A method and system for monitoring the status of equipment in a power distribution station based on the Internet of Things

Abstract

This invention discloses an IoT-based method and system for detecting the status of equipment in a power distribution substation, relating to the field of power distribution substation equipment detection technology. The method includes the following steps: collecting environmental monitoring data and equipment status monitoring data of the power distribution substation, and combining them to generate real-time monitoring data; based on the real-time monitoring data and equipment normal status monitoring data, analyzing whether the status of the power distribution substation equipment corresponding to the real-time monitoring data is normal, and if so, generating predicted equipment normal status monitoring data; this IoT-based method and system for detecting the status of equipment in a power distribution substation can determine whether the equipment is normal by comparing the characteristics of the time series of real-time monitoring data over a period of time, and can predict whether there are any abnormalities in the equipment within a period of time, thereby reminding staff before the equipment malfunctions.

Description

Technical Field

[0001] This invention relates to the field of power distribution room equipment testing technology, specifically to a method and system for detecting the status of power distribution room equipment based on the Internet of Things. Background Technology

[0002] Traditional power distribution substations suffer from incomplete environmental monitoring equipment and low levels of construction, renovation, and intelligence. With the advancement of power distribution datafication, more and more substations are adding various environmental and equipment monitoring sensors to achieve more comprehensive monitoring of equipment and analyze its operational status through threshold judgment and data analysis. Existing technology, with publication number CN118365315A, discloses a method for fault operation and maintenance management of power distribution substations based on multi-source data analysis. This method involves monitoring multi-source data from various devices and calculating the Z-value for each device. Data with an absolute Z-value greater than a preset threshold is marked as anomalies. After obtaining the number of anomalies for each device, weights are assigned to all devices based on their correlation. The overall operating index of the power distribution substation is obtained by weighted calculation of the number of anomalies from all abnormal devices. The comparison between the overall operating index and the index threshold predicts whether future faults will occur in the substation, and corresponding control strategies are generated based on the prediction results. This management method can combine multi-source data from multiple devices to determine the abnormal status of multiple devices, and then perform an overall analysis to determine whether there is a possibility of failure in the power distribution room. This allows for early warning before power distribution room failures occur, ensuring the stability and safety of power distribution room operation.

[0003] However, relying solely on weight allocation and threshold analysis can easily lead to false alarms or missed alarms due to equipment aging or different operating conditions. Furthermore, threshold-based judgments only trigger alarms when relevant monitoring data or parameters exceed the threshold, failing to provide early warnings and resulting in a certain lag in alarms. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for monitoring the status of equipment in a power distribution station based on the Internet of Things (IoT), in order to address the aforementioned shortcomings in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting the status of equipment in a power distribution station based on the Internet of Things, comprising the following steps:

[0006] S1. Collect environmental monitoring data and equipment status monitoring data of the power distribution room, and combine them to generate real-time monitoring data;

[0007] S2. Based on the real-time monitoring data and the equipment normal status monitoring data, analyze whether the status of the substation equipment corresponding to the real-time monitoring data is normal. If so, generate predicted equipment normal status monitoring data.

[0008] S3. If abnormal, the real-time monitoring data is marked as abnormal monitoring data. The abnormal monitoring feature standardized vector obtained by feature extraction and standardization of the abnormal monitoring data is input into the pre-trained abnormal type prediction model to generate the predicted abnormal type label and the corresponding prediction label confidence data.

[0009] S4. Based on the predicted label confidence data, determine whether the confidence level of the predicted anomaly category label is less than the set confidence level standard. If not, output the predicted anomaly type label. If yes, output the predicted anomaly type label and the corresponding predicted label confidence data at the same time, and issue a confidence level alarm. The confidence level alarm can be implemented by sending an alarm pop-up, sending emails to managers, SMS messages, etc.

[0010] There can be multiple confidence levels, and each confidence level corresponds to a confidence standard. The confidence standard can be a threshold for the output probability of the anomaly category label, a threshold for entropy or binary entropy, a variance threshold for the average probability of the output of each anomaly category label after multiple forward propagations, etc.

[0011] For example, the output layer of the anomaly type prediction model is a softmax function. The confidence level for high confidence is that the threshold of the anomaly category label output probability is greater than 0.9 or the entropy is less than 0.5. The confidence level for medium confidence is that the threshold of the anomaly category label output probability is less than or equal to 0.9 and greater than 0.6 or the confidence level is less than 1.2 and greater than or equal to 0.5. The confidence level for low confidence (which requires confidence alarm) is that the threshold of the anomaly category label output probability is less than or equal to 0.6 or the confidence level is greater than or equal to 1.2.

