Transformer load warning method, electronic device, storage medium, and program product

Abstract

This application provides a method, electronic device, storage medium, and program product for load early warning of a transformer. The method includes: acquiring multi-source operating data of the target transformer; performing probabilistic interval prediction of the transformer's future load based on the operating data to obtain a prediction sequence; extracting load rate and environmental parameters from the operating data; calculating the top oil temperature based on the load rate and environmental parameters; calculating the hot spot temperature based on the load rate and top oil temperature; dynamically determining the transformer's load-bearing capacity sequence for future periods based on the hot spot temperature, top oil temperature, and preset safety constraints, wherein the safety constraints include a top oil temperature limit, a hot spot temperature limit, an upper limit for auxiliary equipment capacity, and an insulation life loss threshold; comparing the prediction sequence and the load-bearing capacity sequence point by point in the same time sequence; and triggering a corresponding overload or heavy load early warning when the predicted load exceeds the corresponding early warning threshold of the load-bearing capacity at multiple consecutive time points.

Description

Technical Field

[0001] This application relates to the field of intelligent operation and maintenance of power systems, and in particular to a load early warning method for transformers, electronic equipment, storage medium and program products. Background Technology

[0002] With the accelerated construction of new power systems, transformers, as key equipment in the power grid, are crucial to the reliability of power supply due to their safe and stable operation. Against the backdrop of high-proportion renewable energy integration and increased load fluctuations, traditional overload early warning methods based on fixed thresholds or static thermal models are insufficient to accurately reflect the real-time load-bearing capacity of transformers under complex operating conditions, easily leading to false alarms or missed alarms, thus affecting power grid dispatching decisions and equipment lifespan management.

[0003] Existing technologies do not fully integrate the time-frequency characteristics of multi-source operational data and lack the ability to collaboratively model future load uncertainties and equipment dynamic safety boundaries. Furthermore, existing early warning mechanisms mostly consider only instantaneous thermal states and do not incorporate multi-dimensional safety constraints such as cumulative insulation life loss and auxiliary equipment capacity into a unified assessment framework. This results in conservative or aggressive assessments of load-bearing capacity, which cannot support refined and forward-looking operational control.

[0004] Therefore, there is an urgent need for a transformer load early warning method that can integrate multi-scale load forecasting, dynamic thermal state calculation and multi-constraint collaborative optimization to achieve high-precision and highly adaptable identification of heavy load and overload risks. Summary of the Invention

[0005] This application provides a load warning method, electronic device, storage medium, and program product for transformers, which can achieve dynamic load capacity assessment and probabilistic early warning based on multi-source data-driven, multi-scale feature fusion, and multi-dimensional safety constraint coordination.

[0006] In a first aspect, embodiments of this application provide a load warning method for a transformer, comprising: acquiring multi-source operating data of a target transformer; performing probabilistic interval prediction of the future load of the transformer based on the operating data to obtain a prediction sequence; extracting load rate and environmental parameters from the operating data; calculating the top oil temperature based on the load rate and environmental parameters; and calculating the hot spot temperature based on the load rate and the top oil temperature; dynamically determining the transformer's load-bearing capacity sequence for a future period based on the hot spot temperature, the top oil temperature, and preset safety constraints, wherein the safety constraints include a top oil temperature limit, a hot spot temperature limit, an upper limit for auxiliary equipment capacity, and an insulation life loss threshold; comparing the prediction sequence with the load-bearing capacity sequence point by point in the same time sequence; and triggering a corresponding overload warning or overload warning when the predicted load exceeds the corresponding warning threshold of the load-bearing capacity at multiple consecutive time points.

[0007] In one possible implementation, multiple input feature variables for load forecasting are extracted from the operational data. Multi-scale feature extraction is performed on the time series corresponding to each input feature variable to generate a multi-scale feature representation that integrates low-frequency trends and high-frequency fluctuations. The multi-scale feature representation of each input feature variable is mapped to a fixed-dimensional vector through an embedding layer. All embedding vectors are stacked along the variable dimension to form a joint sequence representation. A sparse self-attention mechanism is constructed based on the joint sequence representation to calculate the dynamic temporal dependencies between variables. Load forecast values ​​for multiple quantiles are generated based on the output of the sparse self-attention mechanism through a quantile regression model to form the forecast sequence.

[0008] In one possible implementation, the load rate and environmental parameters are input into a pre-built first fuzzy inference model to obtain a calculated value for the top oil temperature; the load rate and the calculated value for the top oil temperature are input into a pre-built second fuzzy inference model to obtain a calculated value for the hot spot temperature.

[0009] In one possible implementation, both the first fuzzy inference model and the second fuzzy inference model employ Gaussian membership functions, and the center parameter and width parameter of the Gaussian membership function are adaptively adjusted based on historical operating data.

[0010] In one possible implementation, the maximum allowable load value of the transformer at each point in time within a future period is determined, such that when the transformer operates at the maximum load value, the operating state of the transformer meets the safety constraints, wherein the safety constraints include at least one of the following: top oil temperature is less than or equal to a preset top oil temperature limit, hot spot temperature is less than or equal to a preset hot spot temperature limit, operating current is less than or equal to a preset upper limit of auxiliary equipment capacity, and cumulative insulation life loss is less than or equal to an insulation life loss threshold; the maximum load values ​​are arranged in chronological order to form the carrying capacity sequence.

[0011] In one possible implementation, the future period is divided into multiple time intervals; for each time interval, a relative aging rate is determined based on the hot spot temperature corresponding to the time interval; the cumulative insulation life loss is obtained by multiplying and summing the relative aging rates of each time interval with the time length of the corresponding time interval.

