A method for preventing maloperation of a relay protection

Abstract

The application provides a method for preventing relay protection misoperation, and relates to the technical field of electrical protection. According to all historical power consumption data, the method simultaneously corrects the error judgment result of a relay protector, extracts features and zero sequence components for training, obtains a fault prediction model, constructs a basic condition sufficient to cope with various grounding faults, adjusts the protection threshold of the relay protector according to the output result of the fault prediction model, constructs a floating judgment standard, compares the output of the fault prediction model, strengthens the identification of grounding faults, further assists the adjustment of the protection threshold through historical use records, realizes the judgment sufficient to cope with various grounding faults, and avoids the misjudgment of not tripping when it should be tripped and tripping when it should not be tripped.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical protection, in particular to a method for preventing misoperation of relay protection. BACKGROUND

[0002] As the infrastructure of modern society, the stable operation of power system is directly related to the national economy and people's livelihood safety. The relay protection device is the core equipment for detecting and isolating faults in the power system, which can remove faulty components in a short time to avoid the expansion of the accident. However, in actual operation, especially in the process of ground fault, the phenomenon of relay protection misoperation (i.e. not tripping when it should trip, or refusing to operate when it should) is common. The main reasons include external electromagnetic interference (such as lightning, high-voltage cable switch operation or adjacent equipment interference), insulation performance degradation due to line aging, non-standard secondary circuit construction, equipment component performance degradation, imperfect system design, and human operation errors, etc. These factors often lead to a decrease in the accuracy of the protection device, especially in the detection of ground anomalies, the extraction of zero sequence components is easily affected by noise, leading to misjudgment, and thus causing power interruption, economic loss and even equipment damage.

[0003] In the prior art, the disclosure number CN113300338A discloses a method, device, medium and equipment for suppressing misoperation of relay protection of a power supply system. The method detects the zero sequence current signal in real time, determines the cause of the zero sequence current signal when the zero sequence current signal is detected, and controls the relay protection device to perform a trip action when the cause of the zero sequence current signal is the bus grounding of the power supply system. When the zero sequence current signal is detected, the cause of the zero sequence current signal needs to be further determined. Only when the cause of the zero sequence current signal is the bus grounding of the power supply system, the relay protection device is controlled to perform a trip action.

[0004] Due to the variety of ground fault forms, the manifestations in the zero sequence current signal are also diverse. Although the disclosed technical document has achieved a reduction in the probability of misoperation of the relay protection device by tightening the trip generation condition, the setting condition of the trip is too high, making it very easy for the relay protection device to fail to trip when it should, which will also cause losses and lack of practicality.

[0005] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a method for preventing misoperation of relay protection to solve the problems raised in the background.

[0007] To achieve the above object, the present application provides the following technical solutions:

[0008] A method for preventing relay protection misoperation, comprising the following specific steps:

[0009] Step 1: Obtain historical power consumption data, the power consumption data including current curve and voltage curve, set a sliding window and a sampling rate in the power consumption data, obtain features and zero sequence components from the power consumption data in the sliding window, form a sample set from the features and the zero sequence components, and associate the sample set with whether the relay protector performs a tripping action, the tripping action being a correct action after correction;

[0010] Step 2: input the sample set as features and the corresponding tripping action as features, train in a gradient boosting tree model to obtain a fault prediction model, obtain real-time power consumption data, extract features and zero sequence components, input the features and the zero sequence components into the fault prediction model, obtain a judgment result of the fault prediction model on whether to trip, and generate a probability of the judgment result being tripping through an output layer of the fault prediction model, and obtain a confidence of the judgment result through the probability;

[0011] Step 3: input the real-time power consumption data into the relay protector and obtain a judgment result of the relay protector on whether to trip, compare the judgment result of the relay protector with the judgment result of the fault prediction model, if the two judgment results are the same, directly execute, if the two judgment results are different, obtain a protection threshold of the relay protector, adjust the protection threshold through the confidence, generate a judgment result of the relay protector on whether to trip again according to the adjusted protection threshold, and execute according to the result;

[0012] Step 4: set a protection threshold adjustment upper limit when adjusting the protection threshold, obtain historical use records of the relay protector and generate a state evaluation index, and adjust the protection threshold adjustment upper limit through the state evaluation index.

[0013] Further, all historical power consumption data are obtained through a secondary circuit, the power consumption data including voltage curves and current curves of three-phase A, B and C, the same sliding window and sampling rate are set in the voltage curves and the current curves, and the start and end points of each sliding window in the voltage curves and the current curves of three-phase A, B and C are the same, the length of the sliding window is greater than 5 phase change periods, and the sampling rate of each sliding window is greater than 1 kHz, zero sequence components corresponding to each sliding window are obtained through a symmetrical component method, the zero sequence components including zero sequence complex vectors and zero sequence amplitudes.