[0012] The output layer of the anomaly type prediction model is a Sigmoid function, and MC Diopout is introduced. The confidence level for high confidence is that the output probability of each anomaly category label is greater than 0.9 or less than 0.1 and the binary entropy is less than 0.2, and the probability variance is less than 0.005. The confidence level for low confidence is that the output probability of each anomaly category label is less than or equal to 0.55 and greater than or equal to 0.45, or the probability variance is less than 0.005 or the binary entropy is greater than 0.9. The confidence level for low confidence is not either high confidence or low confidence.

[0013] S5. Collect the anomaly monitoring data and historical anomaly monitoring data corresponding to each anomaly type label, mix them according to the set mixing rules to generate new anomaly monitoring data samples, and train and update the anomaly type prediction model based on the new anomaly monitoring data samples.

[0014] Furthermore, S1 includes the following steps:

[0015] S1.1 Monitoring the environment of the power distribution room based on environmental monitoring sensors to generate environmental monitoring data of the power distribution room; wherein, the environmental monitoring sensors include temperature and humidity sensors, smoke sensors, SF6 sensors, water immersion sensors, noise sensors and gas sensors, etc.

[0016] S1.2. Monitor the operating status of the equipment in the substation based on equipment status monitoring sensors to generate equipment status monitoring data. This data may include: battery voltage, current, internal resistance, temperature (cell / group), and charging / discharging status; cable head / busbar temperature; transformer oil temperature, winding temperature, oil level, dissolved gases in the oil (e.g., H2, CH4, C2H4, C2H6, C2H2, CO, CO2, O2, N2), core / clamping grounding current, vibration, and load current / voltage; and partial discharge quantity (pC), discharge phase (φ), number of discharges, discharge type, and ultrasonic / ultra-high frequency signal strength.

[0017] S1.3 Set up a sliding window to select the most recently set duration of environmental monitoring data and equipment status monitoring data of the power distribution room, and generate real-time monitoring data.

[0018] Furthermore, S2 includes the following steps:

[0019] S2.1 Collect environmental monitoring data and equipment status monitoring data of the power distribution room under normal conditions, and generate normal equipment status monitoring data;

[0020] S2.2. The real-time monitoring data and the equipment normal status monitoring data are subjected to feature extraction and standardization processing respectively to generate feature standardization vectors for real-time monitoring data and equipment normal status monitoring data. During feature extraction, time-domain and frequency-domain features can be extracted as needed. Time-domain features include statistical features such as mean, median, variance, standard deviation skewness (measuring distribution asymmetry), kurtosis (measuring distribution sharpness), maximum value, minimum value, range (Max-Min), quantiles, etc.; dynamic characteristics such as zero-crossing rate, autocorrelation function, lag difference statistics (such as mean absolute rate of change); extreme value features such as peak count, peak-valley amplitude, etc.; approximate entropy, sample entropy, permutation entropy, etc.; frequency-domain features include spectral statistics after converting the time-domain signal to the frequency domain through Fourier transform, such as spectral mean, variance, centroid, spectral bandwidth, spectral flatness; energy distribution, phase information, etc.; and time-frequency domain features such as wavelet coefficient energy, wavelet entropy, statistical features of the time-spectrum graph, energy of intrinsic mode functions, and instantaneous frequency, etc. This is to ensure the accuracy of subsequent similarity assessments.

[0021] S2.3. Based on the cosine similarity algorithm, calculate the similarity between the standardized vector of the real-time monitoring data features and the standardized vector of the equipment normal status monitoring data features, and determine whether it is greater than or equal to the set normal similarity threshold. If so, generate the predicted equipment normal status monitoring data.

[0022] Furthermore, the training of the anomaly type prediction model includes the following steps:

[0023] Collect historical anomaly monitoring data and corresponding anomaly category labels;

[0024] The historical anomaly monitoring data is subjected to feature extraction and standardization processing to generate a standardized vector of historical anomaly monitoring features;

[0025] Based on the standardized vector of historical anomaly monitoring features as input and the corresponding anomaly category label as output, the parameters of the selected neural network model are trained to generate an anomaly type prediction model. The neural network model can be selected as needed, such as SVM, Random Forest, XGBoost, LightGBM, CatBoost, CNN, RNN, LSTM, etc. Specifically:

[0026] Furthermore, the step of training the selected neural network model based on the standardized vector of the historical anomaly monitoring features as input and the corresponding anomaly category label as output to generate an anomaly type prediction model includes the following steps:

[0027] A1. The standardized vectors of the historical anomaly monitoring features and the corresponding anomaly category labels are divided into a training set and a validation set according to the set training ratio;

[0028] A2. Input the standardized vector of historical anomaly monitoring features in the training set into the anomaly type prediction model, and output the predicted anomaly category label;

[0029] A3. Calculate the error between the predicted anomaly category label and the corresponding anomaly category label in the training set based on the set loss function, and determine whether the error is less than the set error threshold. If yes, determine the parameters of the anomaly type prediction model and execute A5; otherwise, execute A4.