[0012] In one possible implementation, a heavy load warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of a first preset percentage range at multiple consecutive time points; an overload warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of a second preset percentage range at multiple consecutive time points; wherein the total duration of the multiple consecutive time points is not less than the lower limit of the preset time range, and the first preset percentage is less than the second preset percentage.

[0013] Secondly, embodiments of this application provide a load early warning device for a transformer, comprising: an acquisition module, configured to acquire multi-source operating data of a target transformer, and perform probabilistic interval prediction of the future load of the transformer based on the operating data to obtain a prediction sequence; an extraction module, configured to extract load rate and environmental parameters from the operating data, calculate the top oil temperature based on the load rate and environmental parameters, and calculate the hot spot temperature based on the load rate and the top oil temperature; a determination module, configured to dynamically determine the load-bearing capacity sequence of the transformer in the future time period according to the hot spot temperature, the top oil temperature, and preset safety constraints, wherein the safety constraints include a top oil temperature limit, a hot spot temperature limit, an upper limit for auxiliary equipment capacity, and an insulation life loss threshold; and an early warning module, configured to compare the prediction sequence with the load-bearing capacity sequence point by point in the same time sequence, and trigger a corresponding overload warning or overload warning when the predicted load exceeds the corresponding early warning threshold of the load-bearing capacity at multiple consecutive time points.

[0014] Thirdly, embodiments of this application provide a load warning device for a transformer, including: a memory and a processor;

[0015] The memory stores computer-executed instructions;

[0016] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0018] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0019] The transformer load early warning method, electronic device, storage medium, and program product provided in this application acquire multi-source operating data of the target transformer, generate a probabilistic interval prediction sequence characterizing load uncertainty based on the operating data, and extract load rate and environmental parameters from the operating data to calculate top oil temperature and hot spot temperature in sequence. Then, combined with multi-dimensional safety constraints such as top oil temperature limit, hot spot temperature limit, auxiliary equipment capacity limit, and insulation life loss threshold, the load-bearing capacity sequence for future periods is dynamically determined. Finally, by comparing the prediction sequence with the load-bearing capacity sequence point by point with time alignment, a graded early warning is triggered when the limit is exceeded at multiple consecutive time points. Thus, while taking into account the thermal state of the equipment, electrical capacity, and cumulative effects of insulation aging, it achieves a forward-looking, dynamic, and highly reliable early warning effect for heavy load and overload risks. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] Figure 1 Flowchart of the transformer load warning method provided in this application Figure 1 ;

[0022] Figure 2 Flowchart of the transformer load warning method provided in this application Figure 2 ;

[0023] Figure 3 A schematic diagram of multi-scale feature extraction provided for this application;

[0024] Figure 4 A schematic diagram illustrating the correlation between the features provided in this application and the predicted load;

[0025] Figure 5 A schematic diagram of the load warning device for the transformer provided in this application;

[0026] Figure 6 This is a structural schematic diagram of the transformer load warning device provided in this application.

[0027] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0028] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0029] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0030] Figure 1 Flowchart of the transformer load warning method provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0031] S201. Obtain multi-source operating data of the target transformer, and make probabilistic interval predictions of the future load of the transformer based on the operating data to obtain a prediction sequence;

[0032] Multi-source operating data refers to various real-time or historical monitoring data from the transformer itself and its related systems, including but not limited to data that can reflect the operating status of the equipment and external conditions, such as current, voltage, power, active / reactive load, and ambient temperature.

[0033] Probabilistic interval forecasting refers to forecasting future load values ​​not as a fixed point, but as an interval containing upper and lower bounds. This interval represents the possible range of load values ​​at a certain confidence level, and is used to quantify the uncertainty of the forecast.

[0034] A forecast sequence refers to a set of load forecast intervals for future periods arranged in chronological order, with each time point corresponding to a forecast interval.

[0035] This step first collects multi-source operational data related to the target transformer. Then, using this data as input, a predictive model is used to model the load over a future period, outputting the probabilistic load interval for each future time point, thus forming a complete prediction sequence. This sequence is used for subsequent risk comparison with the carrying capacity.

[0036] S202. Extract load rate and environmental parameters from the operating data, calculate the top oil temperature based on the load rate and environmental parameters, and calculate the hot spot temperature based on the load rate and top oil temperature.

[0037] Load factor refers to the ratio of the current actual load of a transformer to its rated capacity. It is calculated from the operating current or power and reflects the load level of the equipment.

[0038] Environmental parameters mainly refer to external environmental factors that affect the heat dissipation of transformers, with ambient temperature being a typical example.

[0039] Top layer oil temperature refers to the temperature of the upper layer of insulating oil in the transformer tank, and is a key indicator reflecting the overall thermal state.

[0040] Hot spot temperature refers to the temperature at the hottest point in the transformer winding. It directly determines the aging rate of the insulation material and is a core parameter for safety assessment.

[0041] This step extracts the load rate and environmental parameters from the acquired operational data. First, using the load rate and environmental parameters as input, the top oil temperature is calculated using a temperature calculation model. Then, using the load rate and the newly calculated top oil temperature as input, the hot spot temperature is calculated using another temperature calculation model. The two calculations are completed sequentially, forming a complete thermal state assessment chain.

[0042] S203. Based on the hot spot temperature, top oil temperature and preset safety constraints, dynamically determine the transformer's load-bearing capacity sequence for future periods.

[0043] Preset safety constraints: These refer to the pre-defined physical or operational boundaries used to ensure the safe operation of transformers, including: top oil temperature limit, hot spot temperature limit, auxiliary equipment capacity limit, and insulation life loss threshold.

