Cable current-carrying capacity dynamic prediction method based on residual learning

Abstract

The invention relates to the field of cable thermal stability evaluation and dynamic capacity regulation and control, and particularly discloses a residual learning-based cable current-carrying capacity dynamic prediction method, which comprises the following steps of: acquiring multi-source cable data including global environment characteristics, local characteristics and physical model output, and preprocessing the multi-source cable data to obtain a pre-processed multi-source cable data; performing feature enhancement processing on the preprocessed multi-source cable data through an attention mechanism module; the attention mechanism module comprises a global attention mechanism module and a local attention mechanism module, inputting a result processed by the attention mechanism to the residual prediction model, obtaining a final temperature prediction value, calculating the maximum current-carrying capacity of the cable based on the temperature prediction value, determining the available range of the current-carrying capacity of the cable, and outputting the available range of the current-carrying capacity of the cable. And performing real-time updating and parameter correction on the residual prediction model based on the residual deviation. According to the method, the thermal circuit physical model and the residual learning algorithm are fused, and the method has high prediction accuracy and good dynamic response capability.

Description

Technical Field

[0001] The present invention relates to the field of power system operation status monitoring and intelligent dispatching, and in particular to a method for dynamic prediction of cable current carrying capacity based on residual learning. Background Art

[0002] The current carrying capacity of a power cable refers to the amount of current allowed to flow through the cable during stable operation. It is a key parameter that determines its transmission efficiency. Its upper limit is determined by the long-term temperature tolerance of the cable's main insulation. When the current carrying capacity is too high, the cable generates more heat, and the XLPE insulation material (cross-linked polyethylene insulation material) is prone to thermal oxidation and thermal cracking, accelerating insulation degradation and shortening its service life. However, if the current carrying capacity is too low, the cable's power transmission capacity will not be fully utilized, resulting in a waste of resources.

[0003] However, the complex underground cable installation environment, with dynamic changes in soil structure, moisture, and ambient temperature and humidity, significantly increases the difficulty of accurately assessing the current carrying capacity. Cable current carrying capacity is a core boundary condition for safe cable operation, and its accurate assessment is directly related to the dispatch efficiency and safety margin of the power system.

[0004] Therefore, in view of the characteristics and needs of the power system, the present invention proposes a dynamic prediction method for cable current carrying capacity based on residual learning, which learns and dynamically corrects the prediction error of the traditional physical model in real time, improves the calculation accuracy, and predicts the available range of cable current carrying capacity. Summary of the Invention

[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a dynamic prediction method for cable current carrying capacity based on residual learning to achieve accurate prediction of cable current carrying capacity and adapt to external environmental factors of the cable.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions: The present invention provides a method for dynamically predicting cable current carrying capacity based on residual learning, comprising: Acquire multi-source cable data including global environmental features, local features and physical model output, and pre-process the multi-source cable data; Performing feature enhancement processing on the pre-processed multi-source cable data through an attention mechanism module; the attention mechanism module includes a global attention mechanism module and a local attention mechanism module; The result processed by the attention mechanism is input into the residual prediction model to obtain the final temperature prediction value; The maximum current carrying capacity of the cable is calculated based on the temperature prediction value, and the available current carrying capacity range of the cable is determined; The residual prediction model is updated and parameters are corrected in real time based on the residual deviation.

[0007] Preferably, the global environmental characteristics include ambient air temperature, soil temperature, tunnel wind speed, solar radiation intensity, relative humidity and cable load; the local characteristics include hotspot temperature, average temperature along the line, hotspot temperature difference and temperature spatial gradient, wherein the hotspot temperature and average temperature along the line are respectively the highest temperature value and average temperature value of the distributed optical fiber temperature detection system, and the hotspot temperature difference is the difference between the hotspot temperature and the average temperature along the line.

[0008] Preferably, the physical model output is the cable temperature in the thermal steady-state formula, specifically:

[0009] Where, and The cables and soil are The temperature at the moment, is the physical model output, is the thermal resistance, for The load current at that moment.

[0010] Preferably, the method for preprocessing multi-source cable data is specifically as follows: Time-align and normalize multi-source cable data according to timestamps; The normalized multi-source cable data are integrated into a global feature subset and a local feature subset, where the global feature subset is the data corresponding to the global environmental features, and the local feature subset is the data corresponding to the local features and the physical model output.