[0030] A4. Calculate the gradient of the parameter through backpropagation, and optimize and update the parameter based on the gradient and the optimization algorithm, then return to A2;

[0031] A5. Input the standardized vector of historical anomaly monitoring features in the validation set into the anomaly type prediction model, and output the predicted anomaly category label;

[0032] A6. Based on the anomaly category labels corresponding to the validation set and the predicted anomaly category labels, the performance of the anomaly type prediction model is validated. The performance validation metrics include accuracy, F1 score, AUC-ROC, precision, and recall.

[0033] Furthermore, generating the corresponding predicted label confidence data includes the following steps:

[0034] Based on the anomaly type prediction model, the standardized vector of the anomaly monitoring features is propagated forward once or multiple times;

[0035] Obtain the output probability of each anomaly category label from the anomaly type prediction model during each forward propagation, and generate an anomaly category label output probability dataset.

[0036] Based on the probability dataset of the anomaly category label data, the confidence level of the output of each anomaly category label data is analyzed to generate corresponding prediction label confidence data.

[0037] In one embodiment, when the anomaly monitoring feature standardized vector is forward-propagated once based on the anomaly type prediction model, the output probability of each anomaly category label data is obtained, and an anomaly category label output probability data set is generated. If the output layer of the anomaly type prediction model uses the softmax function, the highest probability value is obtained, and the entropy of each anomaly type label data is calculated; if the output layer of the anomaly type prediction model uses the sigmoid function, the binary entropy of each anomaly type label is calculated, or the average confidence of all anomaly type labels is calculated, etc.

[0038] In one embodiment, when implementing an anomaly type prediction model based on Monte Carlo Dropout for a single prediction, the normalized vector of anomaly monitoring features is forward-propagated multiple times to obtain the output probability of each anomaly category label in each forward propagation, generating an anomaly category label output probability dataset. Based on the anomaly category label output probability dataset, the average output probability of each anomaly category label in multiple forward propagations is calculated, the variance of the output probability of each anomaly category label in multiple forward propagations is calculated, and the entropy of the average probability distribution of each anomaly category label is calculated. Furthermore, the anomaly category labels output in each forward propagation can be obtained, and the consistency of the anomaly type labels output by the anomaly type prediction model can be analyzed.

[0039] Furthermore, S5 includes the following steps:

[0040] S5.1 Collect historical anomaly monitoring data and corresponding anomaly type labels, and collect the predicted anomaly type labels corresponding to the anomaly monitoring dataset;

[0041] S5.2. Classify the historical anomaly monitoring data and the anomaly monitoring data according to the different anomaly type labels, and generate the historical anomaly type anomaly monitoring data set and the anomaly type anomaly monitoring data set respectively.

[0042] S5.3 Determine whether the number of abnormal monitoring data in the abnormal monitoring data set of each abnormality type is greater than or equal to the corresponding set abnormality type update quantity threshold; wherein an abnormality type abnormal monitoring data set can have a separate abnormality type update quantity threshold set;

[0043] S5.4 If so, the historical anomaly monitoring data of the corresponding historical anomaly type anomaly monitoring data set is mixed with the anomaly monitoring data of the corresponding anomaly type update number threshold in the anomaly type anomaly monitoring data set according to the set ratio x, to generate an anomaly type anomaly monitoring update data set, where x = number of anomaly monitoring data / number of historical anomaly monitoring data.

[0044] S5.5 If not, then set the corresponding historical anomaly monitoring data set (corresponding to the set anomaly type update quantity threshold / x) of the historical anomaly monitoring data set as the anomaly type anomaly monitoring update data set;

[0045] S5.6 Collect the anomaly monitoring update data set corresponding to all anomaly type labels, and train and update the parameters of the anomaly type prediction model.

[0046] Furthermore, S5.4 also includes the following steps:

[0047] S5.4.1 If so, perform feature extraction and standardization processing on the historical anomaly type anomaly monitoring data set and the anomaly type anomaly monitoring data set respectively to generate historical anomaly type anomaly monitoring feature standardization vector and anomaly type anomaly monitoring feature standardization vector.

[0048] S5.4.2 Calculate the absolute value of the similarity error between the standardized vector of the historical anomaly type anomaly monitoring feature and the standardized vector of the anomaly type anomaly monitoring feature based on cosine similarity, and generate anomaly type feature change data;

[0049] S5.4.3 The setting ratio increases as the abnormal type feature change data corresponding to the corresponding abnormal category label increases and decreases as the abnormal type feature change data decreases. That is, the larger the abnormal type feature change data, the fewer the number of historical abnormal type monitoring data in the corresponding abnormal type monitoring update data set.

[0050] S5.4.4. Mix the historical anomaly monitoring data of the corresponding historical anomaly type anomaly monitoring data set with the anomaly monitoring data of the corresponding anomaly type update quantity threshold in the anomaly type anomaly monitoring data set according to the updated setting ratio x, and generate an anomaly type anomaly monitoring update data set.

