Transformer temperature and fault prediction method based on LSTM technology

Abstract

The invention relates to a transformer temperature and fault prediction method based on an LSTM technology, and the method comprises the steps: carrying out the continuous data collection of a temperature sensor at a monitoring point of a power distribution area, and carrying out the analysis of the temperature change data related to the time, the transformer temperature prediction model of the LSTM with a time sequence as correlation is established for data analysis and prediction, the transformer temperature prediction model of the LSTM comprises an input layer, a hidden layer, an output layer and model training, and the input layer is used for reading and preprocessing a data set and dividing a training test data set; the hidden layer is used for constructing an LSTM hidden layer and a full-connection neural layer; the output layer is used for verification processing of a prediction result; the model training is used for being responsible for loss function calculation, Adam algorithm optimization and iterative calculation of the model training.

Description

Technical Field

[0001] The present invention relates to the field of transformer substation safety technology, and in particular to a transformer temperature and fault prediction method based on LSTM technology. Background Art

[0002] Distribution transformers are key equipment for power transmission and transformation in power systems. If a fault occurs during operation, it will cause large-scale power outages and other accidents, causing serious economic losses and adverse social reactions. Transformer temperature is one of the important indicators of equipment operation. By monitoring its temperature, most safety hazards can be discovered and predicted in advance, thereby reducing the incidence of accidents and ensuring the safe and stable operation of equipment. During the operation of the distribution station area, not only gradual faults will occur, but also sudden faults will occur. Although the temperature prediction of the system can further solve the possibility of gradual faults evolving into more serious accidents, it cannot effectively solve sudden faults. Sudden faults such as high-temperature fires are difficult to handle quickly because they occur suddenly. To address such problems, the project has designed an intelligent linkage fire extinguishing system. When a fire or other accident is detected in the distribution station area, the intelligent fire extinguishing system is activated to quickly extinguish the fire and reduce the loss of the accident.

[0003] Through the study of digital twin technology, the present invention conducts technical analysis on the power and environmental monitoring of transformers in distribution substations, makes full use of its physical model, sensor data update, operation history and other data, integrates multi-disciplinary, multi-physical quantity, multi-scale and multi-probability simulation processes, completes spatial mapping in the computer system, and establishes an information mirror model for distribution substations. Through the information mirror model, real-time monitoring of the operating status of transformers in distribution substations is achieved. This method is simple and the prediction results are highly accurate. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a transformer temperature and fault prediction method based on LSTM technology.

[0005] To solve the above problems, the technical solution adopted by the present invention is:

[0006] A transformer temperature and fault prediction method based on LSTM technology, comprising:

[0007] Continuously collect data from the temperature sensors at the monitoring points in the distribution substation area, analyze the time-related temperature change data, and create a transformer temperature prediction model with LSTM based on time series correlation for data analysis and prediction:

[0008] The transformer temperature prediction model of the LSTM includes: an input layer, a hidden layer, an output layer and model training, wherein the input layer is used to read and preprocess the data set, and divide the training test data set; the hidden layer is used to build the LSTM hidden layer and the fully connected neural layer; the output layer is used as a verification process for the prediction result; and the model training is responsible for the loss function calculation, the Adam algorithm optimization and the iterative calculation of the model training.

[0009] The LSTM transformer temperature prediction model generation method includes:

[0010] Get temperature time series sample data;

[0011] Preprocessing the temperature time series sample data to obtain training data samples and test data samples;

[0012] The training data samples are input into the LSTM long short-term memory network for training to obtain the transformer temperature prediction model of the LSTM.

[0013] As an implementation manner of the present invention, preprocessing the temperature time series sample data includes homogenizing the data.

[0014] As an embodiment of the present invention, at time t, the transformer temperature prediction model input of the LSTM includes the current input variable x t , the hidden layer state variable h at the previous moment t-1 and the cell state variable C at the previous moment t-1 , passing through i t (Input gate), o t (output gate) and f t (Forget gate), and use the tanh activation function to generate a new memory C′ t , the model output includes the current output variable h t and the current cell state variable C t , whose expression is:

[0015] C t =f t C t-1 +i t C′ t

[0016] h t =o t tanh(C t )

[0017] Assume that the input x of the model is x1, x2, x3, ..., x n , and x∈R T, i = 1, 2, ..., n; n is the input dimension, T represents the time step, the top oil temperature is the output, and the output sequence is y = y1, y2, ..., y n , and substitute them into the LSTM model expression respectively to complete the construction of the LSTM model for this scenario.