[0011] Preferably, the method for performing feature enhancement processing on pre-processed multi-source cable data through an attention mechanism module includes: Randomly initialize the weight matrix and calculate the attention score:

[0012] Where, is the attention score, is the weight matrix, for The global feature subset at time t, is the bias phasor; The attention scores are mapped to the interval [0, 1] through the Softmax function to obtain the normalized attention weights, and the sum of the normalized attention weights is guaranteed to be 1; Based on the normalized attention weights, the global feature subset is weighted element-wise:

[0013] Where, express The corresponding attention-enhancing features, is the normalized weight; Based on the above steps, the local feature subset is processed by the local attention module to obtain the attention-enhanced features corresponding to the local feature subset, and the attention-enhanced features corresponding to the global feature subset and the local feature subset are spliced ​​into a vector:

[0014] Where, for The concatenated vector of the moment, and are the attention enhancement features corresponding to the global feature subset and the local feature subset respectively.

[0015] Preferably, the residual prediction model is composed of a three-layer multi-layer perceptron, specifically:

[0016]

[0017]

[0018] Where, and are the output vectors of the first hidden layer and the second hidden layer respectively, is the activation function, 、 and are the weight matrices of the first hidden layer, the second hidden layer, and the third hidden layer, respectively. 、 and are the bias terms of the first hidden layer, the second hidden layer, and the third hidden layer, respectively. For the predicted Temperature residual value at time t; The residual prediction model is trained with the goal of minimizing the physical constraint loss term, where the physical constraint loss term is expressed as:

[0019] Where, is the physical constraint loss value, for The true residual value at time , To control the weight of the physical constraints, for The spatial Laplace gradient penalty of the temperature at the moment; The final temperature prediction value is obtained through the trained residual prediction model:

[0020] Where, for The temperature forecast value at the moment.

[0021] Preferably, the method for calculating the maximum current carrying capacity of the cable based on the temperature prediction value is specifically as follows: Make the temperature prediction value equal to the cable limit temperature and then calculate the maximum current carrying capacity:

[0022] Where, is the maximum current carrying capacity to be calculated, is the residual prediction model Always The temperature residual value under .

[0023] Preferably, the method for real-time updating and parameter correction of the residual prediction model based on the residual deviation is specifically as follows: The actual conductor temperature is obtained at preset intervals, the actual residual value is determined, and the difference between the actual residual value and the model predicted residual value is calculated. :

[0024] Where, for The difference between the actual residual value at time t and the model predicted residual value; like When the preset limit value is exceeded, the residual prediction model is judged to be biased, and the parameters of the residual prediction model are adjusted through the posterior probability update formula in the Bayesian neural network, where the residual loss function is:

[0025] Where, is the residual loss value, is the total number of samples, For the The difference between the true residual value corresponding to the sample and the model predicted residual; Based on the residual loss function, the residual prediction model parameters are locally updated by the gradient descent method:

[0026] Where, and are the model parameters before and after updating, To control the step size of each update, is the gradient of the residual loss function with respect to the residual prediction model parameters; If the prediction result of the residual prediction model still deviates from the preset threshold after local update, the residual prediction model is adjusted through the physical parameter correction method.

[0027] Preferably, the method of adjusting the residual prediction model by the physical parameter correction method is specifically as follows: Calculate the thermal resistance loss value:

[0028] Where, is the thermal resistance loss value, for The actual temperature of the conductor at the time for Thermal resistance The physical model output under Based on the thermal resistance loss value, the gradient descent method is used to correct the equivalent thermal resistance parameters:

[0029] Where, is the corrected thermal resistance parameter.

[0030] The embodiments of the present invention have at least the following beneficial effects: (1) The present invention adopts a dual-channel residual learning structure to learn and dynamically correct the prediction error of the traditional physical model in real time, which can significantly improve the calculation accuracy.

[0031] (2) The present invention introduces a physical constraint loss term to ensure that the prediction results meet the cable thermal safety threshold requirements from the algorithm level.

[0032] (3) The present invention integrates the Bayesian online update mechanism and adaptive parameter iteration to effectively solve the drift problem of the residual prediction model in long-term operation, thereby achieving continuous optimization and long-term stable operation of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention provides a flow chart of a method for dynamically predicting cable current carrying capacity based on residual learning.

[0034] Figure 2 Flowchart of real-time update and parameter correction in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0036] Herein, suffixes such as "module," "component," or "unit" used to represent elements are only used to facilitate description of the present invention and have no specific meaning. Therefore, "module," "component," or "unit" may be used interchangeably.