[0044] The load-bearing capacity sequence refers to the sequence of the maximum load values ​​that a transformer can safely withstand at various points in the future, arranged in chronological order. This sequence changes dynamically with the current thermal state, environmental conditions, and safety constraints.

[0045] Dynamic determination means that the carrying capacity is not a fixed value, but is calculated online based on the real-time or predicted operating status and constraints.

[0046] In other words, based on the currently estimated top oil temperature and hot spot temperature, and combined with four preset safety constraints, the maximum allowable load value of the transformer under the premise of satisfying all constraints in the future time period is calculated; the maximum load values ​​at each time point are arranged in chronological order to form a sequence of carrying capacity. This process embodies the closed-loop logic of "state perception - constraint verification - capacity output".

[0047] S204. The predicted sequence and the carrying capacity sequence are compared point by point in the same time sequence. When the predicted load exceeds the corresponding early warning threshold of the carrying capacity at multiple consecutive time points, the corresponding heavy load warning or overload warning is triggered.

[0048] Same time series refers to two sequences whose time axes are aligned, that is, the predicted load and carrying capacity are compared at the same time.

[0049] The corresponding warning threshold refers to the threshold (e.g., heavy load threshold, overload threshold) associated with the carrying capacity and used to distinguish the warning level. Its value can be a fixed proportion or absolute value of the carrying capacity.

[0050] Multiple consecutive time points refer to a continuous window of time, rather than isolated moments, and are used to avoid false alarms caused by instantaneous fluctuations.

[0051] This step aligns the predicted sequence obtained in S201 with the carrying capacity sequence obtained in S203 in the time dimension, and compares whether the predicted load exceeds the warning threshold at each time point. If the load continues to exceed the limit for a continuous period of time (such as several hours), the corresponding level of warning signal is triggered according to the degree of exceeding the limit (such as whether it reaches the heavy load or overload threshold) for operation and maintenance personnel to make decisions.

[0052] The transformer load early warning method provided in this application acquires multi-source operating data of the target transformer, generates a probabilistic interval prediction sequence characterizing load uncertainty based on the operating data, and extracts load rate and environmental parameters from the operating data to calculate top oil temperature and hot spot temperature in sequence. Then, it dynamically determines the load-bearing capacity sequence for future periods by combining multi-dimensional safety constraints such as top oil temperature limit, hot spot temperature limit, auxiliary equipment capacity limit, and insulation life loss threshold. Finally, it triggers graded early warning when the predicted sequence and the load-bearing capacity sequence are time-aligned and compared point by point. This achieves a forward-looking, dynamic, and highly reliable early warning effect for heavy load and overload risks while taking into account the thermal state of the equipment, electrical capacity, and cumulative effects of insulation aging.

[0053] Figure 2 Flowchart of the transformer load warning method provided in this application Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the load warning method for transformers is described in detail. The method includes:

[0054] S301. Extract multiple input feature variables for load forecasting from the operational data. Perform multi-scale feature extraction on the time series corresponding to each input feature variable to generate a multi-scale feature representation that integrates low-frequency trends and high-frequency fluctuations. Map the multi-scale feature representation of each input feature variable to a fixed-dimensional vector through an embedding layer. Stack all embedding vectors along the variable dimension to form a joint sequence representation. Construct a sparse self-attention mechanism based on the joint sequence representation to calculate the dynamic temporal dependencies between variables. Generate load forecast values ​​for multiple quantiles based on the output of the sparse self-attention mechanism through a quantile regression model to form a forecast sequence.

[0055] Input feature variables refer to observable variables selected from multi-source operational data that have predictive value for the future load of the transformer. Optionally, these include historical active power, reactive power, current, voltage, ambient temperature, and timestamps (such as hourly or weekday types). Each variable constitutes an independent time series.

[0056] Multi-scale feature extraction refers to simultaneously capturing the dynamic characteristics of a single time series at different time granularities, including slowly changing low-frequency trends (such as daily load baseline) and rapidly changing high-frequency fluctuations (such as abrupt load changes and noise). This process aims to preserve the global structure and local details of the original signal.

[0057] Specifically, multi-scale feature extraction using one-dimensional wavelet transform convolution (1D-WTC) is employed to preprocess the input historical multivariate time series data, such as load, temperature, and humidity, separating and extracting features at different scales, and then performing denoising processing. Figure 3 As shown, Figure 3 This diagram illustrates the multi-scale feature extraction provided in this application. First, the input features are processed using 1D-WTC, where each feature undergoes discrete wavelet transform to decompose it into low-frequency components. and high frequency components Secondly, the low-frequency components are recursively decomposed into finer-grained components through deeper wavelet transforms until a predefined maximum decomposition level l is reached. At each layer, these two components are depthwise convolved to obtain two convolutional quantities. and Then, these two convolutional quantities are reconstructed using an inverse wavelet transform to obtain the result of the current level. Finally, the low-frequency components of the next stage are... This is combined to form cross-level fusion features, a process that propagates from the last level back to the first level by using the initial low-frequency convolution results. With the output of the second stage Add them together to get the final output. .

[0058] Joint sequence representation refers to a three-dimensional tensor formed by concatenating or stacking the vectors of all input feature variables after embedding them along the variable dimension (rather than the time dimension). Its structure is [time step, number of variables, embedding dimension], and it is used to characterize the collaborative evolution of multiple variables in a unified spatiotemporal framework.

[0059] In other words, from the acquired multi-source operational data, several variables strongly correlated with the load are selected as input features, such as historical active power, ambient temperature, and time characteristics. Each variable forms an independent time series. For the time series of each input feature variable, time-frequency analysis methods (such as wavelet transform, multi-resolution analysis, etc.) are used to decompose it into different frequency components, extracting low-frequency components (reflecting long-term trends) and high-frequency components (reflecting short-term fluctuations), and then generating a feature representation containing multi-scale information through reconstruction or fusion operations. This representation integrates trend and fluctuation information at each time point.

