Power transmission channel hidden danger occurrence trend prediction analysis method and device and computer readable storage medium
By combining the two-way LSTM model, adaptive parameter random forest model and Holt-Winters method prediction methods, the problems of low patrol efficiency, lagging alarm response and unreasonable resource allocation in the monitoring of hidden dangers of transmission channels are solved, and more accurate hidden dangers are achieved Forecasting and resource allocation have improved the operation and maintenance management effect of transmission channels.
Patent Information
- Application Number
- CN202411201330.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-11-29
AI Technical Summary
The existing hidden danger monitoring methods for power transmission channels have problems such as low patrol efficiency, lagging alarm response, unreasonable resource allocation and insufficient use of alarm data, resulting in high patrol pressure, insufficient prediction accuracy and poor operation and maintenance results.
A prediction method combining the bidirectional LSTM model, adaptive parameter random forest model and Holt-Winters method is adopted to deeply learn and pattern recognition of historical alarm data, improve prediction accuracy through feature engineering and attention mechanism, and optimize model parameters To improve the accuracy of patrol resource allocation.
It significantly improves the accuracy of hidden danger prediction, optimizes the allocation of inspection resources, improves the decision-making support for inspection efficiency and operation and maintenance management, and reduces the pressure on inspection.
Smart Images

Figure CN119046882A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power transmission channel data processing, and more specifically, relates to a method, device and computer-readable storage medium for predicting and analyzing the trend of hidden dangers in power transmission channels. Background Art
[0002] In a region, hidden dangers in power transmission channels may occur repeatedly or multiple times within a certain period of time. Based on historical alarm information, it is possible to predict the hidden dangers that may occur at a certain time in the future, determine the areas where equipment that needs to be inspected in the future is located, and reduce the inspection pressure. Existing hidden danger monitoring and response strategies for power transmission channels are often based on fixed-cycle inspections and basic statistical analysis of historical alarm data.
[0003] Chinese invention patent CN117522147A discloses a method and system for assessing power transmission safety risks in a complex power grid environment, including obtaining the electrical signal operation data of the transmission line in each historical period, the diagnostic data of power grid fault events and the corresponding historical environmental data, and constructing a transmission line safety hazard prediction model; obtaining environmental forecast data within a preset time period in the future, and obtaining real-time predictions of transmission line safety hazards through the prediction model; combining the obtained real-time predictions of transmission line safety hazards with preset warning thresholds, and making a risk assessment of power transmission safety within the preset time period of the transmission line; the system includes: a data acquisition module, a future environment acquisition module, a transmission line safety hazard prediction model, a risk assessment module and an output module.
[0004] However, this approach has the following significant drawbacks:
[0005] Low inspection efficiency: The same inspection intensity is adopted in all areas of the current region, resulting in a large amount of manpower and material resources being invested in relatively low-risk areas, while high-risk areas may not receive sufficient attention due to the dispersion of resources, resulting in low inspection efficiency.
[0006] Delayed alarm response: Relying on manual or conventional monitoring systems to identify and report hidden dangers, there may be delays, omissions or misjudgments in the transmission of alarm information, affecting the timely discovery and effective handling of hidden dangers.
[0007] Unreasonable resource scheduling: The lack of accurate prediction of the probability and severity of hidden dangers results in the inability to accurately match the risk level of hidden dangers in the allocation of inspection resources. This is especially true in areas with frequent construction, complex environments, or high incidence of seasonal hidden dangers. This may cause an uneven distribution of inspection forces and affect the overall operation and maintenance effect.
[0008] Ignoring in-depth mining of alarm data: The use of historical alarm information remains at the surface statistical level, and the implicit patterns and correlations in the data are not fully utilized, which makes it impossible to reveal the interactions between hidden dangers and potential risk chains. Summary of the invention
[0009] The present invention aims to overcome at least one of the defects of the above-mentioned prior art and provide a method for predicting and analyzing the trend of hidden dangers in power transmission channels, so as to solve the defects of the prior art such as low inspection efficiency, delayed alarm response, insufficient prediction accuracy, unreasonable resource allocation and insufficient utilization of alarm data, and provide a more intelligent, accurate and resource-efficient method for predicting and analyzing the trend of hidden dangers in power transmission channels.