[0018] As an implementation mode of the present invention, two loss functions are used for calculation, which makes the prediction results and the temperature melon degree higher and expands the difference between the relationships.

[0019] As an implementation mode of the present invention, the loss function is obtained by using the following formula:

[0020] loss=max(0,-y*(S(r;q)-S(r';q))+margin)

[0021] loss2=y*|cos(r,r')|*max(0,cos(r,r)-γ).

[0022] As an embodiment of the present invention, the method comprises:

[0023] After determining that the predicted temperature value is greater than the temperature threshold range associated with the current distribution transformer fault strategy, the distribution transformer continues to be overloaded based on the current distribution transformer management strategy until the predicted temperature value is within the temperature threshold range.

[0024] The beneficial effects of adopting the above technical solution are:

[0025] The transformer temperature and fault prediction method based on LSTM technology provided by the present invention includes: continuous data collection of temperature sensors at monitoring points in distribution substations, and by analyzing the temperature change data related to time, creating a transformer temperature prediction model of LSTM with time series as correlation for data analysis and prediction: the transformer temperature prediction model of LSTM includes: input layer, hidden layer, output layer and model training, wherein the input layer is used to read and preprocess the data set, and the division of the training test data set; the hidden layer is used to build the LSTM hidden layer and the fully connected neural layer; the output layer is used as the verification processing of the prediction result; the model training is responsible for the loss function calculation, the Adam algorithm optimization and the iterative calculation of the model training. The method makes full use of its physical model, the update of sensor data, the operation history and other data, integrates multi-disciplinary, multi-physical quantity, multi-scale, multi-probability simulation processes, completes the spatial mapping in the computer system, establishes the distribution substation information mirror model, and realizes the real-time monitoring of the operation status of the transformer in the distribution substation through the information mirror model. The method is simple and the prediction result has high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a structural block diagram of an LSTM provided by the implementation of the present invention. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is clearly and completely described below in conjunction with specific embodiments.

[0028] The embodiment of the present invention provides a transformer temperature and fault prediction method based on LSTM technology, the method comprising:

[0029] Continuously collect data from the temperature sensors at the monitoring points in the distribution substation area, analyze the time-related temperature change data, and create a transformer temperature prediction model with LSTM based on time series correlation for data analysis and prediction:

[0030] The transformer temperature prediction model of the LSTM includes: an input layer, a hidden layer, an output layer and model training, wherein the input layer is used to read and preprocess the data set, and to divide the training and test data set; the hidden layer is used to build the LSTM hidden layer and the fully connected neural layer; the output layer is used as a verification process for the prediction result; the model training is responsible for the loss function calculation, the Adam algorithm optimization and the iterative calculation of the model training;

[0031] The LSTM transformer temperature prediction model generation method includes:

[0032] Get temperature time series sample data;

[0033] Preprocessing the temperature time series sample data to obtain training data samples and test data samples; the preprocessing includes homogenizing the data;

[0034] The training data samples are input into the LSTM long short-term memory network for training to obtain the transformer temperature prediction model of the LSTM.

[0035] At time t, the transformer temperature prediction model input of the LSTM includes the current input variable x t , the hidden layer state variable h at the previous moment t-1 and the cell state variable C at the previous moment t-1 , passing through i t (Input gate), o t (output gate) and f t (Forget gate), and use the tanh activation function to generate a new memory C′ t , the model output includes the current output variable h t and the current cell state variable C t, whose expression is:

[0036] C t =f t C t-1 +i t C′ t

[0037] h t =o t tanh(C t )

[0038] Assume that the input x of the model is x1, x2, x3, ..., x n , and x∈R T , i = 1, 2, ..., n; n is the input dimension, T represents the time step, the top oil temperature is the output, and the output sequence is y = y1, y2, ..., y n , and substitute them into the LSTM model expression respectively to complete the construction of the LSTM model for this scenario.

[0039] As for the loss function in model training, in one possible implementation, two loss functions are used to make the prediction results more accurate and expand the differences in the relationship between them.

[0040] The loss function is obtained using the following formula:

[0041] loss=max(0,-y*(S(r;q)-S(r';q))+margin)

[0042] loss2=y*|cos(r,r')|*max(0,cos(r,r)-γ).