[0060] The multi-scale feature representation of each variable is input into an embedding layer containing learnable parameters, which uniformly maps variable-length or multi-dimensional features into fixed-dimensional vectors, ensuring that the features of different variables are in the same semantic space.

[0061] Specifically, the embedding layer will reconstruct the multi-scale feature sequence. (where i is the i-th variable) is mapped to a fixed dimension through linear projection. Embedding vector:

[0062]

[0063] The embedding vectors of all variables at each time step are stacked along the variable dimension to form a three-dimensional tensor, which serves as the joint time series input for multiple variables. This representation preserves the original relationship structure between variables, providing a foundation for subsequent cross-variable modeling. Based on the above joint sequence representation, query, key, and value matrices are calculated. By introducing sparsity constraints (such as filtering high-contribution queries based on KL divergence or using local window attention), only the variable-time point combinations with the greatest influence on the current prediction task are retained, thereby efficiently modeling the dynamic and non-uniform dependencies between multiple variables.

[0064] For example, firstly, the feature sequence after 1D-WTC processing is received. The sequence dimension is ,in Batch size; The time step of the sequence; The total number of variables;

[0065] Then, the complete time series of each individual variable is treated as an independent variable token, i.e., containing There are variables, and the original feature of each variable is a variable of length . Time series;

[0066] Then, the model uses an embedding layer to process this independently. 1 variable, for Any one of the variables Its input data has dimensions of sequence Convert it to a dimension The vector, where It is the embedding dimension of the model's hidden layer;

[0067] Finally, Each independently generated dimension is The variable tokens are stacked along the variable dimension to generate a completely new sequence. Its dimensions are ;

[0068] Based on the above joint sequence representation, the query, key, and value matrix is ​​calculated, including:

[0069] Sequence modeling based on a probabilistic sparse self-attention mechanism enhanced by Wav-KAN firstly uses a wavelet function as a learnable activation function to dynamically generate a query matrix with multi-resolution features based on the multi-scale features of the input data. Key matrix Value matrix The specific calculation process is as follows:

[0070]

[0071] In the formula: , , These represent the query, key, and value matrices, respectively. , , It is an activation function matrix composed of wavelet functions; , , It is a scale and location wavelet function at the location; Represents a variable token.

[0072] The output of the sparse self-attention mechanism is fed into the quantile regression module, which contains multiple parallel output heads, each corresponding to a preset quantile (e.g., α=0.1, 0.5, 0.9). The model directly outputs the load prediction value at each future time point at each quantile. These values ​​are organized in chronological order to form the final probabilistic interval prediction sequence.

[0073] When calculating attention weights, a probabilistic sparsity strategy is adopted. The sparsity of the query vector is quantified by calculating the KL divergence between each query vector and the key vector distribution. The formula for calculating sparsity is defined as follows:

[0074]

[0075] In the formula: for and KL divergence between them; for The length of the vector; Let be the dimension of the key matrix; The sampling factor; for The length of the vector.

[0076] Then, select the top one with the highest sparsity. Use query vectors to construct a sparse query matrix. After performing a complete attention calculation, the final attention output is as follows:

[0077]

[0078] S302. Input the load rate and environmental parameters into the pre-built first fuzzy inference model to obtain the calculated value of the top oil temperature; input the load rate and the calculated value of the top oil temperature into the pre-built second fuzzy inference model to obtain the calculated value of the hot spot temperature.

[0079] Load factor refers to the ratio of the current actual load of a transformer to its rated capacity. It is usually calculated from the operating current or active power and is used to characterize the load level of the equipment.

[0080] Fuzzy inference models, based on fuzzy logic and nonlinear mapping models, transform precise values ​​of input variables into precise estimates of output variables through three stages: fuzzification, rule-based inference, and defuzzification. Their core consists of a pre-defined fuzzy rule base (e.g., "If the load rate is high and the ambient temperature is high, then the top oil temperature is very high") and membership functions.

[0081] The first fuzzy inference model establishes a fuzzy inference system that maps load rate and environmental parameters to top oil temperature. Its input is two-dimensional (load rate, ambient temperature), and its output is top oil temperature (unit: °C).

[0082] The second fuzzy inference model establishes a fuzzy inference system that maps the load rate to the top oil temperature and then to the hot spot temperature. Its input is two-dimensional (load rate, top oil temperature), and its output is the hot spot temperature (unit: °C).

[0083] The calculated values ​​for top-layer oil temperature and hot-spot temperature refer to estimates derived from a fuzzy inference model, rather than measured values. Because the thermal process of a transformer has inertia and nonlinearity, direct measurement is difficult; therefore, model estimation is used.

[0084] In practical applications, the load rate and environmental parameters (e.g., ambient temperature) extracted from the operating data are first fed into the first fuzzy inference model as input.

[0085] The model first fuzzifies the two inputs, that is, it converts the precise input values ​​into the membership degrees of linguistic variables (such as "low", "medium", and "high") according to the preset membership functions (such as triangle and Gaussian).

[0086] Then, based on the built-in fuzzy rule library (e.g., "If the load rate is high and the ambient temperature is high, then the top oil temperature is very high"), fuzzy inference is performed to obtain the fuzzy set of output variables (top oil temperature);

[0087] Finally, the fuzzy output is converted into a single precise value, namely the calculated top oil temperature, by defuzzification (such as the centroid method).