[0010] The detailed technical scheme of the present invention is as follows:
[0011] A method for predicting and analyzing the trend of hidden dangers in power transmission channels, the method comprising:
[0012] S1. Collect all historical alarm information data of the equipment, and then clean the data, such as removing invalid records, processing missing values, and unifying data formats;
[0013] The historical alarm information data includes but is not limited to alarm type, occurrence date, and device number;
[0014] S2. Integrate the data according to the equipment number and date of occurrence, and conduct a secondary cleanup of potential hazards such as fire and foreign matter in wires;
[0015] S3, the device number, time series, and the presence or absence of an alarm are combined into a one-dimensional feature sequence representing all the information of the current device through feature engineering;
[0016] S4. Build a prediction model, including a bidirectional LSTM model with attention mechanism for parallel prediction, an adaptive parameter random forest model, and the Holt-Winters method; finally, combine the prediction results of the three to obtain the final prediction result;
[0017] S5. Model validation and optimization: Evaluate model performance through cross-validation method, and adjust model parameters to improve prediction accuracy and optimize model performance;
[0018] S6. For the future date to be predicted, input the equipment number and the time series characteristics of the corresponding date into the prediction model, output the probability of hidden dangers occurring in the equipment on that day, and set the inspection intensity and frequency based on the historical alarm trend.
[0019] Furthermore, smoke alarms may contain a mixture of real alarms and false alarms, and S2 may also include secondary cleaning for potential hazards such as fire and foreign matter in wires;
[0020] If the same device has 5 or more fire and wire foreign object alarms for 3 consecutive days, it will be marked as a suspected false alarm and removed or downgraded.
[0021] If a fire occurs at the same location for three consecutive days, and if there is a continuous alarm for 15 days forward and backward, it will be considered a false alarm and will be eliminated; if it only exists for a certain period of time within 15 days forward and backward, it will be considered a non-dangerous fire and will be downgraded;
[0022] If a foreign object appears in the wire at the same location for three consecutive days, and if the alarm still exists within 3 days forward and backward, it will be regarded as a false alarm and will be eliminated; if the alarm does not exist within 3 days forward and backward, it will be downgraded.
[0023] Furthermore, the weight reduction operation is to reduce the weight during model training and adjust the weight coefficient in the loss function:
[0024]
[0025] In formula (1), w i represents the weight of the i-th sample, and α (0<α<1) is the weight reduction coefficient.
[0026] Furthermore, the S3 includes: encoding the device number and converting it into a numerical variable;
[0027] Construct time series features from time series so that the model can capture time-related regularities;
[0028] Create binary features to indicate whether an alarm occurs on that day and whether a specific type of alarm occurs.
[0029] Through feature engineering, numerical variables, time series features, and binary features are combined into a one-dimensional feature sequence that represents all the information of the current device.
[0030] Furthermore, the S4 specifically includes:
[0031] S41, the bidirectional LSTM model also includes a pre-processing module, which extracts sliding window statistical features and periodic features from the one-dimensional feature sequence as an input sequence of the bidirectional LSTM model;
[0032] Sliding window statistical features: set a sliding window of fixed length on the time series and extract statistical features within the fixed-length window;
[0033] Periodic characteristics: Use Fourier transform method to analyze the spectrum of data, identify the period, and extract the periodic pattern in the data. The periodic pattern is the daily, weekly, monthly, and annual alarm re-reading pattern. The alarm re-reading pattern is the alarm generated by periodic hidden dangers;
[0034] S42. Use the bidirectional LSTM model to enhance the prediction ability of time series. The bidirectional LSTM model captures information in the time series from both the forward and reverse directions, and can better understand the contextual relationship.
