Transformer hot-spot temperature prediction method based on ACO optimization CNN-GRU combined neural network
By using ACO-optimized CNN-GRU combined neural network in transformer hotspot temperature prediction, the problems of insufficient prediction accuracy and local optimization in the prior art are solved, and a more efficient and scientific transformation hotspot temperature prediction is achieved.
Patent Information
- Application Number
- CN202411215128.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art has insufficient accuracy in the prediction of transformer hot spot temperature, is prone to local optimality, and is not scientific enough to process multi-source information, which affects the normal operation and prediction accuracy of the transformer.
Using ACO optimization-based CNN-GRU combination neural network, a two-layer one-dimensional CNN network and a two-layer gated loop unit are used to optimize weights and thresholds, and a hot spot temperature prediction model is built to improve prediction accuracy and avoid local maximums. Excellent question.
It significantly improves the accuracy and accuracy of the transformer hot spot temperature prediction, and realizes more scientific and reasonable multi-source information processing, which is suitable for practical engineering applications.
Smart Images

Figure CN119004078A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power equipment monitoring, and more specifically to a transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network. Background Art
[0002] With the development of social economy and the improvement of comprehensive national strength, the stable operation of power equipment system is directly related to the future development and construction of the country. Transformers play an important role in the entire power transmission system and distribution system, and their operating status is directly related to the stable operation of the entire power equipment system. Once a transformer operation failure occurs, it will not only threaten the safety of the power transmission system and distribution system, but also have a certain impact on the benefits of some production, processing and other related enterprises, thereby causing immeasurable economic losses to the enterprise. Among them, the high hot spot temperature of the winding will accelerate the aging of the insulation material and shorten the service life of the transformer. According to the guideline GB / T 1094.7-2008, the aging rate of the transformer will double for every 6°C above the temperature, so it is extremely important to implement the hot spot temperature monitoring and prediction of the transformer, which is of great significance to ensure the safe operation of the entire power system equipment and the protection of people's lives and property from threats.
[0003] After continuous research and progress, domestic and foreign researchers and experts have proposed many methods and decisions for hot spot temperature monitoring and prediction of transformers. The main methods are as follows: (1) The hot spot temperature can be directly measured by embedding temperature sensors in the coils of the windings. This method is simple and direct, but due to technical limitations in the early days, people could not determine the location of the hot spot, and it was necessary to embed more temperature sensors in the windings, which would affect the distribution of the electric field of the windings, increase electromagnetic interference, reduce the insulation effect of the windings, and affect the operation of the transformer. (2) Using ANSYS software to simulate the temperature field of the transformer, obtain the specific location information of the hot spot, and combine it with fiber grating temperature measurement technology. Although it can monitor the hot spot temperature data well, disassembling and installing the transformer in operation will increase maintenance costs and safety performance issues. (3) The guideline calculation method is a commonly used empirical calculation formula, which is simple to implement and has a wide range of applications. However, this method simplifies the heat transfer process during the operation of the transformer, ignores the nonlinear factors of the entire system, and there are certain errors in the calculation results. Taking into account the load drop and the wind speed in the environment, the wind speed correction factor is introduced to improve the empirical formula. Although the hot spot temperature is calculated more accurately than the standard empirical formula, the selection of characteristic parameters and correction factors has a greater impact on the results. (4) The Kalman filter algorithm is used to establish the state equation and measurement equation of the hot spot temperature, and the hot spot temperature is estimated in real time. This method has time lag and does not consider the influence of noise signals and environmental factors, so the prediction accuracy needs to be improved. (5) The least squares method is used to optimize the dual support vector regression machine. The prediction accuracy and performance are better than the traditional support vector regression machine, but the algorithm has a large computational complexity. (6) The neural network has a simple structure, good self-learning and adaptability, and has a good effect on hot spot temperature prediction. The standard principal component analysis method combined with the neural network is used to predict the hot spot temperature and achieves good results, but the standard principal component analysis method does not consider the differences between the original parameters, and the defect that the neural network is prone to fall into the local optimum has not been solved.