[0051] An IoT-based power distribution room equipment status detection system, used to execute the IoT-based power distribution room equipment status detection method according to any one of claims 1-8, includes a sensor module, an interface, a storage device, a processor, and a display module;

[0052] The sensor module is used to collect environmental monitoring data and equipment status monitoring data of the power distribution room in real time, and transmit them to the storage device for storage via the interface;

[0053] Storage is used to store computer programs;

[0054] The processor is used to execute the computer program to implement a method for detecting the status of equipment in a power distribution station based on the Internet of Things;

[0055] The display module is used to visualize all the data in the system.

[0056] 1. Compared with the prior art, the present invention provides a method and system for detecting the status of equipment in a power distribution room based on the Internet of Things. By extracting the features of real-time monitoring data and equipment normal status monitoring data and comparing their similarity, the present invention can quickly determine whether the working status of the equipment in the power distribution room is normal, thereby reducing the difficulty of determining whether the equipment in the power distribution room is in normal working status.

[0057] 2. Compared with the prior art, the present invention provides a method and system for detecting the status of equipment in a power distribution station based on the Internet of Things. By comparing the characteristics of the time series of real-time monitoring data over a period of time, it can determine whether the equipment is normal. It can predict whether there are any abnormalities in the equipment over a period of time, and thus remind staff before the equipment malfunctions.

[0058] 3. Compared with the prior art, the present invention provides a method and system for detecting the status of equipment in a power distribution station based on the Internet of Things. When the equipment is in an abnormal state, the real-time monitoring data features are extracted and standardized and then input into the constructed anomaly type prediction model, and the corresponding anomaly type label is output. This allows for the rapid identification of the anomaly type of the equipment when it is abnormal, making it convenient for staff to maintain and repair the equipment according to the anomaly type.

[0059] 4. Compared with the prior art, the present invention provides a method and system for detecting the status of equipment in a power distribution station based on the Internet of Things. By analyzing the confidence level of the output anomaly type label when the anomaly type prediction model outputs anomaly type label, it can provide a reminder when the model has a low confidence in the accuracy of the output anomaly type label, thereby preventing staff from over-relying on the prediction results.

[0060] 5. Compared with the prior art, the present invention provides a method and system for detecting the status of equipment in a power distribution station based on the Internet of Things. By collecting the anomaly type labels and corresponding anomaly monitoring data output by the anomaly type prediction model, and analyzing the number of anomaly monitoring data corresponding to each anomaly type label and the feature differences with historical anomaly monitoring data, the ratio of the anomaly monitoring data corresponding to each anomaly type label to historical anomaly monitoring data can be adjusted when the model is updated. This ensures that the model can adaptively adjust according to the individual differences and aging status of the power distribution station equipment, while preventing the model from overfitting to short-term data. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0062] Figure 1 This is a flowchart illustrating the method steps provided in an embodiment of the present invention;

[0063] Figure 2 A system structure block diagram provided for embodiments of the present invention;

[0064] Figure 3 This is a step diagram of S5 provided in an embodiment of the present invention. Detailed Implementation

[0065] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0066] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0067] Exemplary embodiments will be described more fully below with reference to the accompanying drawings; however, these exemplary embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will enable those skilled in the art to fully understand the scope of this disclosure.

[0068] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0069] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0070] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded.

[0071] Please see Figure 1 A method for monitoring the status of equipment in a power distribution station based on the Internet of Things (IoT) includes the following steps:

[0072] S1. Collect environmental monitoring data and equipment status monitoring data of the power distribution room, and combine them to generate real-time monitoring data, including the following steps:

[0073] S1.1 Monitoring the environment of the power distribution room based on environmental monitoring sensors to generate environmental monitoring data of the power distribution room; wherein, the environmental monitoring sensors include temperature and humidity sensors, smoke sensors, SF6 sensors, water immersion sensors, noise sensors and gas sensors, etc.

[0074] S1.2. Monitor the operating status of the equipment in the substation based on equipment status monitoring sensors to generate equipment status monitoring data. This data may include: battery voltage, current, internal resistance, temperature (cell / group), and charging / discharging status; cable head / busbar temperature; transformer oil temperature, winding temperature, oil level, dissolved gases in the oil (e.g., H2, CH4, C2H4, C2H6, C2H2, CO, CO2, O2, N2), core / clamping grounding current, vibration, and load current / voltage; and partial discharge quantity (pC), discharge phase (φ), number of discharges, discharge type, and ultrasonic / ultra-high frequency signal strength.

[0075] S1.3 Set up a sliding window to select the most recently set duration of environmental monitoring data and equipment status monitoring data of the power distribution room, and generate real-time monitoring data.