[0088] Then, the load rate at the same moment and the calculated top oil temperature obtained in the previous step are used as inputs and fed into the second fuzzy inference model;

[0089] The model also performs a three-stage process of fuzzification, rule-based reasoning, and defuzzification.

[0090] Its fuzzy rule base reflects the physical relationship between winding hot spots and oil temperature and load (e.g., "if the load rate is high and the top oil temperature is medium, then the hot spot temperature is high").

[0091] The final output is the calculated hotspot temperature, which serves as the core basis for subsequent safety constraint assessments.

[0092] By using a two-level fuzzy inference structure, the multi-level nonlinear heat conduction process of the transformer from the external environment to the internal hot spot is effectively modeled. Compared with a single model that directly maps "load + environment → hot spot temperature", the two-level structure is more in line with the thermodynamic mechanism and improves the estimation accuracy.

[0093] For example, both the first fuzzy inference model and the second fuzzy inference model adopt Gaussian membership functions, and the center parameter and width parameter of the Gaussian membership function are adaptively adjusted based on historical running data.

[0094] Specifically, a hotspot temperature estimation model is constructed by cascading two Adaptive Neuro-Fuzzy Inference System (ANFIS) sub-models. Each ANFIS sub-model uses a standard five-layer feedforward network structure, including an input and membership function layer, a rule layer, a normalization layer, an inference layer, and a total output layer. The explanation of each layer is as follows:

[0095] Layer 1: Fuzzification layer, which performs a fuzzification process representing a membership function on each input variable, mapping the value of the input variable to the membership degree of the fuzzy set;

[0096] Layer 2: Rule layer, which performs logical operations on the membership degrees of input variables to obtain the incentive intensity of each rule and the fuzzy set of output variables;

[0097] Layer 3: Normalization layer, which normalizes the incentive intensity of each rule obtained from the previous layer to obtain its standardized weight. The number of standardized weights is the same as the number of fuzzy rules.

[0098] Layer 4: Inference layer, which executes the conclusion of the fuzzy rules and infers the output value from the rules;

[0099] Layer 5: Output layer, which calculates the weighted average of the output values ​​of each rule to obtain the exact output result.

[0100] The calculation process is as follows:

[0101]

[0102] In the formula: and They are fuzzy sets and Membership degree; and These are the language tags related to the node functions; and For nodes Input; select the Gaussian function as the membership function. and The maximum value is 1, and the minimum value is 0; and These represent the width and center of the Gaussian function, respectively.

[0103] Furthermore, an ANFIS model for estimating top oil temperature is constructed. Two measurable transformer load rates and ambient temperatures are input into the model, and the model outputs an accurate estimate of the top oil temperature.

[0104] A hot spot temperature estimation ANFIS model is constructed. The model inputs are the transformer load rate and the top oil temperature value obtained from the top oil temperature estimation ANFIS model. The model output is the estimated hot spot temperature value.

[0105] After obtaining the hot spot temperature accurately estimated by the ANFIS model, a dynamic load capacity optimization and evaluation model for the transformer is constructed to evaluate the changes in the transformer's load capacity.

[0106] The model is as follows:

[0107] The optimization objective, which is to maximize the transformer load capacity under safety constraints, is expressed as follows:

[0108]

[0109] S303. Determine the maximum allowable load value of the transformer at each time point in the future period, so that the transformer's operating state meets the safety constraints when operating at the maximum load value; arrange the maximum load values ​​into a load-bearing capacity sequence in chronological order.

[0110] Among them, the safety constraints include at least one of the following: the top oil temperature is less than or equal to the preset top oil temperature limit, the hot spot temperature is less than or equal to the preset hot spot temperature limit, the operating current is less than or equal to the preset upper limit of auxiliary equipment capacity, and the cumulative insulation life loss is less than or equal to the insulation life loss threshold.

[0111] In other words, maximizing transformer load factor requires satisfying the following four core constraints: top oil temperature constraint, hot spot temperature constraint, auxiliary equipment capacity constraint, and relative life loss constraint, as shown below:

[0112]

[0113] In the formula: This refers to the top oil temperature, expressed in °C. This refers to the hotspot temperature, expressed in °C. This refers to the top oil temperature threshold, in °C. This represents the hotspot temperature threshold, in °C. h is the threshold for lifespan loss; pu is the maximum allowable current threshold for auxiliary equipment; For the evaluation duration; The relative aging rate of oil-impregnated insulating paper; Time period to Lifespan loss, h; and The first The relative aging rate and interval time within each time interval.

[0114] For example, the future period is divided into multiple time intervals; for each time interval, the relative aging rate is determined based on the hot spot temperature corresponding to the time interval; the cumulative insulation life loss is obtained by multiplying and summing the relative aging rate of each time interval with the time length of the corresponding time interval.

[0115] The future period to be evaluated (e.g., the next 24 hours) is divided into multiple consecutive time intervals. Each time interval has the same or different duration (e.g., every 10 minutes or every hour is an interval), and the granularity of the division can be determined according to the temporal resolution of the predicted sequence.

[0116] For each time interval, the calculated value of the hot spot temperature within that interval is obtained (obtained from the aforementioned temperature estimation step); based on the hot spot temperature, the relative aging rate of that time interval is determined through a preset temperature-aging rate mapping relationship.

[0117] Here, "relative aging rate" refers to the ratio of the aging rate of the insulation material at the current hot spot temperature to the aging rate at a reference temperature (e.g., 98°C). For example, at a higher hot spot temperature, a relative aging rate greater than 1 indicates accelerated aging; at a lower temperature, a relative aging rate less than 1 indicates slower aging. The relative aging rate for each time interval is multiplied by its duration to obtain the aging contribution within that interval; subsequently, the aging contributions for all time intervals are summed, and the total sum represents the cumulative insulation life loss.