[0037] As used herein, terms such as "upper," "lower," "inner," "outer," "front," "back," "one end," and "the other end" indicate positions or locations based on those shown in the accompanying drawings. These terms are intended solely to facilitate and simplify the description of the present invention and are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0038] As used herein, "and / or" includes any and all combinations of one or more of the associated listed items.

[0039] Herein, "plurality" means two or more than two, ie, it includes two, three, four, five, etc.

[0040] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0041] As used herein, unless otherwise expressly specified or limited, the terms "installed," "provided with," and "connected" should be understood broadly. For example, "connected" may refer to a fixed connection, a detachable connection, or an integral connection; it may refer to a mechanical connection, a direct connection, an indirect connection via an intermediate medium, or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention on a case-by-case basis.

[0042] At present, the following three methods are mainly used in engineering practice to evaluate the current carrying capacity: (1) Experimental method: The experimental method constructs a temperature-rise test platform and measures cable temperature changes under different installation and loading conditions, thereby inferring the cable's maximum current-carrying capacity under those conditions. This method offers high accuracy and directness, making it suitable for small-scale, specific studies. However, it has the following significant drawbacks: Poor versatility: It is only applicable to specific scenarios corresponding to experimental conditions and cannot be generalized to other working conditions; High resource consumption: The experimental cycle is long, the cost is high, and a large amount of manpower, equipment and site resources are required; Difficult to apply in real time: Unable to adapt to dynamic environmental changes during cable operation, and its practicality is limited.

[0043] (2) Analytical method: Analytical methods typically construct a cable thermal circuit model based on thermoelectric analogy theory and utilize international standards such as IEC 60287 (steady-state) and IEC 60853 (transient) to estimate cable conductor temperature and rated current. Essentially, this method simplifies the complex cable heat conduction problem into a series of equivalent circuit calculations involving thermal resistance, current, and heat capacitance. This method provides an important foundation for industry standards, but its limitations are becoming increasingly apparent: Serious physical simplification: The standard does not provide detailed modeling for complex boundary conditions such as soil moisture, water migration, and cable heat sources; Conservative parameter assumptions: For example, parameters such as thermal resistance and resistivity are set as constants, without considering their dynamic changes with temperature; Model static rigidity: Difficult to respond to dynamic changes such as environmental conditions, installation status, and cable aging; Significant prediction error: Under special laying methods (such as multiple circuits, shallow burial, heat source interference, etc.), the prediction error can exceed 10~15°C, seriously affecting the system scheduling accuracy.

[0044] (3) Numerical method: Numerical methods benefit from the development of modern computing power. By discretizing the cable structure and applying methods such as the finite difference method (FDM), finite element method (FEM), and boundary element method (BEM) to solve the cable temperature field distribution, they can theoretically simulate more complex physical processes. However, there are multiple obstacles in practical application: High computational complexity: The cable structure and surrounding environment need to be discretized into tens of thousands of units to solve a system of nonlinear partial differential equations, resulting in long computation time and low efficiency. The model is highly dependent on parameters: it is sensitive to input parameters such as soil thermal conductivity and insulation material thermal resistance. If the measured data is inaccurate, it is easy to cause systematic errors. Lack of real-time performance: It is difficult to adapt to dynamic situations such as rapid load fluctuations and sudden heat sources, and cannot meet the actual scheduling requirements of second-level response; High hardware threshold: The model requires a high-performance computing platform and lacks the ability to be widely deployed; Difficulty in coupling modeling: The electromagnetic-thermal-environment coupling process is highly nonlinear, and existing numerical methods make it difficult to construct a universal and transferable cross-scenario model.

[0045] In general, while experimental methods offer high accuracy, they are also costly, time-consuming, and lack adaptability, making them difficult to meet the real-time and universal requirements of power grid operation. Analytical methods offer an engineering reference path, but rely heavily on simplified physical assumptions and struggle to handle dynamic load variations and soil condition uncertainties. Numerical methods offer high accuracy but rely on extensive computing resources, are challenging to deploy, and are difficult to apply to rapid assessment and real-time scheduling in real-world scenarios.

[0046] To overcome the above problems, see Figure 1 As shown, the present invention provides a method for dynamic prediction of cable current carrying capacity based on residual learning, including S1 to S5: S1 obtains multi-source cable data including global environmental features, local features and physical model output, and preprocesses the multi-source cable data.