[0076] S2. Based on real-time monitoring data and equipment normal status monitoring data, analyze and process whether the status of the substation equipment corresponding to the real-time monitoring data is normal. If so, generate predicted equipment normal status monitoring data, including the following steps:

[0077] S2.1 Collect environmental monitoring data and equipment status monitoring data of the power distribution room under normal conditions, and generate normal equipment status monitoring data;

[0078] S2.2. Feature extraction and standardization are performed on real-time monitoring data and equipment normal status monitoring data respectively, generating standardized feature vectors for real-time monitoring data and equipment normal status monitoring data. During feature extraction, time-domain and frequency-domain features can be extracted as needed. Time-domain features include statistical features such as mean, median, variance, standard deviation skewness (measuring distribution asymmetry), kurtosis (measuring distribution sharpness), maximum, minimum, range (Max-Min), quantiles, etc.; dynamic characteristics such as zero-crossing rate, autocorrelation function, lag difference statistics (such as mean absolute rate of change); extreme value features such as peak count, peak-valley amplitude, etc.; approximate entropy, sample entropy, permutation entropy, etc. Frequency-domain features include spectral statistics after converting the time-domain signal to the frequency domain using Fourier transform, such as spectral mean, variance, centroid, spectral bandwidth, spectral flatness; energy distribution, phase information, etc.; and time-frequency domain features such as wavelet coefficient energy, wavelet entropy, statistical characteristics of the time-frequency spectrum, energy of intrinsic mode functions, and instantaneous frequency, etc. This is to ensure the accuracy of subsequent similarity assessments.

[0079] S2.3. Based on the cosine similarity algorithm, calculate the similarity between the standardized vector of real-time monitoring data features and the standardized vector of equipment normal status monitoring data features, and determine whether it is greater than or equal to the set normal similarity threshold. If so, generate predicted equipment normal status monitoring data.

[0080] In one embodiment, training the anomaly type prediction model includes the following steps:

[0081] Collect historical anomaly monitoring data and corresponding anomaly category labels;

[0082] Feature extraction and standardization are performed on historical anomaly monitoring data to generate a standardized vector of historical anomaly monitoring features;

[0083] Using standardized vectors of historical anomaly monitoring features as input and corresponding anomaly category labels as output, a selected neural network model is trained to generate an anomaly type prediction model. The neural network model can be selected as needed, such as SVM, Random Forest, XGBoost, LightGBM, CatBoost, CNN, RNN, LSTM, etc. Specifically:

[0084] A1. Standardize the historical anomaly monitoring feature vectors and corresponding anomaly category labels into training set and validation set according to the set training ratio;

[0085] A2. Input the standardized vector of historical anomaly monitoring features in the training set into the anomaly type prediction model, and output the predicted anomaly category label;

[0086] A3. Calculate the error between the predicted anomaly category label and the corresponding anomaly category label in the training set based on the set loss function. Determine whether the error is less than the set error threshold. If yes, determine the parameters of the anomaly type prediction model and execute A5; otherwise, execute A4.

[0087] A4. Calculate the gradient of the parameters through backpropagation, and optimize and update the parameters based on the gradient and the optimization algorithm, then return A2;

[0088] A5. Input the standardized vector of historical anomaly monitoring features in the validation set into the anomaly type prediction model, and output the predicted anomaly category label;

[0089] A6. Based on the anomaly category labels corresponding to the validation set and the predicted anomaly category labels, the performance of the anomaly type prediction model is validated. The performance validation metrics include accuracy, F1 score, AUC-ROC, precision, and recall.

[0090] S3. If abnormal, the real-time monitoring data is marked as abnormal monitoring data. The abnormal monitoring feature standardization vector obtained by feature extraction and standardization of the abnormal monitoring data is input into the pre-trained abnormal type prediction model to generate the predicted abnormal type label and the corresponding prediction label confidence data.

[0091] In one embodiment, generating corresponding predicted label confidence data includes the following steps:

[0092] Based on the anomaly type prediction model, the standardized vector of anomaly monitoring features is propagated forward once or multiple times;

[0093] Obtain the output probability of each anomaly category label from the anomaly type prediction model at each forward propagation, and generate an anomaly category label output probability dataset.

[0094] Based on the probability dataset of anomaly category label data, the confidence level of the output of each anomaly category label data is analyzed to generate corresponding prediction label confidence data.

[0095] In one embodiment, when the anomaly monitoring feature normalization vector is propagated forward once based on the anomaly type prediction model, the output probability of each anomaly category label data is obtained, and an anomaly category label output probability data set is generated. If the output layer of the anomaly type prediction model uses the softmax function, the highest probability value is obtained, and the entropy of each anomaly type label data is calculated; if the output layer of the anomaly type prediction model uses the sigmoid function, the binary entropy of each anomaly type label or the average confidence of all anomaly type labels is calculated, etc.