[0118] In an exemplary embodiment, a heavy load warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of a first preset percentage range at multiple consecutive time points; an overload warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of a second preset percentage range at multiple consecutive time points; wherein the total duration of the multiple consecutive time points is not less than the lower limit of the preset time range, and the first preset percentage is less than the second preset percentage.

[0119] In an exemplary embodiment of this application, in order to achieve graded early warning while taking into account both operational safety and scheduling flexibility, a first preset percentage range, a second preset percentage range, and a preset time range are set to define the triggering conditions for overload early warning and overload early warning.

[0120] in:

[0121] The first preset percentage range represents the load threshold range used to determine the risk of overload, and its typical value can be 75% to 85% (i.e. 75% to 85% of the transformer's load-bearing capacity).

[0122] The second preset percentage range represents the load threshold range used to determine overload risk, and its typical value can be 95% to 100%.

[0123] In practical applications, the predicted load curve for the next week is compared with the assessed dynamic load capacity curve for the next week; a situation where the load continuously reaches 80% of the dynamic load capacity for 2 hours or more is judged as heavy load; a situation where the load continuously reaches 100% of the dynamic load capacity for 2 hours or more is judged as overload.

[0124] The preset time range represents the minimum duration window that the warning must meet, and its typical value can be 1 hour to 4 hours.

[0125] The specific warning logic is as follows:

[0126] Heavy load warning triggering conditions: When the load value in the predicted sequence exceeds the upper limit of "capacity × first preset percentage range" (e.g., capacity × 85%) at multiple consecutive time points, and the total duration of the consecutive time points is not less than the lower limit of the preset time range (e.g., 1 hour), a heavy load warning is triggered, prompting operators to pay attention to the heat accumulation trend of the equipment and consider load transfer or enhanced cooling.

[0127] Overload warning trigger conditions: When the predicted load exceeds the upper limit of "capacity × second preset percentage range" (e.g., capacity × 100%) at multiple consecutive time points, and the duration is not less than the lower limit of the preset time range, an overload warning is triggered, indicating that the equipment is about to enter or has already entered a dangerous operating state, and control measures must be taken immediately.

[0128] In addition, to ensure the rationality of the warning level, the first preset percentage is set to be less than the second preset percentage (e.g., 85% < 100%), thereby ensuring that the heavy load warning occurs before the overload warning, forming an effective risk gradient response mechanism.

[0129] It is important to emphasize that the percentage ranges (e.g., 75%~85%, 95%~100%) and time ranges (e.g., 1~4 hours) mentioned above are merely exemplary values. In actual applications, they can be adjusted according to transformer type, insulation class, operation and maintenance strategy, or power grid regulations. For example, a more conservative threshold (e.g., a first preset percentage upper limit of 80%) can be used for older equipment, while a more lenient threshold can be applied to newer equipment with strong cooling capabilities. Therefore, the core of this application lies in dynamically setting multi-level early warning thresholds based on the load-bearing capacity and combining them with continuous over-limit time windows for judgment, rather than the specific numerical values ​​themselves.

[0130] The transformer load early warning method provided in this application acquires multi-source operating data of the target transformer, generates a probabilistic interval prediction sequence characterizing load uncertainty based on the operating data, and extracts load rate and environmental parameters from the operating data to calculate top oil temperature and hot spot temperature in sequence. Then, it dynamically determines the load-bearing capacity sequence for future periods by combining multi-dimensional safety constraints such as top oil temperature limit, hot spot temperature limit, auxiliary equipment capacity limit, and insulation life loss threshold. Finally, it triggers graded early warning when the predicted sequence and the load-bearing capacity sequence are time-aligned and compared point by point. This achieves a forward-looking, dynamic, and highly reliable early warning effect for heavy load and overload risks while taking into account the thermal state of the equipment, electrical capacity, and cumulative effects of insulation aging.

[0131] The dataset used in this paper was provided by a municipal power supply company in southern China, covering the operating load data of typical distribution substations in the region from 2022 to 2024. The data was sampled at 15-minute intervals, with 96 sampling points per day, for a total of 104,352 samples. During modeling, the original data was divided into training and test sets in a 4:1 ratio, with the last 10% of the training set used as a validation set for model structure adjustment and hyperparameter selection. The training set was used for parameter learning, and the test set was used for performance evaluation to ensure the model's generalization ability.

[0132] To improve the reliability of the experimental results, each prediction model was independently trained and tested 10 times, and four evaluation metrics—R², MAE, MAPE, and RMSE—were calculated. To reduce the impact of random errors, the maximum and minimum values ​​of each metric were removed during the calculation, and the average of the remaining results was taken as the final evaluation metric. This effectively improves the stability and comparability of the experimental results.

[0133] Based on the original feature data, this paper calculates the correlation between each feature and the system load, and selects key input variables that have a significant impact on the prediction results to improve the accuracy and efficiency of model training.

[0134] To further analyze the relationship between various features and predicted load, this paper selects the load data of a certain week as an example, sets the load sequence of the last day of that week as the prediction object, and calculates the correlation coefficients between the load data of the previous few days of that week and auxiliary features such as weather, date, and holidays and the predicted load. The correlation results between the feature factors and the predicted load are as follows: Figure 4 As shown, Figure 4 A schematic diagram illustrating the correlation between the features provided in this application and the predicted load.

