity degradation factor respectively, it is possible to obtain factors that conform to various different physical laws. Furthermore, by performing first to third coupling factors and a total coupling factor, it is possible to construct a hierarchical, multi-dimensional, and strongly coupled total coupling factor. The second proportional value is calculated using this total coupling factor, making the proportional value more consistent with the actual situation of the target transformer and effectively reflecting the synergistic degradation of different parameters, thus making the heavy overload prediction of the target transformer more accurate.

[0103] Furthermore, in this embodiment of the invention, after completing the heavy overload prediction of the target transformer, drift detection can be performed on the spatiotemporal attention bidirectional feedback network to update the network and make it more applicable. Specifically, this includes the following steps: Step e1: Calculate the KL divergence index of the target transformer using the load data and meteorological data of the target transformer, and calculate the PSI index of the target transformer using the meteorological data and load data of the target transformer; Step e2: Use the KL divergence index and the PSI index to perform drift detection, determine the degree of drift, and update the spatiotemporal attention bidirectional feedback network based on the degree of drift.

[0104] In this embodiment of the invention, the load data of the target transformer, including first load data for performing a spatiotemporal attention bidirectional feedback network and second load data for obtaining the predicted load rate, can be used to determine the load distribution of the first load data and the load distribution of the second load data, and the KL divergence index can be calculated using the load distribution of the first load data and the load distribution of the second load data.

[0105] Furthermore, data segments of equal duration can be taken from the first ambient temperature used to train the aforementioned bidirectional feedback network and the second ambient temperature used to obtain the predicted load rate from meteorological data, and divided into multiple intervals, each interval being a bin. The bins of the first ambient temperature and the second ambient temperature are aligned, and the load data corresponding to each bin is determined based on the time corresponding to each bin. The meteorological-load correlation characteristics corresponding to that bin are calculated, and the PSI index is calculated according to the following formula:

[0106] in, This represents the PSI index value. This represents the frequency of the meteorological-load correlation characteristics corresponding to a certain bin of the second ambient temperature. The frequency of meteorological correlation features with the bin aligned with a certain bin in the first ambient temperature.

[0107] After obtaining the KL divergence and PSI indices mentioned above, these two indices will be used for drift detection, specifically including: If the KL divergence index is less than or equal to the first threshold and the PSI index is less than or equal to the second threshold, then a slight drift is identified.

[0108] If the KL divergence index is greater than the first threshold and less than or equal to the third threshold, or the PSI index is greater than the second threshold and less than or equal to the fourth threshold, then a moderate drift is determined to have occurred.

[0109] If the KL divergence index is greater than the third threshold, or the PSI index is greater than the fourth threshold, then severe drift is determined to have occurred.

[0110] If a slight drift is detected, the model parameters in the spatiotemporal attention bidirectional feedback network are updated using the following formula:

[0111] in, This indicates the updated parameters. This indicates the parameters before the update. This indicates the magnitude of the KL divergence index.

[0112] If moderate drift is determined, the gradient of the adjustment parameter used during training of the bidirectional attention feedback network is adjusted. The adjustment method can be to subtract the preset gradient value from the existing gradient to obtain the updated gradient value. Based on the updated gradient value, the above process of training the initial bidirectional attention feedback network is re-executed using sample data obtained from newly collected multi-source data to obtain the retrained bidirectional attention feedback network.

[0113] If severe drift is identified, local transfer retraining is performed on the bidirectional attention feedback network. That is, 70% of the existing model parameters in the network are retained, and then the bidirectional attention feedback network with only 70% of the parameters is trained using sample data obtained from newly collected multi-source data to obtain the updated bidirectional attention feedback network.

[0114] In this embodiment of the invention, the above-mentioned transformer overload prediction method calculates geographical correlation parameters, load coordination parameters, meteorological response coefficients, and equipment monitoring indexes, and performs weighted fusion and normalization operations to obtain a multi-dimensional dynamically coupled adjacency matrix of the transformer. This allows the adjacency matrix to work synergistically with a spatiotemporal attention bidirectional feedback network, effectively improving the accuracy of overload prediction and reducing the false judgment rate under extreme conditions. It is suitable for load change scenarios. Furthermore, both the multi-dimensional dynamically coupled adjacency matrix and the spatiotemporal attention bidirectional feedback network are lightweight parallel computing architectures, resulting in a short prediction time for the target transformer's load rate and low resource consumption, which can meet the real-time scheduling requirements of the power grid. In addition, the overload benchmark threshold and overload benchmark threshold are obtained based on the historical overload count of the target transformer's equipment static dataset. This allows for the prediction of whether the target transformer is overloaded or overloaded based on the overload benchmark threshold and overload benchmark threshold, enabling the setting of thresholds based on the target transformer's condition, reflecting individual differences, and further improving the accuracy of prediction.