[0035] The output of the bidirectional LSTM is the concatenation of the forward and backward hidden states h t :
[0036]
[0037] In formula (2), h t is the output of the bidirectional LSTM, is the forward hidden state at the tth time step, is the backward hidden state at the t-th time step, and the forward and backward hidden states are concatenated by merging in dimension;
[0038] S43. The bidirectional LSTM model combines the attention mechanism to calculate the importance of each time step, so that the bidirectional LSTM focuses on predicting more important time points, which helps to improve the model's ability to capture important features:
[0039] Attention score:
[0040]
[0041] In formula (3), e t represents the attention score at the current time step t, v a , W, b a is a learnable parameter, T represents transposition,
[0042] Attention weights:
[0043]
[0044] In formula (4), α t represents the attention weight of the current time step t;
[0045] Context vector:
[0046]
[0047] In formula (5), c is the context vector, and the attention-weighted context vector c is passed as input to the subsequent prediction layer;
[0048] Generate the predicted output y of the bidirectional LSTM model:
[0049] y=W y ·c+b y (6);
[0050] In formula (6), y is the predicted output of the bidirectional LSTM model, W y and b y are learnable weights and biases;
[0051] The bidirectional LSTM model tends to be comprehensive in prediction, including analysis and prediction of routine events, emergencies, and time series;
[0052] S44. Use the adaptive parameter random forest model and the Holt-Winters method for prediction:
[0053] Use random forest models for predictions and adopt dynamic hyperparameter adjustment strategies, using grid search or Bayesian optimization to automatically find the best parameter combination to improve the prediction accuracy of random forest models. Adaptive parameter random forests tend to be sudden and have no periodic changes;
[0054] The Holt-Winters method is used to process time series data with trend and seasonal components to obtain forecast results. The Holt-Winters method tends to favor long-term seasonal changes.
[0055] S45, weighted fusion of the LSTM prediction results, the random forest model prediction results, and the prediction results of the Holt-Winters method to capture seasonal changes to obtain the final prediction results;
[0056] Final Prediction=ω1Random Forest Prediction+ω2LSTM Prediction+ω3Holt-Winters Prediction (7);
[0057] In formula (7), Final Prediction is the final output, ω1, ω2, ω3 are the weights of the random forest model, LSTM model and Holt-Winters method, and ω1+ω2+ω3=1.
[0058] Furthermore, the adjustment model parameters include but are not limited to:
[0059] If the model performs poorly on the validation set, increase the number of LSTM units to improve the model's expressiveness or adjust the learning rate;
[0060] For model overfitting, increase the regularization term or reduce the complexity of the model;
[0061] For underfitting models, increase the number of trees to improve model stability or adjust the maximum depth of the tree to control model complexity;
[0062] If the model has poor prediction accuracy for any category, adjust the number of attention heads to change the model's ability to capture information from different aspects or change the model initialization method.
[0063] Furthermore, the S6 specifically includes: for a future date to be predicted, inputting a device number and a time series feature of a corresponding date into a prediction model to obtain a prediction result;
[0064] Secondly, all historical alarm data of the equipment are plotted as a line graph to show the trend of hidden dangers and analyze the change of the probability of hidden dangers over time;
[0065] Based on the prediction results, hidden danger trend chart and the actual situation of the area where the equipment is located, determine the intensity and frequency of inspections of the equipment and surrounding areas.
[0066] In another aspect of the present invention, there is also provided a device for predicting and analyzing the trend of hidden dangers in a power transmission channel, the device comprising:
[0067] processor;
[0068] a memory having stored thereon a computer program executable on the processor;
[0069] Wherein, when the computer program is executed by the processor, the steps of the method for predicting and analyzing the trend of hidden dangers in power transmission channels are implemented.
[0070] In another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a method for predicting and analyzing the trend of hidden dangers in power transmission channels is implemented.