[0135] Depend on Figure 4It can be seen that the correlation coefficients between the historical load of the six days prior to the forecast date and the load of the forecast date are all above 0.6, indicating a strong correlation. The correlation coefficients between the highest, lowest, and average temperatures, as well as the historical load of the same period during the forecast period, and the forecast load are all above 0.6, indicating a strong correlation. The correlations between rainfall, average humidity, and holiday factors and the forecast load are all below 0.4, indicating a weak correlation. Based on the above analysis, the historical load of the six days prior to the forecast date, the historical load of the same period during the forecast period, the highest temperature, the lowest temperature, and the average temperature are ultimately selected as the model input features.

[0136] To ensure the fairness of the comparative experiments, the parameters of the proposed WTC-iPST model and each benchmark model were determined with reference to relevant literature and through trial-and-error tuning. These included the Improved Transformer (iTransformer), Transformer, Temporal Convolutional Network (TCN), and Long Short-Term Memory Network (LSTM), ensuring superior performance on the same dataset. All models output prediction results through fully connected layers and were trained using the Adam optimizer with a learning rate of 0.0001, a batch size of 128, and 100 training epochs.

[0137] This invention uses the root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R²), and mean absolute error (MAE) as indicators to evaluate prediction performance. The specific formulas are as follows:

[0138]

[0139] To verify the prediction accuracy and reliability of the proposed model, this invention compares the performance metrics of these models, and the results are shown in Table 1:

[0140] Table 1

[0141]

[0142] As shown in Table 1, the WTC-iPST model proposed in this invention achieved the best results in transformer load forecasting. Its MAE, RMSE, MAPE, and R² all outperform the comparative models such as iTransformer, Transformer, TCN, and LSTM. Specifically, the MAE, RMSE, and MAPE of the proposed model are 1.21kW, 1.52kW, and 0.86%, respectively, with an R² of 0.9974, demonstrating a significant improvement in overall forecasting accuracy.

[0143] While ensuring the performance of the load forecasting model, it is also necessary to verify the performance of the transformer oil temperature forecasting model. The ANFIS model is compared with the Tranformer model, the Informer model, and the LSTM model. The results are shown in Table 2.

[0144] Table 2

[0145]

[0146] As shown in the table above, the ANFIS model of this invention is more accurate in predicting transformer oil temperature compared to other models.

[0147] Based on the above transformer load and oil temperature prediction results, a 110kV transformer with ODAF cooling was selected as the research object. Based on the load prediction results, hot spot temperature, and transformer equipment status, the transformer overload warning results were evaluated, as shown in Table 3.

[0148] Table 3

[0149]

[0150] As shown in Table 3, the prediction models show differences in their performance in heavy overload early warning. The Transformer+ANFIS model and the Transformer+Transformer model only identify one heavy load period, which is difficult to fully reflect the risk of distribution transformers operating under high load. The WTC-iPST+ANFIS model proposed in this paper is better than other models in terms of the accuracy of heavy overload early warning.

[0151] Figure 5 A schematic diagram of the load warning device for the transformer provided in this application is shown below. Figure 5 As shown, the transformer load warning device 40 provided in this embodiment includes:

[0152] The acquisition module 401 is used to acquire multi-source operating data of the target transformer, and to perform probabilistic interval prediction of the future load of the transformer based on the operating data to obtain a prediction sequence.

[0153] The extraction module 402 is used to extract load rate and environmental parameters from the operating data, calculate the top oil temperature based on the load rate and environmental parameters, and calculate the hot spot temperature based on the load rate and top oil temperature.

[0154] The determination module 403 is used to dynamically determine the transformer's load-bearing capacity sequence in the future period based on the hot spot temperature, top oil temperature and preset safety constraints. The safety constraints include the top oil temperature limit, the hot spot temperature limit, the upper limit of auxiliary equipment capacity and the insulation life loss threshold.

[0155] The early warning module 404 is used to compare the predicted sequence with the carrying capacity sequence point by point in the same time sequence. When the predicted load exceeds the corresponding early warning threshold of the carrying capacity at multiple consecutive time points, the corresponding heavy load warning or overload warning is triggered.

[0156] In one possible implementation, multiple input feature variables for load forecasting are extracted from the operational data. Multi-scale feature extraction is performed on the time series corresponding to each input feature variable to generate a multi-scale feature representation that integrates low-frequency trends and high-frequency fluctuations. The multi-scale feature representation of each input feature variable is mapped to a fixed-dimensional vector through an embedding layer. All embedding vectors are stacked along the variable dimension to form a joint sequence representation. A sparse self-attention mechanism is constructed based on the joint sequence representation to calculate the dynamic temporal dependencies between variables. Load forecast values ​​for multiple quantiles are generated based on the output of the sparse self-attention mechanism through a quantile regression model to form a forecast sequence.

[0157] In one possible implementation, the load rate and environmental parameters are input into a pre-built first fuzzy inference model to obtain the calculated value of the top oil temperature; the load rate and the calculated value of the top oil temperature are input into a pre-built second fuzzy inference model to obtain the calculated value of the hot spot temperature.

[0158] In one possible implementation, both the first fuzzy inference model and the second fuzzy inference model employ Gaussian membership functions, and the center parameter and width parameter of the Gaussian membership function are adaptively adjusted based on historical operating data.

[0159] In one possible implementation, the maximum allowable load value of the transformer at each point in time within a future period is determined, such that when the transformer operates at the maximum load value, the transformer's operating state meets safety constraints. These safety constraints include at least one of the following: top oil temperature is less than or equal to a preset top oil temperature limit, hot spot temperature is less than or equal to a preset hot spot temperature limit, operating current is less than or equal to a preset upper limit for auxiliary equipment capacity, and cumulative insulation life loss is less than or equal to an insulation life loss threshold. The maximum load values ​​are then arranged in chronological order to form a sequence of load-bearing capacities.