[0004] Therefore, how to improve the prediction accuracy of transformer hot spot temperature without affecting the normal operation of the transformer is an urgent problem to be solved by technical personnel in this field. Summary of the invention
[0005] In view of the above problems, the present invention provides a transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network to at least solve some of the technical problems mentioned in the above background technology.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network, characterized by comprising the following steps:
[0008] S1, obtaining the load rate, top oil temperature and ambient temperature of the target transformer;
[0009] S2, normalizing the load rate, top oil temperature and ambient temperature;
[0010] S3, input the normalized load rate, top oil temperature and ambient temperature into the CNN-GRU model optimized based on the ACO algorithm, and output the hot spot temperature prediction value of the target transformer.
[0011] Furthermore, the CNN-GRU model includes a double-layer one-dimensional CNN network unit and a double-layer gated recurrent unit;
[0012] In the double-layer one-dimensional CNN network unit, the normalized load rate, top oil temperature and ambient temperature are subjected to two consecutive convolutions by the convolution layer, and then the convolution results are subjected to pooling by the pooling layer. Finally, feature extraction is performed based on the pooling results by the fully connected layer to obtain a feature vector; the feature vector is denormalized to obtain a hot spot temperature prediction value of the target transformer;
[0013] The double-layer gated recurrent unit learns the feature vector and continuously updates the weight and bias of each neuron through momentum and adaptive learning rate, so that the output value of the loss function of the CNN-GRU model remains optimal.
[0014] Furthermore, it is characterized in that, in the two-layer one-dimensional CNN network unit, the number of convolution kernels in the first layer network is 16, and the number of convolution kernels in the second layer network is 32; the convolution mode is same convolution; and the convolution step size is 1;
[0015] Furthermore, the loss function of the CNN-GRU model is expressed as:
[0016]
[0017] Among them, Floss represents the loss function of the prediction model; n represents the total number of cycles; y t (i) represents the actual value of the hot spot temperature of the transformer during the i-th cycle; y p (i) represents the predicted value of the transformer hot spot temperature at the i-th cycle.
[0018] Furthermore, the CNN-GRU model is optimized based on the ACO algorithm, which specifically includes the following steps:
[0019] (1) Determine the M weights to be optimized and the thresholds corresponding to the weights in the CNN-GRU model. Each weight and the corresponding threshold are set to n random non-zero values and form a set I Pi ;
[0020] (2) At the initial moment, the kth ant among the m ants is from the set I Pi Starting from set I according to the state transition probability Pi Select an element j from the set; after selecting elements from all sets, a set of weights and thresholds of the CNN-GRU model are formed;
[0021] (3) When all m ants complete the cycle, m groups of initial weights and thresholds are obtained, and the CNN-GRU model is constructed and trained; the group of weights and thresholds with the smallest error during CNN-GRU model training is recorded, and the minimum error is compared with the expected error. If it is greater than the expected error, step (4) is executed, otherwise step (6) is executed;
[0022] (4) For set I Pi The pheromone amount of each element in is adjusted, where 1≤i≤M; i represents the i-th cycle; M represents the total number of cycles;
[0023] (5) Repeat steps (2) and (3) until all ants converge on the same optimal path or the maximum number of iterations is reached;
[0024] (6) Using the initial weights and thresholds of the optimal CNN-GRU model selected by the ACO algorithm, a CNN-GRU model optimized based on the ACO algorithm is constructed.
[0025] Furthermore, in step (4), the adjustment formula is expressed as:
[0026]
[0027] Among them, τ j (t+M) represents the amount of pheromone on each path after the ant has traversed all the cycles; Δτ j (I Pi ) indicates that each time the loop is completed, the set I P The pheromone increment of element j in i; η represents the residual rate of pheromone; t represents time t; Δτj k (I P i) indicates that the kth ant stays in set I in each cycle P The formula for the pheromone quantity of element i and j is:
[0028]
[0029] Among them, e k represents the error between the network output and the expected output obtained by training the CNN-GRU model with the weights and thresholds selected by ant k; Q represents the total amount of pheromone released by ant k.