[0071] Compared with the prior art, the present invention has the following beneficial effects:
[0072] (1) The present invention provides a method, device and computer-readable storage medium for predicting and analyzing the trend of hidden dangers in power transmission channels, which improves the accuracy of hidden danger prediction: by using multiple machine learning models and methods to perform deep learning and pattern recognition on historical alarm data, it can accurately predict the probability of occurrence of different types of hidden dangers in future power transmission channels. Compared with traditional statistical analysis methods, it significantly improves the accuracy of prediction and provides a reliable basis for preventive maintenance and emergency response.
[0073] (2) The present invention provides a method, device and computer-readable storage medium for predicting and analyzing the trend of hidden dangers in power transmission channels, which optimize the allocation of inspection resources: based on the prediction results, it is possible to achieve differentiated assessment of the hidden danger risks of various areas and equipment, guide the formulation of targeted inspection plans, concentrate limited inspection forces on high-risk areas, avoid waste of resources, effectively reduce inspection pressure, and improve inspection efficiency.
[0074] (3) The present invention provides a method, device and computer-readable storage medium for predicting and analyzing the trend of hidden dangers in power transmission channels. Through comprehensive analysis of multi-dimensional data such as hidden danger types, occurrence frequencies, and spatiotemporal distribution, the method provides detailed data support for the operation and maintenance management of power transmission channels, helps decision makers to accurately identify key risk factors and formulate more reasonable operation and maintenance strategies and emergency plans. BRIEF DESCRIPTION OF THE DRAWINGS
[0075] Figure 1 It is a flow chart of a method for predicting and analyzing the trend of hidden dangers in power transmission channels described in the present invention.
[0076] Figure 2 It is a schematic diagram of the prediction model structure described in Example 1 of the present invention.
[0077] Figure 3 This is Example 1 of the trend chart analysis of hidden dangers in power transmission channels in Example 1 of the present invention.
[0078] Figure 4 This is Example 2 of the trend chart analysis of hidden dangers in power transmission channels in Example 1 of the present invention. DETAILED DESCRIPTION
[0079] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0080] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.
[0081] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0082] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.
[0083] Example 1
[0084] Ginseng Figure 1 This embodiment provides a method for predicting and analyzing the trend of hidden dangers in power transmission channels, the method comprising:
[0085] S1. Data preparation: Collect all historical alarm information data of the equipment, i.e., the transmission channel visualization equipment, including but not limited to the alarm type, occurrence date, and equipment number; then, clean the data, such as removing invalid records, processing missing values, and unifying the data format;
[0086] The alarm types include but are not limited to: tower cranes, cranes, smoke, fire, excavators, foreign objects in wires, cement mixers, cement pump trucks, pile drivers, forklifts, and bulldozers.
[0087] S2. Data preprocessing: Integrate the data according to the equipment number and date of occurrence, and perform secondary cleaning of potential hazards such as fire and foreign matter in wires;
[0088] Specifically, smoke alarms have a mixture of real alarms and false alarms, so only fire and foreign matter hazards are cleaned up twice: if there are 5 or more smoke alarms on the same device for 3 consecutive days, they are marked as suspected false alarms and handled appropriately (such as removal or downgrading);
[0089] If a fire occurs at the same location for three consecutive days, it may be a false alarm caused by car lights, street lights, reflections, etc., or it may be a non-dangerous fire that is under human supervision, such as the burning of Platycodon grandiflorum. Therefore, for such hidden dangers, if there is a continuous alarm within 15 days forward and backward, it will be regarded as a false alarm and eliminated; if it only exists for a certain period of time within 15 days forward and backward, it will be regarded as a non-dangerous fire and will be downgraded.
[0090] If foreign objects appear on the wires at the same location for three consecutive days, it may be a false alarm caused by the hardware and firmware on the wires. Therefore, for such hidden dangers, if the alarm still exists within 3 days forward and backward, it will be regarded as a false alarm caused by the hardware and firmware on the wires and will be eliminated; if the alarm does not exist within 3 days forward and backward, it will be downgraded.