[0047] In this embodiment, data on global environmental characteristics are obtained from a Supervisory Control and Data Acquisition (SCADA) system and a meteorological data interface, and data containing local characteristics are obtained through a distributed temperature sensing (DTS) system with spatial resolution (one point per meter).

[0048] Among them, the global environmental characteristics include ambient air temperature, soil temperature, tunnel wind speed, solar radiation intensity, relative humidity and cable load; the local characteristics include hotspot temperature, average temperature along the line, hotspot temperature difference and temperature spatial gradient, among which the hotspot temperature and average temperature along the line are the maximum temperature value and average temperature value of the distributed optical fiber temperature detection system respectively, and the hotspot temperature difference is the difference between the hotspot temperature and the average temperature along the line.

[0049] Specifically, the output of the physical model is the cable temperature in the thermal steady-state formula, which is:

[0050] Where, and The cables and soil are The temperature at the moment, is the physical model output, is the thermal resistance, for The load current at that moment.

[0051] Among them, the resistance needs to be determined according to the cable structure and laying conditions.

[0052] Furthermore, all data are aligned by timestamp, and in order to prevent the dimension from affecting model training, each data item is normalized, and the three types of data are integrated into global feature subsets and local feature subsets. The global feature subset contains 6 items, which are the data corresponding to the global environmental features, and the local feature subset contains 5 items, which are the data corresponding to the local features and physical model outputs. The final model input vector has a total of 11 items.

[0053] S2 performs feature enhancement processing on the preprocessed multi-source cable data through an attention mechanism module; the attention mechanism module includes a global attention mechanism module and a local attention mechanism module.

[0054] The current carrying capacity and temperature of the cable are affected by many factors, but the influencing factors are different in different time, space and environment. The attention mechanism module can identify and focus on the input factors that have the greatest impact on the error. Therefore, a global attention module and a local attention module are set to perform feature enhancement processing.

[0055] For example, the air velocity in one duct has a greater impact on temperature, while humidity has almost no effect, while in another environment the opposite is true, with the air velocity having a greater impact on temperature at high loads.

[0056] Specifically, before training begins, the weight matrix is ​​randomly initialized, where the elements are randomly sampled and distributed in a small range, and the uniform distribution (-0.1, 0.1) and the bias vector (set to 0) are determined; Calculate the attention score:

[0057] Where, is the attention score, is the weight matrix, for The global feature subset at time t, is the bias phasor; The attention scores are mapped to the interval [0, 1] through the Softmax function to obtain the normalized attention weights, and the sum of the normalized attention weights is guaranteed to be 1; Based on the normalized attention weights, the global feature subset is weighted element-wise:

[0058] Where, express The corresponding attention-enhancing features, is the normalized weight; Based on the above steps, the local feature subset is processed by the local attention module to obtain the attention-enhanced features corresponding to the local feature subset, and the attention-enhanced features corresponding to the global feature subset and the local feature subset are spliced ​​into a vector:

[0059] Where, for The concatenated vector of the moment, and are the attention enhancement features corresponding to the global feature subset and the local feature subset respectively.

[0060] S3 inputs the results processed by the attention mechanism into the residual prediction model to obtain the final temperature prediction value.

[0061] Specifically, the residual prediction model is composed of a three-layer multi-layer perceptron, specifically:

[0062]

[0063]

[0064] Where, and are the output vectors of the first hidden layer and the second hidden layer respectively, is the activation function, 、 and are the weight matrices of the first hidden layer, the second hidden layer, and the third hidden layer, respectively. 、 and are the bias terms of the first hidden layer, the second hidden layer, and the third hidden layer, respectively. For the predicted Temperature residual value at time t; The residual prediction model is trained with the goal of minimizing the physical constraint loss term, where the physical constraint loss term is expressed as:

[0065] Where, is the physical constraint loss value, for The true residual value at time , To control the weight of the physical constraints, for Spatial Laplace gradient penalty of the temperature at each moment; In the above formula, the first term is the square of the difference between the predicted residual and the true residual, is the data residual term, and the second term is the spatial Laplace gradient penalty of temperature, which is used to suppress non-physical drastic temperature mutations, is the physical constraint term, where Requires experimental tuning.

[0066] The final temperature prediction value is obtained through the trained residual prediction model:

[0067] Where, for The temperature forecast value at the moment.

[0068] S4 calculates the maximum current carrying capacity of the cable based on the temperature prediction value and determines the available current carrying capacity range of the cable.