[0096] In one embodiment, when implementing an anomaly type prediction model based on Monte Carlo Dropout for a single prediction, the normalized vector of anomaly monitoring features is forward-propagated multiple times to obtain the output probability of each anomaly category label in each forward propagation, generating an anomaly category label output probability dataset. Based on the anomaly category label output probability dataset, the average output probability of each anomaly category label in multiple forward propagations is calculated, the variance of the output probability of each anomaly category label in multiple forward propagations is calculated, and the entropy of the average probability distribution of each anomaly category label is calculated. Furthermore, the anomaly category labels output in each forward propagation can be obtained, and the consistency of the anomaly type labels output by the anomaly type prediction model can be analyzed.

[0097] S4. Based on the predicted label confidence data, determine whether the confidence level of the predicted anomaly category label is less than the set confidence level standard. If not, output the predicted anomaly type label; if so, output the predicted anomaly type label and the corresponding predicted label confidence data at the same time, and issue a confidence level alarm. The confidence level alarm can be implemented by sending an alarm pop-up, sending emails to administrators, SMS messages, etc.

[0098] There can be multiple confidence levels, and each confidence level corresponds to a confidence standard. The confidence standard can be a threshold for the output probability of the anomaly category label, a threshold for entropy or binary entropy, a variance threshold for the average probability of the output of each anomaly category label after multiple forward propagations, etc.

[0099] For example, the output layer of the anomaly type prediction model is a softmax function. The confidence level for high confidence is that the threshold of the anomaly category label output probability is greater than 0.9 or the entropy is less than 0.5. The confidence level for medium confidence is that the threshold of the anomaly category label output probability is less than or equal to 0.9 and greater than 0.6 or the confidence level is less than 1.2 and greater than or equal to 0.5. The confidence level for low confidence (which requires confidence alarm) is that the threshold of the anomaly category label output probability is less than or equal to 0.6 or the confidence level is greater than or equal to 1.2.

[0100] The output layer of the anomaly type prediction model is a Sigmoid function, and MC Diopout is introduced. The confidence level for high confidence is that the output probability of each anomaly category label is greater than 0.9 or less than 0.1 and the binary entropy is less than 0.2, and the probability variance is less than 0.005. The confidence level for low confidence is that the output probability of each anomaly category label is less than or equal to 0.55 and greater than or equal to 0.45, or the probability variance is less than 0.005 or the binary entropy is greater than 0.9. The confidence level for low confidence is not either high confidence or low confidence.

[0101] S5. Collect anomaly monitoring data and historical anomaly monitoring data corresponding to each anomaly type label, mix them according to the set mixing rules to generate new anomaly monitoring data samples, and train and update the anomaly type prediction model based on the new anomaly monitoring data samples, including the following steps:

[0102] S5.1 Collect historical anomaly monitoring data and corresponding anomaly type labels, and collect the predicted anomaly type labels corresponding to the anomaly monitoring dataset;

[0103] S5.2 Classify the historical anomaly monitoring data and the anomaly monitoring data according to the different anomaly type labels, and generate the historical anomaly type anomaly monitoring data set and the anomaly type anomaly monitoring data set respectively;

[0104] S5.3 Determine whether the number of abnormal monitoring data in the abnormal monitoring data set of each abnormality type is greater than or equal to the corresponding set abnormality type update quantity threshold; wherein an abnormality type abnormal monitoring data set can have a separate abnormality type update quantity threshold set;

[0105] S5.4 If so, the historical anomaly monitoring data of the corresponding historical anomaly type anomaly monitoring data set is mixed with the anomaly monitoring data of the corresponding anomaly type anomaly type anomaly monitoring data set according to the set ratio x, and anomaly monitoring data of the corresponding set anomaly type update number threshold is generated to generate anomaly type anomaly monitoring update data set, where x = number of anomaly monitoring data / number of historical anomaly monitoring data.

[0106] In one embodiment, S5.4 may further include the following steps:

[0107] S5.4.1 If so, perform feature extraction and standardization on the historical anomaly type anomaly monitoring dataset and the anomaly type anomaly monitoring dataset respectively to generate the historical anomaly type anomaly monitoring feature standardization vector and the anomaly type anomaly monitoring feature standardization vector.

[0108] S5.4.2 Calculate the absolute value of the similarity error between the standardized vector of historical anomaly type anomaly monitoring features and the standardized vector of anomaly type anomaly monitoring features based on cosine similarity, and generate anomaly type feature change data;

[0109] S5.4.3 The setting ratio increases as the change data of the anomaly type feature corresponding to the anomaly category label increases and decreases as the change data decreases. That is, the larger the change data of the anomaly type feature, the fewer the number of historical anomaly type anomaly monitoring data in the corresponding anomaly type anomaly monitoring update data set.

[0110] For example, increasing with increasing and decreasing with decreasing can be done in a stepwise manner. For instance, if the change in the corresponding anomaly type feature data is between 0 and 0.1 (excluding 0.1), x is 1:1; if the change in the anomaly type feature data is between 0.1 and 0.3 (excluding 0.3), x is 2:3; if the change in the anomaly type feature data is between 0.3 and 0.5 (excluding 0.5), x is 1:3; if the change in the anomaly type feature data is greater than 0.5, x is 1:5, and so on.