[0091] Preferably, for data marked as suspicious false positives, the weight is reduced during model training, and the specific implementation method is to adjust the weight coefficient in the loss function:
[0092]
[0093] In formula (1), w i represents the weight of the i-th sample, and α (0<α<1) is the weight reduction coefficient.
[0094] S3, feature engineering: The device number, time series, and alarm status are combined into a one-dimensional feature sequence representing all the information of the current device through feature engineering;
[0095] In this field, the device number can be a character variable with English, a numeric variable, or a mixed variable. Preferably, the device number is encoded and converted into a numeric variable, for example, the device number 11M220330PVSH0001 is encoded as 00000001;
[0096] Construct time series features such as date, day of the week, month, quarter, year, etc., so that the model can capture time-related patterns;
[0097] Create a binary feature to indicate whether an alarm occurs on that day and whether a specific type of alarm occurs. 0 indicates no such alarm, and 1 indicates such alarm occurs.
[0098] Through feature engineering, numerical variables, time series features, and binary features are combined into a one-dimensional feature sequence that represents all the information of the current device. For example, the feature vector of a certain device on a certain day is: (device code, device information, year, quarter, month, date, whether there is a fireworks alarm, whether there is a mechanical hazard) (device code, device information, year, quarter, month, date, whether there is a fireworks alarm, whether there is a mechanical hazard).
[0099] For example, the one-dimensional feature sequence of a certain device on a certain day includes device code, device information, year, quarter, month, date, whether there is a smoke alarm, whether there is a fire alarm, whether there is a wire foreign body alarm, and whether there is a mechanical hidden danger. Specifically: (00000001, 1, 436, 538, 2023, 1, 3, 20, 1, 1, 1, 1) = (encoded device number, whether the device is operating normally (1 indicates normal operation, 0 indicates that the device is not operating normally), longitude, latitude, year, quarter, month, date, whether there is a smoke alarm, whether there is a fire alarm, whether there is a wire foreign body alarm, and whether there is a mechanical hidden danger).
[0100] S4. Build a prediction model, such as Figure 2 As shown, it includes a bidirectional LSTM model with attention mechanism for parallel prediction, an adaptive parameter random forest model and the Holt-Winters method. Finally, the prediction results of the three are fused to obtain the final prediction result.
[0101] Preferably, the S4 specifically includes:
[0102] S41, the bidirectional LSTM model also includes a pre-processing module, which extracts sliding window statistical features and periodic features from the one-dimensional feature sequence as an input sequence of the bidirectional LSTM model;
[0103] Sliding window statistical features: A sliding window of fixed length is set on the time series. In this embodiment, the fixed length is set to 1 week, and the statistical features (mean, standard deviation, maximum value, minimum value, and quartiles) within the fixed-length window are extracted.
[0104] Periodic characteristics: Use the Fourier transform method to analyze the spectrum of the data, identify the period, and extract the periodic pattern in the data. The periodic pattern in this embodiment is a daily, weekly, monthly, and yearly alarm re-reading pattern, and the alarm re-reading pattern is an alarm generated by periodic hidden dangers.
[0105] S42. Use the bidirectional LSTM model to enhance the prediction ability of time series. The bidirectional LSTM model captures information in the time series from both the forward and reverse directions, and can better understand the contextual relationship.
[0106] The output of the bidirectional LSTM is the concatenation of the forward and backward hidden states h t :
[0107]
[0108] In formula (1), h t is the output of the bidirectional LSTM, is the forward hidden state at the tth time step, is the backward hidden state at the t-th time step, and the forward and backward hidden states are concatenated by merging in dimension.
[0109] S43. The bidirectional LSTM model combines the attention mechanism to calculate the importance of each time step, so that the bidirectional LSTM focuses on predicting more important time points, which helps to improve the model's ability to capture important features:
[0110] Attention score:
[0111]
[0112] In formula (3), e t represents the attention score at the current time step t, v a , W, b a is a learnable parameter, T represents transposition,
[0113] Attention weights:
[0114]
[0115] In formula (4), α t represents the attention weight of the current time step t;
[0116] Context vector:
[0117]
[0118] In formula (5), c is the context vector, and the attention-weighted context vector c is passed as input to the subsequent prediction layer;
[0119] Generate the predicted output y of the bidirectional LSTM model:
[0120] y=W y ·c+b y (6);
[0121] In formula (6), y is the predicted output of the bidirectional LSTM model, W y and b y are learnable weights and biases;
[0122] The bidirectional LSTM model tends to be comprehensive in prediction, including analysis and prediction of regular events, emergencies, and time series.