[0115] Please see Figure 2 The diagram below shows the structure of a transformer overload prediction device according to an embodiment of the present invention. The device includes: The acquisition module 201 is used to acquire multi-source data from multiple transformers. The multi-source data includes: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range. The parameter calculation module 202 is used to calculate the geographic correlation parameter using the geographic topology data of the multiple transformers, calculate the load coordination parameter using the load data of the multiple transformers, calculate the meteorological response coefficient using the meteorological data of the multiple transformers, and calculate the equipment health index using the equipment static data and equipment status data of the multiple transformers. The matrix calculation module 203 is used to perform weighted fusion and normalization operations based on the geographical correlation parameters, load coordination parameters, meteorological response coefficients and equipment health index to obtain the multi-dimensional dynamic coupling adjacency matrix of each transformer. The sample module 204 is used to preprocess and normalize the load data and meteorological data to obtain a sequence of continuous characteristics of each transformer based on load and meteorology, and use it as sample data. Training module 205 is used to train the initial spatiotemporal attention bidirectional feedback network based on the sample data, device static data and the adjacency matrix to obtain the trained spatiotemporal attention bidirectional feedback network. The prediction module 206 is used to input the target sequence obtained from the real-time load data and meteorological data based on the target transformer and the target adjacency matrix obtained from the multi-source data based on the target transformer into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for a future preset time period. The threshold module 207 is used to input the static equipment data and historical heavy overload count of the target transformer into the trained GBDT model to obtain the heavy load benchmark threshold and overload benchmark threshold corresponding to the target transformer. The heavy overload prediction module 208 is used to predict whether the target transformer is under heavy load or overload based on the heavy load reference threshold, the overload reference threshold and the predicted load rate.

[0116] In the aforementioned transformer overload prediction device, geographical correlation parameters, load coordination parameters, meteorological response coefficients, and equipment monitoring index are calculated, and then weighted, fused, and normalized to obtain a multi-dimensional dynamically coupled adjacency matrix of the transformer. This adjacency matrix can be used in conjunction with a spatiotemporal attention bidirectional feedback network to effectively improve the accuracy of overload prediction, with a low misjudgment rate under extreme conditions, making it suitable for load change scenarios. Furthermore, both the multi-dimensional dynamically coupled adjacency matrix and the spatiotemporal attention bidirectional feedback network are lightweight parallel computing architectures, resulting in short prediction time for the target transformer's load rate and low resource consumption, which can meet the real-time dispatching requirements of the power grid. In addition, the overload benchmark threshold and overload benchmark threshold are obtained based on the historical overload count of the target transformer's equipment static dataset, so as to predict whether the target transformer is overloaded or overloaded based on the overload benchmark threshold and overload benchmark threshold. This allows for setting thresholds based on the target transformer's condition, reflecting individual differences, and further improving the accuracy of prediction.

[0117] Figure 3 An internal structural diagram of a computer device in one embodiment is shown. This computer device can specifically be a terminal or a server. Figure 3 As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program causes the processor to perform the steps in the above-described method embodiments. The internal memory may also store a computer program, which, when executed by the processor, causes the processor to perform the steps in the above-described method embodiments. Those skilled in the art will understand that... Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0118] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the contents of the method in the embodiment of the present invention.

[0119] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the contents of the method in the embodiment of the invention.

[0120] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0121] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0122] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting transformer overload, characterized in that, The method includes: Collect multi-source data from multiple transformers, including: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range; The geographic correlation parameter is obtained by using the geographic topology data of the multiple transformers, the load coordination parameter is obtained by using the load data of the multiple transformers, the meteorological response coefficient is obtained by using the meteorological data of the multiple transformers, and the equipment health index is obtained by using the equipment static data and equipment status data of the multiple transformers. Based on the aforementioned geographical correlation parameters, load synergy parameters, meteorological response coefficients, and equipment health index, a weighted fusion and normalization operation is performed to obtain the adjacency matrix of each transformer with multi-dimensional dynamic coupling. The load data and meteorological data are preprocessed and normalized to obtain a sequence of continuous characteristics of each transformer based on load and meteorological conditions, which is then used as sample data. Based on the sample data, device static data, and the adjacency matrix, the initial spatiotemporal attention bidirectional feedback network is trained to obtain the trained spatiotemporal attention bidirectional feedback network. The target sequence obtained from real-time load data and meteorological data based on the target transformer, and the target adjacency matrix obtained from multi-source data based on the target transformer are input into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for a future preset time period. The static data of the target transformer and the number of historical heavy overloads are input into the trained GBDT model to obtain the heavy load benchmark threshold and overload benchmark threshold corresponding to the target transformer. The target transformer is predicted to be overloaded or overloaded based on the heavy load reference threshold, the overload reference threshold, and the predicted load rate.