[0160] In one possible implementation, the future period is divided into multiple time intervals; for each time interval, the relative aging rate is determined based on the hot spot temperature corresponding to the time interval; the cumulative insulation life loss is obtained by multiplying and summing the relative aging rates of each time interval with the corresponding time interval length.

[0161] In one possible implementation, a heavy load warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of a first preset percentage range at multiple consecutive time points; an overload warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of a second preset percentage range at multiple consecutive time points; wherein the total duration of the multiple consecutive time points is not less than the lower limit of the preset time range, and the first preset percentage is less than the second preset percentage.

[0162] The transformer load warning device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0163] Figure 6 This is a structural schematic diagram of the transformer load warning device provided in this application. Figure 6 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0164] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0165] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0166] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0167] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0168] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0169] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0170] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0171] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0172] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0173] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0174] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0175] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0176] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0177] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0178] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A load pre-warning method of a transformer, characterized by, include: Obtain multi-source operating data of the target transformer, and perform probabilistic interval prediction of the future load of the transformer based on the operating data to obtain a prediction sequence; The load rate and environmental parameters are extracted from the operating data. The top oil temperature is calculated based on the load rate and environmental parameters, and the hot spot temperature is calculated based on the load rate and the top oil temperature. Based on the hot spot temperature, the top oil temperature, and preset safety constraints, the transformer's carrying capacity sequence for future periods is dynamically determined. The safety constraints include the top oil temperature limit, the hot spot temperature limit, the upper limit of auxiliary equipment capacity, and the insulation life loss threshold. The predicted sequence and the carrying capacity sequence are compared point by point in the same time sequence. When the predicted load exceeds the corresponding early warning threshold of the carrying capacity at multiple consecutive time points, the corresponding heavy load warning or overload warning is triggered.

2. The method of claim 1, wherein, Based on the aforementioned operating data, a probabilistic interval prediction of the transformer's future load is performed to obtain a prediction sequence, including: Multiple input feature variables for load forecasting are extracted from the operational data. Multi-scale feature extraction is performed on the time series corresponding to each input feature variable to generate a multi-scale feature representation that integrates low-frequency trends and high-frequency fluctuations. The multi-scale feature representation of each input feature variable is mapped to a fixed-dimensional vector through an embedding layer; Stack all embedded vectors along the variable dimension to form a joint sequence representation; A sparse self-attention mechanism is constructed based on the joint sequence representation to calculate the dynamic temporal dependencies between variables. The load prediction values ​​for multiple quantiles are generated based on the output of the sparse self-attention mechanism using a quantile regression model, forming the prediction sequence.

3. The method of claim 1, wherein, The calculation of the top oil temperature based on the load rate and environmental parameters, and the calculation of the hot spot temperature based on the load rate and the top oil temperature, include: The load rate and environmental parameters are input into a pre-built first fuzzy inference model to obtain the calculated value of the top oil temperature. The load rate and the calculated top oil temperature are input into a pre-built second fuzzy inference model to obtain the hot spot temperature calculation value.

4. The method of claim 3, wherein, Both the first fuzzy inference model and the second fuzzy inference model adopt Gaussian membership functions, and the center parameter and width parameter of the Gaussian membership function are adaptively adjusted based on historical running data.

5. The method of claim 1, wherein, Based on the hot spot temperature, the top oil temperature, and preset safety constraints, the transformer's load-bearing capacity sequence for future periods is dynamically determined, including: The maximum allowable load value of the transformer at each time point in the future period is determined, so that when the transformer is running at the maximum load value, the operating state of the transformer meets the safety constraints. The safety constraints include at least one of the following: top oil temperature is less than or equal to a preset top oil temperature limit, hot spot temperature is less than or equal to a preset hot spot temperature limit, operating current is less than or equal to a preset upper limit of auxiliary equipment capacity, and cumulative insulation life loss is less than or equal to an insulation life loss threshold. The maximum load values ​​are arranged in chronological order to form the carrying capacity sequence.

6. The method of claim 5, wherein, The method further includes: Divide the future period into multiple time intervals; For each time interval, the relative aging rate is determined based on the hotspot temperature corresponding to that time interval; The cumulative insulation life loss is obtained by multiplying and summing the relative aging rate of each time interval with the corresponding time length.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: When the predicted load exceeds the product of the carrying capacity and the upper limit of the first preset percentage range at multiple consecutive time points, an overload warning is triggered. An overload warning is triggered when the predicted load exceeds the product of the carrying capacity and the upper limit of the second preset percentage range at multiple consecutive time points. Wherein, the total duration of the multiple consecutive time points is not less than the lower limit of the preset time range, and the first preset percentage is less than the second preset percentage.

8. A load warning device for a transformer, characterized by include: The acquisition module is used to acquire multi-source operating data of the target transformer, and perform probabilistic interval prediction of the future load of the transformer based on the operating data to obtain a prediction sequence; The extraction module is used to extract load rate and environmental parameters from the operating data, calculate the top oil temperature based on the load rate and environmental parameters, and calculate the hot spot temperature based on the load rate and the top oil temperature. The determination module is used to dynamically determine the transformer's load-bearing capacity sequence in the future time period based on the hot spot temperature, the top oil temperature, and preset safety constraints. The safety constraints include the top oil temperature limit, the hot spot temperature limit, the upper limit of auxiliary equipment capacity, and the insulation life loss threshold. The early warning module is used to compare the predicted sequence with the carrying capacity sequence point by point in the same time sequence. When the predicted load exceeds the corresponding early warning threshold of the carrying capacity at multiple consecutive time points, the corresponding heavy load warning or overload warning is triggered.

9. A load warning device for a transformer, characterized by include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterised in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images