[0123] S44. Use the adaptive parameter random forest model and the Holt-Winters method for prediction:
[0124] Use random forest models for predictions and adopt dynamic hyperparameter adjustment strategies, using grid search or Bayesian optimization to automatically find the best parameter combination to improve the prediction accuracy of random forest models. Adaptive parameter random forests tend to be sudden and have no periodic changes;
[0125] The Holt-Winters method is used to process time series data with trend and seasonal components to obtain forecast results. The Holt-Winters method tends to favor long-term seasonal changes.
[0126] S45, model integration: The LSTM prediction results, the random forest model prediction results, and the prediction results of the Holt-Winters method to capture seasonal changes are weighted and integrated to obtain the final prediction results:
[0127] Final Prediction=ω1Random Forest Prediction+ω2LSTM Prediction+ω3Holt-Winters Prediction (7);
[0128] In formula (7), Final Prediction is the final output, ω1, ω2, ω3 are the weights of the random forest model, LSTM model and Holt-Winters method, and ω1+ω2+ω3=1.
[0129] S5. Model validation and optimization: Evaluate model performance through cross-validation method and adjust model parameters to improve prediction accuracy.
[0130] The model performance was optimized by adjusting parameters such as the number of LSTM units, learning rate, regularization term, number of trees, maximum depth of trees, and number of attention heads.
[0131] Preferably, the adjustment model parameters include but are not limited to:
[0132] When the performance of model evaluation is poor, increase the number of LSTM units to improve the expressiveness of the model or adjust the learning rate;
[0133] For model overfitting, increase the regularization term or reduce the complexity of the model;
[0134] For underfitting models, increase the number of trees to improve model stability or adjust the maximum depth of the tree to control model complexity;
[0135] If the model has poor prediction accuracy for any category, adjust the number of attention heads to change the model's ability to capture information from different aspects or change the model (prediction model) initialization method.
[0136] The initialization methods here include but are not limited to adjustments of Xavier / Glorot initialization, He initialization, uniform initialization and other methods.
[0137] S6. Predict hidden dangers and set inspection strategies: For the future date to be predicted, input the equipment number and the time series characteristics of the corresponding date into the prediction model, output the probability of hidden dangers occurring on the equipment on that day, and set the inspection intensity and frequency based on the historical alarm occurrence trend;
[0138] Preferably, for a future date to be predicted, the device number and the time series characteristics of the corresponding date are input into the prediction model to obtain a prediction result;
[0139] Secondly, all historical alarm data of the equipment are plotted as a line graph to show the trend of hidden dangers and analyze the change of the probability of hidden dangers over time, such as Figure 3 , Figure 4 As shown;
[0140] Based on the prediction results, hidden danger trend chart and the actual situation of the area where the equipment is located, determine the inspection intensity and frequency of the equipment and surrounding areas. Specifically, the inspection intensity and frequency can be set independently according to different environments and equipment.
[0141] In summary, a method for predicting and analyzing the trend of hidden dangers in power transmission channels of the present invention collects and processes historical alarm data, and uses a bidirectional LSTM model combined with an attention mechanism, an adaptive parameter random forest model, and the Holt-Winters method for integrated modeling. It can accurately predict the trend of hidden dangers in power transmission channels, and analyze the hidden danger trends in combination with the prediction results, so as to formulate more reasonable inspection strategies, effectively improve the hidden danger prediction accuracy and inspection efficiency, and solve the problems of low inspection efficiency, delayed alarm response, insufficient prediction accuracy, unreasonable resource allocation, and insufficient utilization of alarm data in the prior art.