[0069] Specifically, the method for calculating the maximum current carrying capacity of the cable based on the temperature prediction value is as follows: Make the temperature prediction value equal to the cable limit temperature and then calculate the current carrying capacity:

[0070] Where, is the maximum current carrying capacity to be calculated, is the residual prediction model Always The temperature residual value under .

[0071] In this embodiment, dynamic prediction is performed on XLPE cable (cross-linked polyethylene insulated cable), and its limit temperature is 90 , so that the temperature prediction value is equal to the cable limit temperature, and then the formula of the residual prediction model is combined to solve the maximum current carrying capacity.

[0072] Subsequently, the available current carrying capacity range of the cable is cut off based on the maximum current carrying capacity obtained by the solution to ensure that the current load current does not exceed the maximum current carrying capacity.

[0073] S5 performs real-time updates and parameter corrections on the residual prediction model based on the residual deviation.

[0074] It should be understood that in actual conditions, the soil may cause changes in the thermal conductivity of the cable due to rain or drought, which in turn affects the prediction accuracy of the residual prediction model. Therefore, the model parameters are updated in real time to solve the problems of decreased accuracy and systematic deviation caused by environmental changes, cable aging and other factors during the long-term operation of the model.

[0075] Specifically, see Figure 2As shown in FIG, the method for real-time updating and parameter correction of the residual prediction model based on the residual deviation is specifically as follows: The actual conductor temperature is obtained at preset intervals, the actual residual value is determined, and the difference between the actual residual value and the model predicted residual value is calculated. :

[0076] Where, for The difference between the actual residual value at time t and the model predicted residual value; like When the preset limit value is exceeded, the residual prediction model is judged to be biased, and the parameters of the residual prediction model are adjusted through the posterior probability update formula in the Bayesian neural network, where the residual loss function is:

[0077] Where, is the residual loss value, is the total number of samples, For the The difference between the true residual value corresponding to the sample and the model predicted residual; Based on the residual loss function, the residual prediction model parameters are locally updated by the gradient descent method:

[0078] Where, and are the model parameters before and after updating, To control the step size of each update, is the gradient of the residual loss function with respect to the residual prediction model parameters; If the residual prediction model still deviates from the preset threshold after the local update of the preset number of times, the residual prediction model is adjusted by the physical parameter correction method.

[0079] Optionally, the preset interval time needs to be adjusted according to actual working conditions.

[0080] In this embodiment, the total number of samples is 10.

[0081] Furthermore, if local updates reveal that the model's predictions are always too high or too low and do not meet the accuracy requirements, this indicates that the key parameters in the physical model are no longer accurate. For example, if the thermal conductivity of the soil increases due to seasonal changes and increased rainfall, then a correction of the physical parameters will be triggered.

[0082] Among them, the method of adjusting the residual prediction model by the physical parameter correction method is specifically as follows: Calculate the thermal resistance loss value:

[0083] Where, is the thermal resistance loss value, for The actual temperature of the conductor at the time for Thermal resistance The physical model output under Based on the thermal resistance loss value, the gradient descent method is used to correct the equivalent thermal resistance parameters:

[0084] Where, is the corrected thermal resistance parameter.

[0085] Among them, the step size of each update needs to be set according to the actual situation. The smaller the value, the more stable it is; the larger the value, the faster the convergence but the more unstable it is.

[0086] In general, compared with traditional methods, the present invention can not only reflect the impact of environmental and load changes on cable temperature rise in real time, but also achieve continuous optimization and long-term stable operation of the model through Bayesian incremental update and parameter self-calibration mechanism.

[0087] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. A dynamic prediction method for cable current carrying capacity based on residual learning, characterized in that: The following steps are involved: Acquire multi-source cable data including global environmental features, local features and physical model output, and pre-process the multi-source cable data; The pre-processed multi-source cable data is subjected to feature enhancement through the attention mechanism module; The attention mechanism module includes a global attention mechanism module and a local attention mechanism module; The result processed by the attention mechanism is input into the residual prediction model to obtain the final temperature prediction value; The maximum current carrying capacity of the cable is calculated based on the temperature prediction value, and the available current carrying capacity range of the cable is determined; The residual prediction model is updated and parameters are corrected in real time based on the residual deviation.