[0014] Further, the current curve and the voltage curve of each sliding window are preprocessed, and features are extracted from the voltage curve and the current curve of the A, B and C three-phase within each sliding window, the features including current mean value, voltage mean value, current main frequency amplitude, voltage main frequency amplitude, current high frequency energy ratio, voltage high frequency energy ratio and system load;

[0015] The features in a single sliding window and the zero sequence component are taken as a sample set, and the sample set logic is as follows:

[0016] The elements in the sample set include current mean value, voltage mean value, voltage main frequency amplitude, current main frequency amplitude, current high frequency energy ratio, voltage high frequency energy ratio, system load, zero sequence complex vector and zero sequence amplitude of the zero sequence complex vector;

[0017] A label is added to each sample set, the label being the trip condition corresponding to the sliding window, when the label is trip, it is recorded as 1, and when the label is not trip, it is recorded as 0;

[0018] All sample sets and corresponding labels are summarized and normalized, and all sample sets are divided into a training set and a verification set according to 8:2.

[0019] Further, the training set is input into a gradient boosting tree model for training to obtain a fault prediction model, and the output is a judgment of whether to trip, and at the same time, the output layer generates a trip probability.

[0020] Further, real-time power consumption data is collected through the secondary circuit, and features and zero sequence components are obtained according to a set sliding window, the features and the zero sequence components are input into the fault prediction model to obtain a judgment of whether to trip and a trip probability, and a judgment confidence is generated according to the trip probability, and the logic is as follows:

[0021] The trip probability is multiplied by a trip probability logarithmic function to form a first intermediate quantity, the trip probability logarithmic function being a logarithm of the trip probability and a constant as a base, a second intermediate quantity is formed by multiplying the not trip probability by a not trip probability logarithmic function, the not trip probability logarithmic function being a logarithm of the not trip probability and a constant as a base, the not trip probability being equal to 1 minus the trip probability, the first intermediate quantity, the second intermediate quantity and a first constant are added to equal the judgment confidence, the trip probability being closer to 1 and 0, the judgment confidence being higher, and the trip probability being closer to 0.5, the judgment confidence being lower.

[0022] Further, the power consumption data of the secondary circuit is analyzed by a relay protector, and a judgment result of the relay protector is obtained, the judgment of the relay protector including trip and not trip;

[0023] The output of the fault prediction model is compared with the judgment result of the relay protector. If the results are the same, the judgment result of the relay protector is directly executed.

[0024] Further, if the output of the fault prediction model is not the same as the judgment result of the relay protector, the protection threshold value set in the relay protector is obtained, which represents the tripping condition set by the system at the design time to ensure the safety of the equipment. The protection threshold value of the relay protector is adjusted by judging the confidence, and the logic is as follows:

[0025] When the judgment result of the fault prediction model is tripping, the judgment result is calibrated as a constant 1, and when the judgment result of the fault prediction model is not tripping, the judgment result is calibrated as a constant 0. A third intermediate quantity is obtained by subtracting the constant 0.5 from the judgment result. The third intermediate quantity controls the adjustment direction of the threshold value, and when the judgment result is tripping, the protection threshold value is reduced, and when the judgment result is not tripping, the protection threshold value is increased. A fourth intermediate quantity is obtained by multiplying the third intermediate quantity and the confidence of the judgment result of the fault prediction model. The adjusted protection threshold value is obtained by multiplying the intermediate quantity by the negative and the protection threshold value before adjustment. The fourth intermediate quantity controls the adjustment amplitude of the threshold value, and the higher the confidence, the greater the adjustment amplitude.

[0026] At the same time, an upper limit of the protection threshold value adjustment is set, which represents the relative change of the adjusted protection threshold value relative to the protection threshold value before adjustment.

[0027] Further, the relay protector uses the adjusted protection threshold value to judge the power consumption data again, and executes tripping or non-tripping operation according to the judgment result.

[0028] Further, the historical use record of the relay protector is obtained, which includes the running years, the rated service life, the number of tripping operations, and the number of judgment errors, the number of judgment errors including the number of times that should trip but did not trip and the number of times that should not trip but did trip. The state evaluation index of the relay protector is obtained, and the logic is as follows:

[0029] The fifth intermediate quantity is formed by multiplying the ratio of the running years to the rated service life by the number of judgment errors. The more the running years or the number of judgment errors, the larger the fifth intermediate quantity. The smaller the rated service life, the larger the fifth intermediate quantity. The sixth intermediate quantity is formed by the logarithmic function of the ratio of the number of tripping operations to the running years. The ratio of the number of tripping operations to the running years is taken as the logarithm, and the constant is taken as the base number. The larger the number of tripping operations, the larger the sixth intermediate quantity. The smaller the running years, the larger the sixth intermediate quantity. The state evaluation index is obtained by assigning weights to the fifth intermediate quantity and the sixth intermediate quantity and summing them up.