[0142] Example 2
[0143] This embodiment provides a device for implementing a method for predicting and analyzing the trend of hidden dangers in a power transmission channel, the device comprising:
[0144] at least one processor; and
[0145] A memory storing instructions, which, when executed by the at least one processor, causes the at least one processor to execute the method for predicting and analyzing the trend of hidden dangers in a power transmission channel as described above.
[0146] In this embodiment, the device includes but is not limited to: personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile computing devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDAs), handheld devices, messaging devices, wearable computing devices, consumer electronic devices, etc.
[0147] Example 3
[0148] This embodiment also provides a computer-readable storage medium storing executable instructions, which, when executed, enable the machine to execute the above-mentioned method for predicting and analyzing the trend of hidden dangers in power transmission channels.
[0149] Specifically, a system or device equipped with a readable storage medium can be provided, on which software program codes that implement the functions of any of the above-mentioned embodiments are stored, and a computer or processor of the system or device can read and execute instructions stored in the readable storage medium.
[0150] In this case, the program code itself read from the machine-readable medium can realize the function of any one of the above embodiments, and thus the machine-readable code and the machine-readable storage medium storing the machine-readable code constitute part of this specification.
[0151] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or a cloud via a communication network.
[0152] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0153] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0154] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0156] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation methods of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the claims of the present invention shall be included in the protection scope of the claims of the present invention.
Claims
1. A method for predicting and analyzing the trend of hidden dangers in power transmission channels, characterized in that: The method comprises: S1. Collect all historical alarm information data of the device and then clean the data. The historical alarm information data includes but is not limited to alarm type, occurrence date, and device number; S2. Integrate the data according to the equipment number and date of occurrence, and conduct a secondary cleanup of potential hazards such as fire and foreign matter in wires; S3, the device number, time series, and the presence or absence of an alarm are combined into a one-dimensional feature sequence representing all the information of the current device through feature engineering; S4. Build a prediction model, including a bidirectional LSTM model with attention mechanism for parallel prediction, an adaptive parameter random forest model, and the Holt-Winters method; finally, combine the prediction results of the three to obtain the final prediction result; S5. Evaluate the performance of the prediction model through cross-validation method and adjust the model parameters; S6. For the future date to be predicted, input the equipment number and the time series characteristics of the corresponding date into the prediction model, output the probability of hidden dangers occurring in the equipment on that day, and set the inspection intensity and frequency based on the historical alarm trend.
2. A method for predicting and analyzing the trend of hidden dangers in power transmission channels according to claim 1, characterized in that: The secondary cleaning of fire and foreign matter hazards of wires specifically includes: If the same device has 5 or more fire and wire foreign object alarms for 3 consecutive days, it will be marked as a suspected false alarm and the following processing will be performed: If a fire occurs at the same location for three consecutive days, and if there is a continuous alarm for 15 days forward and backward, it will be considered a false alarm and will be eliminated; if it only exists for a certain period of time within 15 days forward and backward, it will be considered a non-dangerous fire and will be downgraded; If foreign objects appear in the wires at the same location for three consecutive days, and if the alarm still exists within 3 days forward and backward, it will be regarded as a false alarm caused by the hardware and firmware on the wires and will be eliminated; if the alarm does not exist within 3 days forward and backward, it will be downgraded.
3. A method for predicting and analyzing the trend of hidden dangers in power transmission channels according to claim 2, characterized in that: The weight reduction operation is to reduce the weight during model training and adjust the weight coefficient in the loss function: In formula (1), w i represents the weight of the i-th sample, and α (0<α<1) is the weight reduction coefficient.
4. A method for predicting and analyzing the trend of hidden dangers in power transmission channels according to claim 1 or 2, characterized in that: The S3 includes: encoding the device number and converting it into a numerical variable; constructing a time series feature from the time series; creating a binary feature to indicate whether an alarm occurs on that day and whether a specific type of alarm occurs; Through feature engineering, numerical variables, time series features, and binary features are combined into a one-dimensional feature sequence that represents all the information of the current device.