[0111] Increasing and decreasing can also be functional, such as x=k*R (R is not equal to 0, x is 1 when R is 0), where k is a coefficient and R is the abnormal type characteristic change data.

[0112] S5.4.4. Mix the historical anomaly monitoring data of the corresponding historical anomaly type anomaly monitoring data set with the anomaly monitoring data of the corresponding anomaly type update quantity threshold in the anomaly type anomaly monitoring data set according to the updated setting ratio x, and generate an anomaly type anomaly monitoring update data set.

[0113] S5.5 If not, then set the corresponding historical anomaly monitoring data set (corresponding to the set anomaly type update quantity threshold / x) of the historical anomaly monitoring data set as the anomaly type anomaly monitoring update data set;

[0114] S5.6 Collect the anomaly monitoring update data set corresponding to all anomaly type labels, and train and update the parameters of the anomaly type prediction model.

[0115] Please see Figure 2 The present invention also provides an IoT-based equipment status detection system for power distribution substations, which is used in the above-mentioned IoT-based equipment status detection method for power distribution substations provided by the present invention, including a sensor module, an interface, a storage device, a processor, and a display module;

[0116] The sensor module is used to collect environmental monitoring data and equipment status monitoring data of the power distribution room in real time, and transmit them to the storage device for storage through the interface. The sensor module includes various types of sensors, and there can be several of each type of sensor. For example, the sensors may include temperature and humidity sensors, smoke sensors, SF6 sensors, water immersion sensors, noise sensors, gas sensors, voltage sensors, current sensors, vibration sensors, partial discharge sensors, etc.

[0117] Storage is used to store computer programs;

[0118] The processor is used to execute computer programs to implement the IoT-based equipment status detection method for power distribution rooms provided by this invention;

[0119] The display module is used to visualize all data in the system.

[0120] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A power distribution station equipment state detection method based on the Internet of Things, characterized in that, The method comprises the following steps: S1, collecting power distribution station environment monitoring data and power distribution station equipment state monitoring data, and combining to generate real-time monitoring data; S2, based on the real-time monitoring data and the normal state monitoring data of the equipment, whether the power distribution station equipment state corresponding to the real-time monitoring data is normal is analyzed and processed, if yes, the predicted equipment normal state monitoring data is generated; S3, if not, the real-time monitoring data is marked as abnormal monitoring data, and the abnormal monitoring feature standardization vector obtained by feature extraction and standardization processing of the abnormal monitoring data is input into the pre-trained abnormal type prediction model to generate the predicted abnormal type label and the corresponding prediction label confidence data; S4, based on the prediction label confidence data, whether the confidence of the predicted abnormal class label is less than the set confidence standard is judged, if not, the predicted abnormal type label is output; if yes, the predicted abnormal type label and the corresponding prediction label confidence data are output, and confidence alarm is performed; S5, collecting the abnormal monitoring data and historical abnormal monitoring data corresponding to each abnormal type label, mixing to generate abnormal monitoring data new samples according to the set mixing rule, and training and updating the abnormal type prediction model based on the abnormal monitoring data new samples; The S5 comprises the following steps: S5.1, collecting historical abnormal monitoring data and corresponding abnormal type label, and collecting the predicted abnormal type label corresponding to the abnormal monitoring data set; S5.2, classifying the historical abnormal monitoring data and the abnormal monitoring data according to the different abnormal type labels respectively, and generating the historical abnormal type abnormal monitoring data set and the abnormal type abnormal monitoring data set respectively; S5.3, judging whether the number of abnormal monitoring data in each abnormal type abnormal monitoring data set is greater than or equal to the corresponding set abnormal type update quantity threshold value; S5.4, if yes, mixing the historical abnormal monitoring data of the corresponding historical abnormal type abnormal monitoring data set with the corresponding set abnormal type update quantity threshold value abnormal monitoring data in the abnormal type abnormal monitoring data set according to the set proportion x, to generate the abnormal type abnormal monitoring update data set, wherein x=abnormal monitoring data quantity / historical abnormal monitoring data quantity; S5.5, if not, setting the corresponding set abnormal type update quantity threshold value / x historical abnormal monitoring data corresponding to the corresponding historical abnormal type abnormal monitoring data set as the abnormal type abnormal monitoring update data set; S5.6, collecting the abnormal type abnormal monitoring update data set corresponding to all abnormal type labels, and training and updating the parameters of the abnormal type prediction model. 2.The power distribution station equipment state detection method based on the Internet of Things according to claim 1, characterized in that, The S1 comprises the following steps: S1.1, monitoring the power distribution station environment based on the environment monitoring sensor to generate the power distribution station environment monitoring data; S1.2, monitoring the running state of the power distribution station equipment based on the equipment state monitoring sensor to generate the power distribution station equipment state monitoring data; S1.3, setting a sliding window, selecting the power distribution station environment monitoring data and the power distribution station equipment state monitoring data in the latest set time period, and generating the real-time monitoring data. 3.The power distribution station equipment state detection method based on the Internet of Things according to claim 1, characterized in that, The S2 comprises the following steps: S2.1, collect power distribution station environment monitoring data and power distribution station equipment state monitoring data in normal state of power distribution station equipment, and generate equipment normal state monitoring data; S2.2, the feature extraction and standardization processing are carried out on the real-time monitoring data and the equipment normal state monitoring data respectively, and the real-time monitoring data feature standardization vector and the equipment normal state monitoring data feature standardization vector are generated; S2.3, based on cosine similarity algorithm, the similarity of the real-time monitoring data feature standardization vector and the equipment normal state monitoring data feature standardization vector is calculated, and it is judged whether it is greater than or equal to the set normal similarity threshold, if yes, the predicted equipment normal state monitoring data is generated.