2. A cable current carrying capacity dynamic prediction method based on residual learning according to claim 1, characterized in that: The global environmental characteristics include ambient air temperature, soil temperature, tunnel wind speed, solar radiation intensity, relative humidity and cable load; the local characteristics include hotspot temperature, average temperature along the line, hotspot temperature difference and temperature spatial gradient, where the hotspot temperature and average temperature along the line are respectively the highest temperature value and average temperature value of the distributed optical fiber temperature detection system, and the hotspot temperature difference is the difference between the hotspot temperature and the average temperature along the line.

3. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 1 is characterized in that: The output of the physical model is the cable temperature in the thermal steady-state formula, specifically: Where, and The cables and soil are The temperature at the moment, is the physical model output, is the thermal resistance, for The load current at that moment.

4. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 1 is characterized in that: The method for preprocessing multi-source cable data is specifically as follows: Time-align and normalize multi-source cable data according to timestamps; The normalized multi-source cable data are integrated into a global feature subset and a local feature subset, where the global feature subset is the data corresponding to the global environmental features, and the local feature subset is the data corresponding to the local features and the physical model output.

5. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 1, characterized in that: The method for performing feature enhancement processing on pre-processed multi-source cable data through an attention mechanism module includes: Randomly initialize the weight matrix and calculate the attention score: Where, is the attention score, is the weight matrix, for The global feature subset at time t, is the bias phasor; The attention scores are mapped to the interval [0, 1] through the Softmax function to obtain the normalized attention weights, and the sum of the normalized attention weights is guaranteed to be 1; Based on the normalized attention weights, the global feature subset is weighted element-wise: Where, is the attention enhancement feature corresponding to the global feature subset, is the normalized weight; Based on the above steps, the local feature subset is processed by the local attention module to obtain the attention-enhanced features corresponding to the local feature subset, and the attention-enhanced features corresponding to the global feature subset and the local feature subset are spliced ​​into a vector: Where, for The concatenated vector of the moment, is the attention enhancement feature corresponding to the local feature subset.

6. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 1, characterized in that: The residual prediction model is composed of a three-layer multi-layer perceptron, specifically: Where, and are the output vectors of the first hidden layer and the second hidden layer respectively, is the activation function, 、 and are the weight matrices of the first hidden layer, the second hidden layer, and the third hidden layer, respectively. 、 and are the bias terms of the first hidden layer, the second hidden layer, and the third hidden layer, respectively. For the predicted Temperature residual value at time t; The residual prediction model is trained with the goal of minimizing the physical constraint loss term, where the physical constraint loss term is expressed as: Where, is the physical constraint loss value, for The true residual value at time , To control the weight of the physical constraints, for The spatial Laplace gradient penalty of the temperature at the moment; The temperature residual value is obtained through the trained residual prediction model to determine the final temperature prediction value: Where, for The temperature forecast value at the moment.

7. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 1, characterized in that: The method for calculating the maximum current carrying capacity of the cable based on the temperature prediction value is specifically as follows: Make the temperature prediction value equal to the cable limit temperature and then calculate the maximum current carrying capacity: Where, is the maximum current carrying capacity to be calculated, is the residual prediction model Always The temperature residual value under .

8. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 1, characterized in that: The method for real-time updating and parameter correction of the residual prediction model based on the residual deviation is specifically as follows: The actual conductor temperature is obtained at preset intervals, the actual residual value is determined, and the difference between the actual residual value and the model predicted residual value is calculated. : Where, for The difference between the actual residual value at time t and the residual predicted by the model; like When the preset limit value is exceeded, the residual prediction model is judged to be biased, and the parameters of the residual prediction model are adjusted through the posterior probability update formula in the Bayesian neural network, where the residual loss function is: Where, is the residual loss value, is the total number of samples, For the The difference between the true residual value corresponding to the sample and the model predicted residual; Based on the residual loss function, the residual prediction model parameters are locally updated by the gradient descent method: Where, and are the model parameters before and after updating, To control the step size of each update, is the gradient of the residual loss function with respect to the residual prediction model parameters; If the prediction result of the residual prediction model still deviates from the preset threshold after local update, the residual prediction model is adjusted through the physical parameter correction method.

9. The method for dynamic prediction of cable current carrying capacity based on residual learning according to claim 8, characterized in that: The method of adjusting the residual prediction model by the physical parameter correction method is specifically as follows: Calculate the thermal resistance loss value: Where, is the thermal resistance loss value, for The actual temperature of the conductor at the time for Thermal resistance The physical model output under Based on the thermal resistance loss value, the thermal resistance parameters are corrected using the gradient descent method: Where, is the corrected thermal resistance parameter.

Images