5. A method for predicting and analyzing the trend of hidden dangers in power transmission channels according to claim 1, characterized in that: The S4 specifically includes: S41, the bidirectional LSTM model also includes a pre-processing module, which extracts sliding window statistical features and periodic features from the one-dimensional feature sequence as an input sequence of the bidirectional LSTM model; Sliding window statistical features: set a sliding window of fixed length on the time series and extract statistical features within the fixed-length window; Periodic characteristics: Use Fourier transform method to analyze the spectrum of data, identify the period, and extract the periodic pattern in the data. The periodic pattern is the daily, weekly, monthly, and annual alarm re-reading pattern. The alarm re-reading pattern is the alarm generated by periodic hidden dangers; S42. Use the bidirectional LSTM model to enhance the prediction ability of time series. The bidirectional LSTM model captures information in the time series from both the forward and reverse directions, and can better understand the contextual relationship. The output of the bidirectional LSTM is the concatenation of the forward and backward hidden states h t : In formula (2), h t is the output of the bidirectional LSTM, is the forward hidden state at the tth time step, is the backward hidden state at the t-th time step, and the forward and backward hidden states are concatenated by merging in dimension; S43, the bidirectional LSTM model combines the attention mechanism to calculate the importance of each time step so that the bidirectional LSTM focuses on predicting more important time points: Attention score: In formula (3), e t represents the attention score at the current time step t, v a , W, b a is a learnable parameter, T represents transposition; Attention weights: In formula (4), α t represents the attention weight of the current time step t; Context vector: In formula (5), c is the context vector, and the attention-weighted context vector c is passed as input to the subsequent prediction layer; Generate the predicted output y of the bidirectional LSTM model: y=W y ·c+b y (6); In formula (6), y is the predicted output of the bidirectional LSTM model, W y and b y are learnable weights and biases; S44, using both the adaptive parameter random forest model and the Holt-Winters method for prediction; S45, weighted fusion of the LSTM prediction results, the random forest model prediction results, and the prediction results of the Holt-Winters method to capture seasonal changes to obtain the final prediction results; Final Prediction=ω1Random Forest Prediction+ω2LSTM Prediction+ω3Holt-Winters Prediction (7); In formula (7), Final Prediction is the final output, ω1, ω2, ω3 are the weights of the random forest model, LSTM model and Holt-Winters method, and ω1+ω2+ω3=1.
6. A method for predicting and analyzing the trend of hidden dangers in power transmission channels according to claim 1 or 5, characterized in that: The adjustment model parameters include but are not limited to: When the performance of model evaluation is poor, increase the number of LSTM units to improve the expressiveness of the model or adjust the learning rate; For model overfitting, increase the regularization term or reduce the complexity of the model; For underfitting models, increase the number of trees to improve model stability or adjust the maximum depth of the tree to control model complexity; If the model has poor prediction accuracy for any category, adjust the number of attention heads to change the model's ability to capture information from different aspects or change the model initialization method.
7. A device for predicting and analyzing the trend of hidden dangers in power transmission channels, characterized in that: The device comprises: processor; a memory having stored thereon a computer program executable on the processor; Wherein, when the computer program is executed by the processor, the steps of the method for predicting and analyzing the trend of hidden dangers in power transmission channels as claimed in any one of claims 1 to 6 are implemented.
8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
Citation Information
Patent Citations
Power grid cascading fault diagnosis, early warning and evaluation system based on monitoring information
CN108830745A
Bus voltage situation short-term prediction method and system for power grids
CN108964023A
Abnormal discharge identification method based on bidirectional long and short term memory and attention mechanism
CN113589100A
Power transmission safety risk assessment method and system in power grid complex environment
CN117522147A
Application system fault prediction method based on association rule and deep learning integration model
CN117667495A
Cited By
Traffic flow prediction method, device, equipment and product
CN121811640A