4. The power distribution station equipment state detection method based on the Internet of Things according to claim 1, characterized in that, The abnormal type prediction model training includes the following steps: Collecting historical abnormal monitoring data and corresponding abnormal category labels; The feature extraction and standardization processing are carried out on the historical abnormal monitoring data, and the historical abnormal monitoring feature standardization vector is generated; Based on the historical abnormal monitoring feature standardization vector as input, the corresponding abnormal category label as output, the parameter training of the selected neural network model is carried out, and the abnormal type prediction model is generated.

5. The power distribution station equipment state detection method based on the Internet of Things according to claim 4, characterized in that, The abnormal type prediction model training includes the following steps: A1, the historical abnormal monitoring feature standardization vector and the corresponding abnormal category label are divided into training set and validation set according to the set training proportion; A2, the historical abnormal monitoring feature standardization vector in the training set is input into the abnormal type prediction model, and the predicted abnormal category label is output; A3, based on the set loss function, the error of the predicted abnormal category label and the corresponding abnormal category label in the training set is calculated, and it is judged whether the error is less than the set error threshold, if yes, the parameters of the abnormal type prediction model are determined, and A5 is executed; if not, A4 is executed; A4, the gradient of the parameters is calculated by back propagation, and the parameters are optimized and updated based on the gradient and the optimization algorithm, and A2 is returned; A5, the historical abnormal monitoring feature standardization vector in the validation set is input into the abnormal type prediction model, and the predicted abnormal category label is output; A6, the performance of the abnormal type prediction model is verified based on the corresponding abnormal category label and the predicted abnormal category label in the validation set.

6. The power distribution station equipment state detection method based on the Internet of Things according to claim 1, characterized in that, The corresponding prediction label confidence data is generated, including the following steps: Based on the abnormal type prediction model, the abnormal monitoring feature standardization vector is forward propagated one or more times; The output probability of each abnormal category label output by the abnormal type prediction model is obtained, and an abnormal category label output probability data set is generated; Based on the abnormal category label output probability data set, the confidence of each abnormal category label output is analyzed, and the corresponding prediction label confidence data is generated.

7. The power distribution station equipment state detection method based on the Internet of Things according to claim 1, characterized in that, The S5.4 further includes the following steps: S5.4.1, if yes, respectively, the historical abnormal type abnormal monitoring data set and the abnormal type abnormal monitoring data set are extracted and standardized, and the historical abnormal type abnormal monitoring feature standardized vector and the abnormal type abnormal monitoring feature standardized vector are generated; S5.4.2, based on the cosine similarity, the similarity error absolute value of the historical abnormal type abnormal monitoring feature standardized vector and the abnormal type abnormal monitoring feature standardized vector is calculated, and the abnormal type feature change data is generated; S5.4.3, the setting ratio increases with the increase of the abnormal type feature change data corresponding to the corresponding abnormal category label, and decreases with the decrease of the abnormal type feature change data corresponding to the corresponding abnormal category label; S5.4.4, the historical abnormal monitoring data corresponding to the historical abnormal type abnormal monitoring data set is mixed with the abnormal monitoring data in the abnormal type abnormal monitoring data set according to the updated setting ratio x and the corresponding setting abnormal type update quantity threshold, and the abnormal type abnormal monitoring update data set is generated.

8. An Internet of Things-based power distribution station equipment state detection system for performing an Internet of Things-based power distribution station equipment state detection method according to any one of claims 1-7, characterized in that, The system comprises a sensor module, an interface, a storage, a processor, a display module; The sensor module is used to collect power station environment monitoring data and power station equipment state monitoring data in real time, and transmit to the storage through the interface for saving; The storage is used to store computer programs; The processor is used to execute the computer program, realize the power station equipment state detection method based on the Internet of Things as claimed in any one of claims 1-7; The display module is used to visually display all data in the system.

Images