[0030] Through the above technical solution, it can be known that compared with the prior art, the present invention discloses a transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network, which has the following beneficial effects: the present invention establishes a CNN-GRU combined neural network as a hot spot temperature prediction model, combines the ant colony algorithm (ACO algorithm) to globally optimize the network's optimal weights and thresholds, eliminates its defect of being easily trapped in local optimality, further improves the model prediction accuracy, and realizes the prediction of the transformer hot spot temperature. Compared with traditional methods, the present invention can use transformer multi-source information more scientifically and reasonably, and is more suitable for actual engineering applications.
[0031] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.
[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0034] Figure 1 A schematic flow chart of a transformer hot spot temperature prediction method based on an ACO optimized CNN-GRU combined neural network provided in an embodiment of the present invention.
[0035] Figure 2 A schematic diagram of the structure of a double-layer one-dimensional CNN network unit provided in an embodiment of the present invention.
[0036] Figure 3 A schematic diagram of the structure of a double-layer gated cycle unit provided in an embodiment of the present invention.
[0037] Figure 4 A schematic diagram of the ACO algorithm optimization CNN-GRU model process provided in an embodiment of the present invention.
[0038] Figure 5 A schematic diagram of the prediction results of the GRU model provided in an embodiment of the present invention.
[0039] Figure 6 A schematic diagram of the prediction results of the CNN-GRU model provided in an embodiment of the present invention.
[0040] Figure 7A schematic diagram of the ACO-CNN-GRU prediction results provided by an embodiment of the present invention.
[0041] Figure 8 A schematic diagram comparing the absolute errors of predictions of the GRU model, CNN-GRU model, and ACO-CNN-GRU model provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0043] See also Figure 1 As shown, the embodiment of the present invention discloses a transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network, comprising the following steps:
[0044] S1, obtaining the load rate, top oil temperature and ambient temperature of the target transformer;
[0045] S2, normalizing the load rate, top oil temperature and ambient temperature;
[0046] S3. Input the normalized load rate, top oil temperature and ambient temperature into the CNN-GRU model optimized based on the ACO algorithm, and output the hot spot temperature prediction value of the target transformer.
[0047] The above-mentioned CNN-GRU model includes a double-layer one-dimensional CNN network unit and a double-layer gated recurrent unit. Next, the above-mentioned CNN-GRU model optimized based on the ACO algorithm is described respectively.
[0048] 1. Two-layer one-dimensional CNN network unit
[0049] CNN is a feedforward neural network that can automatically learn feature representation from raw data and has strong feature extraction and pattern recognition capabilities. Its basic structure includes multiple convolutional layers, activation functions, pooling layers, and fully connected layers. The convolutional layer extracts features from the input data through convolution operations, and the feature map output by the convolutional layer can be nonlinearly transformed through activation functions. The pooling layer is used to reduce the size of the feature map and reduce the complexity of the model while retaining important feature information.
[0050] In the embodiment of the present invention, a Figure 2The double-layer one-dimensional CNN network unit shown in the figure has 16 convolution kernels in the first layer of the network and 32 convolution kernels in the second layer of the network; the convolution mode is same convolution; the convolution step is 1; in the double-layer one-dimensional CNN network unit, the load rate, top oil temperature and ambient temperature after normalization are convolved twice in a row by the convolution layer, and then the convolution result is pooled by the pooling layer, and finally the feature extraction is performed based on the pooling result by the fully connected layer to obtain the feature vector; the feature vector is denormalized to obtain the hot spot temperature prediction value of the target transformer;
[0051] 2. Two-layer Gated Recurrent Unit (GRU)
[0052] GRU is a variant of RNN and an improved model of LSTM. It is often used to process and predict sequence data. GRU solves the gradient vanishing and gradient exploding problems in traditional RNN by introducing a gating mechanism. It has a simpler structure and better training effect, which can greatly improve the training effect. Figure 3 As shown; where x t represents the input at time t; y t represents the output of the hidden node at time t; h t is the hidden state passed to the next node; h t-1 is the hidden layer state at time t-1, which contains the relevant information of the previous node; r t is the reset gate; t is the update gate; σ is the sigmoid function; tanh is the activation function, which limits the value between -1 and 1; h on tanh t is the candidate hidden layer state.