2. The method according to claim 1, characterized in that, The geographic topology data includes: the topology of the power grid where the transformer is located and the distance of the line paths between transformers. The geographic correlation parameter is calculated according to the following formula: in, The parameter representing the geographical correlation between transformer i and transformer j at time t. This represents the distance of the line path between transformer i and transformer j. α represents the maximum line path distance between two transformers in an area containing multiple transformers. 1、 α2 and σ G Both represent preset constants; The load data includes the transformer load, and the load coordination parameter is calculated using the following formula: in, The load coordination parameters of transformer i and transformer j at time t are represented, where β1 and β2 are preset constants. This represents the load Pearson correlation coefficient between transformer i and transformer j. This represents the difference between the load on transformer i at time t and the load at time t-1. This represents the difference between the load on transformer j at time t and the load at time t-1. The meteorological data includes: ambient temperature, relative humidity, and precipitation. The meteorological response coefficient is calculated using the following formula: This represents the meteorological response coefficient at time t. This represents the ambient temperature at time t. Indicates the preset reference temperature. Indicates the relative humidity of the environment. Indicates precipitation. , and This represents a preset constant; The static data of the equipment includes: years of operation; the status data of the equipment includes: number of failures in the past year; and the health index of the equipment is calculated using the following formula: Indicates the equipment health index. This indicates the number of years that transformer i has been in operation. This indicates the number of failures of transformer i in the past year. This indicates the number of years that transformer j has been in operation. This indicates the number of failures of transformer j in the past year.

3. The method according to claim 1, characterized in that, The multi-dimensional dynamic coupling adjacency matrix is ​​calculated using the following formula: in, This represents the multi-dimensional dynamic coupling adjacency matrix between transformer i and transformer j. , , , These represent the geographical correlation parameter, load coordination parameter, meteorological response coefficient, and equipment health index of transformer i and transformer j at time t, respectively. Let k represent N transformers, where k is any one of the N transformers except for i. , , , These represent the geographical correlation parameter, load coordination parameter, meteorological response coefficient, and equipment health index of transformer i and transformer k at time t, respectively.

4. The method according to claim 1, characterized in that, The static data of the equipment includes the rated capacity. Therefore, predicting whether the target transformer is under heavy load or overload based on the heavy load threshold, the overload threshold, and the predicted load rate includes: Based on the rated capacity, the heavy load reference threshold and the overload reference threshold, safety boundary constraints are applied to determine the heavy load boundary value and the overload boundary value. Predict whether the target transformer is under heavy load or overload based on the heavy load boundary value, the overload boundary value, and the predicted load rate.

5. The method according to claim 4, characterized in that, The step of determining the overload boundary value and the overload boundary value by performing safety boundary constraints based on the rated capacity, the heavy load reference threshold, and the overload reference threshold includes: The maximum value among the product of the heavy-load reference threshold and the rated capacity and the preset first ratio value is selected as the heavy-load boundary value. The maximum value among the products of the overload reference threshold and the rated capacity and the preset second ratio value is selected as the overload boundary value.

6. The method according to claim 4, characterized in that, The step of predicting whether the target transformer is under heavy load or overload based on the heavy load boundary value, the overload boundary value, and the predicted load rate includes: If the predicted load rate is greater than or equal to the heavy load boundary value, less than the overload boundary value, and the duration is greater than a preset first duration, then the target transformer is predicted to be under heavy load. If the predicted load rate is greater than or equal to the overload boundary value, and the duration is greater than a preset second duration, then the target transformer is predicted to be overloaded.

7. The method according to claim 5, characterized in that, The method further includes: The KL divergence index of the target transformer is calculated using the load data and meteorological data of the target transformer, and the PSI index of the target transformer is calculated using the meteorological data and load data of the target transformer. The KL divergence index and the PSI index are used to detect drift, determine the degree of drift, and update the spatiotemporal attention bidirectional feedback network based on the degree of drift.

8. A transformer overload prediction device, characterized in that, The device includes: The data acquisition module is used to collect multi-source data from multiple transformers. The multi-source data includes: geographic topology data, equipment static data, equipment status data, and load data and meteorological data within a preset time range. The parameter calculation module is used to calculate the geographic correlation parameter using the geographic topology data of the multiple transformers, to calculate the load coordination parameter using the load data of the multiple transformers, to calculate the meteorological response coefficient using the meteorological data of the multiple transformers, and to calculate the equipment health index using the equipment static data and equipment status data of the multiple transformers. The matrix calculation module is used to perform weighted fusion and normalization operations based on the geographical correlation parameters, load coordination parameters, meteorological response coefficients and equipment health index to obtain the adjacency matrix of each transformer in a multi-dimensional dynamic coupling. The sample module is used to preprocess and normalize the load data and meteorological data to obtain a sequence of continuous characteristics of each transformer based on load and meteorology, which is then used as sample data. The training module is used to train the initial spatiotemporal attention bidirectional feedback network based on the sample data, device static data and the adjacency matrix, so as to obtain the trained spatiotemporal attention bidirectional feedback network. The prediction module is used to input the target sequence obtained from the real-time load data and meteorological data based on the target transformer and the target adjacency matrix obtained from the multi-source data of the target transformer into the spatiotemporal attention bidirectional feedback network to obtain the predicted load rate for a future preset time period. The threshold module is used to input the static data of the target transformer and the number of historical heavy overloads into the trained GBDT model to obtain the heavy load benchmark threshold and overload benchmark threshold corresponding to the target transformer. The heavy overload prediction module is used to predict whether the target transformer is under heavy load or overload based on the heavy load reference threshold, the overload reference threshold and the predicted load rate.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images