[0043] After determining that the predicted temperature value is greater than the temperature threshold range associated with the current distribution transformer fault strategy, the distribution transformer continues to be overloaded based on the current distribution transformer management strategy until the predicted temperature value is within the temperature threshold range.

[0044] The transformer temperature and fault prediction method based on LSTM technology provided by the present invention includes: continuous data collection of temperature sensors at monitoring points in distribution substations, and by analyzing the temperature change data related to time, creating a transformer temperature prediction model of LSTM with time series as correlation for data analysis and prediction: the transformer temperature prediction model of LSTM includes: input layer, hidden layer, output layer and model training, wherein the input layer is used to read and preprocess the data set, and the division of the training test data set; the hidden layer is used to build the LSTM hidden layer and the fully connected neural layer; the output layer is used as the verification processing of the prediction result; the model training is responsible for the loss function calculation, the Adam algorithm optimization and the iterative calculation of the model training. The method makes full use of its physical model, the update of sensor data, the operation history and other data, integrates multi-disciplinary, multi-physical quantity, multi-scale, multi-probability simulation processes, completes the spatial mapping in the computer system, establishes the distribution substation information mirror model, and realizes the real-time monitoring of the operation status of the transformer in the distribution substation through the information mirror model. The method is simple and the prediction result has high accuracy.

[0045] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A transformer temperature and fault prediction method based on LSTM technology, characterized in that the method include: Continuously collect data from the temperature sensors at the monitoring points in the distribution substation area, analyze the time-related temperature change data, and create a transformer temperature prediction model with LSTM based on time series correlation for data analysis and prediction: The transformer temperature prediction model of the LSTM includes: an input layer, a hidden layer, an output layer and model training, wherein the input layer is used to read and preprocess the data set, and to divide the training and test data set; the hidden layer is used to build the LSTM hidden layer and the fully connected neural layer; the output layer is used as a verification process for the prediction results; and the model training is responsible for the loss function calculation, the Adam algorithm optimization and the iterative calculation of the model training.

2. According to a transformer temperature and fault prediction method based on LSTM technology according to claim 1, it is characterized in that: The LSTM transformer temperature prediction model generation method includes: Get temperature time series sample data; Preprocessing the temperature time series sample data to obtain training data samples and test data samples; The training data samples are input into the LSTM long short-term memory network for training to obtain the transformer temperature prediction model of the LSTM.

3. According to a transformer temperature and fault prediction method based on LSTM technology according to claim 2, it is characterized in that: Preprocessing the temperature time series sample data includes homogenizing the data.

4. According to a transformer temperature and fault prediction method based on LSTM technology according to claim 2, it is characterized in that: At time t, the transformer temperature prediction model input of the LSTM includes the current input variable x t , the hidden layer state variable h at the previous moment t-1 and the cell state variable C at the previous moment t-1 , passing through i t (Input gate), o t (output gate) and f t (Forget gate), and use the tanh activation function to generate a new memory C′ t , the model output includes the current output variable h t and the current cell state variable C t , whose expression is: C t =f t C t-1 +i t C′ t h t =o t fishy(C) t ) Assume that the input x of the model is x1, x2, x3, ..., x n , and x∈R T , i = 1, 2, ..., n; n is the input dimension, T represents the time step, the top oil temperature is the output, and the output sequence is y = y1, y2, ..., y n , and substitute them into the LSTM model expression respectively to complete the construction of the LSTM model for this scenario.

5. The transformer temperature and fault prediction method based on LSTM technology according to claim 1 is characterized in that: Two loss functions are used to make the prediction results more accurate and expand the differences between the relationships.

6. The transformer temperature and fault prediction method based on LSTM technology according to claim 5 is characterized in that: The loss function is obtained using the following formula: loss=max(0,-y*(S(r;q)-S(r';q))+margin) loss2=y*|cos(r,r')|*max(0,cos(r,r)-γ).

7. The transformer temperature and fault prediction method based on LSTM technology according to claim 1 is characterized in that: The method comprises: After determining that the predicted temperature value is greater than the temperature threshold range associated with the current distribution transformer fault strategy, the distribution transformer continues to be overloaded based on the current distribution transformer management strategy until the predicted temperature value is within the temperature threshold range.

Images