[0053] The embodiment of the present invention constructs a double-layer gated recurrent unit to learn the feature vector obtained by the above-mentioned double-layer one-dimensional CNN network unit, and continuously updates the weight and bias of each neuron through momentum and adaptive learning rate, so that the output value of the loss function of the CNN-GRU model remains optimal; the loss function of the CNN-GRU model is expressed as:
[0054]
[0055] Among them, Floss represents the loss function of the prediction model; n represents the total number of cycles; y t (i) represents the actual value of the transformer hot spot temperature during the i-th cycle; y p (i) represents the predicted value of the transformer hot spot temperature at the i-th cycle.
[0056] 3. Optimizing the CNN-GRU model based on the ACO algorithm
[0057] 1. Principle of Ant Colony Algorithm (ACO Algorithm)
[0058] The ant colony algorithm is an optimization algorithm proposed to simulate the path-finding behavior of ants in nature. It has strong global search capabilities. Its principle is:
[0059] M ants are randomly placed on n cities, and ant k (1≤k≤m) chooses the next city based on the distance between the pheromone on the path and the city. At time t, the probability of ant k moving from city o to city z is:
[0060]
[0061] Among them, [τ oz (t)] represents the amount of pheromone on the path at time t; [ω oz ] represents the heuristic information of the path, which is usually inversely proportional to the path length; a represents the information heuristic factor; b represents the expected heuristic factor; Z k represents the city that the kth ant needs to go to.
[0062] When the ants have traversed all cities, the amount of pheromone on each path is readjusted to:
[0063]
[0064] Among them, η represents the residual rate of pheromone; Δτ oz It represents the total amount of pheromone left on the path by m ants after each traversal; represents the amount of pheromone left by ant k on the path each time it traverses, where:
[0065]
[0066] Among them, Q represents the total amount of pheromone released by ant k, which is a constant; L represents the length of the path traversed by ant k during the traversal process;
[0067] When all ants' paths are optimal, or the maximum number of iterations is reached, the ant colony algorithm terminates.
[0068] 2. Principle of ACO algorithm optimization of CNN-GRU model
[0069] The ant colony algorithm has good global optimization capabilities. When optimizing the CNN-GRU combined neural network, the pheromone amount and path selection of the ants during path finding are adjusted according to the comparison results of the output error and expected error of the training samples, and the weights and threshold initial values of the CNN-GRU combined neural network are repeatedly optimized. The basic process can be found in Figure 4 As shown, specifically including:
[0070] (1) Determine the M weights to be optimized and the thresholds corresponding to the weights in the CNN-GRU model. Each weight and the corresponding threshold are set to n random non-zero values and form a set I Pi (1≤i≤M);
[0071] (2) At the initial moment, the kth ant (1≤k≤m) among the m ants comes from the set I Pi Starting from set I according to the state transition probability Pi Select an element j from the set; after selecting elements from all sets, a set of weights and thresholds of the CNN-GRU model are formed;
[0072] (3) When all m ants complete the cycle, m groups of initial weights and thresholds are obtained, and the CNN-GRU model is constructed and trained; the group of weights and thresholds with the smallest error during CNN-GRU model training is recorded, and the minimum error is compared with the expected error. If it is greater than the expected error, step (4) is executed, otherwise step (6) is executed;
[0073] (4) For set I Pi The pheromone amount of each element in is adjusted; where 1≤i≤M; i represents the i-th cycle; M represents the total number of cycles; the adjustment formula is expressed as:
[0074]
[0075] Among them, τ j (t+M) represents the amount of pheromone on each path after the ant has traversed all the cycles; Δτ j (I Pi ) represents the pheromone increment of element j in set IPi in each cycle; η represents the residual rate of pheromone; t represents time t; Indicates that the kth ant stays in set I in each cycle Pi The pheromone amount of element j is:
[0076]
[0077] Among them, e k represents the error between the network output obtained by training the CNN-GRU model with the weights and thresholds selected by ant k and the expected output; Q represents the total amount of pheromone released by ant k;
[0078] (5) Repeat steps (2) and (3) until all ants converge on the same optimal path or the maximum number of iterations is reached;
[0079] (6) Using the initial weights and thresholds of the optimal CNN-GRU model selected by the ACO algorithm, a CNN-GRU model optimized based on the ACO algorithm is constructed.
[0080] Next, a specific example is used to verify the effect of the transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network provided by the present invention.
[0081] The present invention collects real-time monitoring data of SFPSZ-180000 / 220 oil-immersed transformer in a substation, which is cooled by forced oil circulation (OD). 100 sets of temperature rise test data and operation data constitute the prediction sample set, 70 sets of samples are randomly selected as training sets, and the remaining 30 sets of samples are used as test sets. The test set samples are shown in Table 1.
[0082] Table 1 Transformer hot spot temperature test sample
[0083] Sample No. Load rate % Ambient temperature / ℃ Top oil temperature / ℃ Hot spot temperature / ℃ 1 132.7 81.8 33.5 111.4 2 150.1 89.2 33.1 124.9 3 38.7 46.4 26.8 52 4 42.4 53.3 31.2 58.9 5 34.4 46.8 27.3 51.4 6 39.2 49 31.1 52.6 7 81.2 65.2 32.9 82.5 8 37.1 50.1 27.9 57.6 9 80.1 64.3 33 81.5 10 133.1 79.8 32.7 109.3 11 98.1 74.9 33.2 79.7 12 41.3 49.3 26 54.8 13 35.5 47.2 28.1 51.4 14 40.8 50.5 31.7 53.3 15 35.5 50.3 27.7 57 16 99.8 69.2 32.9 92.2 17 34.4 46.6 26.3 51.3 18 131.9 81.8 32.7 111 19 136.6 83.6 32.8 114.1 20 35.5 47.2 28.4 51.4 21 43.4 54.4 28.2 59.7 22 132.1 81.6 32.7 110.8 23 79.8 64 32.9 81.2 24 44.5 55.2 28.6 59.7 25 35.5 46.9 27.8 51.3 26 152.8 84.3 32.7 118.3 27 130.9 83.9 33.2 113.3 28 95.3 73.1 32.9 76.2 29 37.6 47.3 26.6 52.7 30 33.9 46.9 27 51.4
[0084] The hot spot temperature is used as the network output, and the load rate, top oil temperature and ambient temperature are used as the network input. The prediction results of three network models, namely the GRU model, the CNN-GRU model and the CNN-GRU model optimized by the ACO algorithm (referred to as the ACO-CNN-GRU model), are analyzed.
[0085] The GRU model prediction results are as follows Figure 5 As shown, the prediction results of the CNN-GRU model are as follows Figure 6 As shown, the prediction results of the ACO-CNN-GRU model are as follows Figure 7 As shown in the figure, the absolute error comparison of the three models is Figure 8 As shown in the figure, through the comparison of the evaluation indicators of the three methods in Table 2, it can be seen that the ACO-CNN-GRU model is more accurate in predicting the hot spot temperature under the actual operation of the transformer. The fitting degree between the predicted curve and the actual curve is as high as 99.95%, the mean square error (MSE) of the prediction result is: 0.3238, the mean absolute error (MAE): 0.4424, the root mean square error (RMSE): 0.5690, the mean absolute percentage error (MAPE): 0.5764%, the sum of squared errors (SSE): 9.7141, and the mean square percentage error (MSPE): 0.0054%. It can be seen that the ACO-CNN-GRU model has accurate prediction effect, small error, high credibility, and is suitable for actual engineering application.
[0086] Table 2 Comparison of evaluation indicators of three methods
[0087]
[0088] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0089] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network, characterized in that: The steps include: S1, obtaining the load rate, top oil temperature and ambient temperature of the target transformer; S2, normalizing the load rate, top oil temperature and ambient temperature; S3, input the normalized load rate, top oil temperature and ambient temperature into the CNN-GRU model optimized based on the ACO algorithm, and output the hot spot temperature prediction value of the target transformer.
2. According to claim 1, a transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network is characterized in that: The CNN-GRU model includes a double-layer one-dimensional CNN network unit and a double-layer gated recurrent unit; In the double-layer one-dimensional CNN network unit, the normalized load rate, top oil temperature and ambient temperature are subjected to two consecutive convolutions by the convolution layer, and then the convolution results are subjected to pooling by the pooling layer. Finally, feature extraction is performed based on the pooling results by the fully connected layer to obtain a feature vector; the feature vector is denormalized to obtain a hot spot temperature prediction value of the target transformer; The double-layer gated recurrent unit learns the feature vector and continuously updates the weight and bias of each neuron through momentum and adaptive learning rate, so that the output value of the loss function of the CNN-GRU model remains optimal.
3. The transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network according to claim 2 is characterized in that: In the double-layer one-dimensional CNN network unit, the number of convolution kernels in the first layer of the network is 16, and the number of convolution kernels in the second layer of the network is 32; the convolution method is same convolution; and the convolution step size is 1.
4. The transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network according to claim 2 is characterized in that: The loss function of the CNN-GRU model is expressed as: Among them, Floss represents the loss function of the prediction model; n represents the total number of cycles; y t (i) represents the actual value of the transformer hot spot temperature during the i-th cycle; y p (i) represents the predicted value of transformer hot spot temperature at the i-th cycle.
5. The transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network according to claim 1 is characterized in that: The CNN-GRU model is optimized based on the ACO algorithm, which includes the following steps: (1) Determine the M weights to be optimized and the thresholds corresponding to the weights in the CNN-GRU model. Each weight and the corresponding threshold are set to n random non-zero values and form a set I Pi ; (2) At the initial moment, the kth ant among the m ants is from the set I Pi Starting from set I according to the state transition probability Pi Select an element j from the set; after selecting elements from all sets, a set of weights and thresholds of the CNN-GRU model are formed; (3) When all m ants complete the cycle, m groups of initial weights and thresholds are obtained, and the CNN-GRU model is constructed and trained; the group of weights and thresholds with the smallest error during CNN-GRU model training is recorded, and the minimum error is compared with the expected error. If it is greater than the expected error, step (4) is executed, otherwise step (6) is executed; (4) For set I Pi The pheromone amount of each element in is adjusted, where 1≤i≤M; i represents the i-th cycle; M represents the total number of cycles; (5) Repeat steps (2) and (3) until all ants converge on the same optimal path or the maximum number of iterations is reached; (6) Using the initial weights and thresholds of the optimal CNN-GRU model selected by the ACO algorithm, a CNN-GRU model optimized based on the ACO algorithm is constructed.
6. The transformer hot spot temperature prediction method based on ACO optimized CNN-GRU combined neural network according to claim 5 is characterized in that: In step (4), the adjustment formula is expressed as: Among them, τ j (t+M) represents the amount of pheromone on each path after the ant has traversed all the cycles; Δτ j (I Pi ) indicates that each time the loop is completed, the set I P The pheromone increment of element j in i; η represents the residual rate of pheromone; t represents time t; Indicates that the kth ant stays in set I in each cycle Pi The pheromone amount of element j is: Among them, e k represents the error between the network output and the expected output obtained by training the CNN-GRU model with the weights and thresholds selected by ant k; Q represents the total amount of pheromone released by ant k.
Citation Information
Cited By
Monitoring-based deepwater drilling riser system state prediction method, medium and system
CN120524802A
Transformer hot-spot temperature inversion method based on U-Net network architecture
CN120822409A