[0030] The state evaluation index threshold is set, and when the state evaluation index does not exceed the state evaluation index threshold, the state evaluation index does not adjust the protection threshold upper limit, and when the state evaluation index does not exceed the state evaluation index threshold, the state evaluation index adjusts the protection threshold upper limit, and the logic is as follows:

[0031] The product of the state evaluation index and the protection threshold upper limit before adjustment is taken as an adjustment factor, the adjustment factor is added to the protection threshold upper limit before adjustment to obtain the protection threshold upper limit after adjustment, the larger the protection threshold upper limit before adjustment, the larger the protection threshold upper limit after adjustment, and the larger the state evaluation index, the larger the protection threshold after adjustment.

[0032] Compared with the prior art, the beneficial effects of the present application are:

[0033] The present application corrects the error judgment result of the relay protector according to all historical power consumption data, extracts features and zero sequence components for training, obtains a fault prediction model, constructs a basic condition sufficient to cope with various ground faults, adjusts the protection threshold of the relay protector according to the output result of the fault prediction model, constructs a floating judgment standard and compares it with the output of the fault prediction model, strengthens the identification of ground faults, further assists the adjustment of the protection threshold through historical use records, realizes the judgment sufficient to cope with various ground faults, and avoids the misjudgment of not tripping when it should trip and tripping when it should not. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 It is a whole method flow diagram of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with specific embodiments.

[0036] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0037] Embodiment:

[0038] Please refer to Figure 1 The present application provides a technical solution:

[0039] A method for preventing misoperation of a relay protection, comprising the following specific steps:

[0040] Step 1: Obtain historical power consumption data, the power consumption data including current curves and voltage curves, set a sliding window and a sampling rate in the power consumption data, obtain features and zero sequence components from the power consumption data in the sliding window, form a sample set from the features and the zero sequence components, and associate the sample set with whether the relay protection performs a tripping action, the tripping action being a correct action after correction;

[0041] The step 1 includes the following contents:

[0042] Step 101: Obtain all historical power consumption data through a secondary circuit, the power consumption data including voltage curves and current curves of three phases A, B and C, set the same sliding window and sampling rate in the voltage curves and the current curves, and the start and end points of each sliding window are the same in the voltage curves and the current curves of the three phases A, B and C, the length of the sliding window being greater than 5 phase change periods, and the sampling rate of each sliding window being greater than 1 kHz, obtain zero sequence components corresponding to each sliding window respectively by a symmetrical component method, the zero sequence components including zero sequence complex vectors and zero sequence amplitudes;

[0043] The sampling rate represents the frequency of obtaining sampling points in the voltage curves and the current curves per unit time.

[0044] Obtain power consumption data and zero sequence components of all stages, the all stages including a device normal operation non-tripping stage, a device ground abnormality detected and judged successfully tripping stage, a device ground abnormality detected and failed non-tripping stage, and a device ground normality detected and failed mis-tripping stage, manually change the result of the device ground abnormality detected and failed non-tripping stage to tripping, and change the result of the device ground normality detected and failed mis-tripping stage to non-tripping.

[0045] The historical power consumption data is collected through the secondary circuit, the sliding window and the sampling rate are uniformly set, the synchronization and high resolution of the time series data are ensured, the dynamic characteristics of the power system at different stages such as normal operation, successful detection of grounding abnormity and misjudgment scene are captured, the zero sequence component is accurately extracted through the symmetrical component method, and a pure original signal basis is provided for subsequent feature engineering; the fine data collection and stage correction mechanism significantly reduces the interference of data noise on the judgment of the relay protector, improves the early identification ability of the system to the grounding fault, and avoids the blind area of the traditional static threshold method under complex load; at the same time, it seamlessly connects with the subsequent preprocessing link, directly transmits the structured voltage and current curve and zero sequence component as input, and lays the conversion cornerstone from the original waveform to the trainable sample in the overall data flow, thereby enhancing the generalization performance of the fault prediction model, and finally building a solid data-driven framework to prevent the misoperation of the relay protection.

[0046] Step 102: Preprocessing the current curve and voltage curve of each sliding window, extracting features from the voltage curve and current curve of the A, B and C three-phase within each sliding window, the features including current mean value, voltage mean value, current main frequency amplitude, voltage main frequency amplitude, current high frequency energy ratio, voltage high frequency energy ratio and system load;

[0047] The features and zero sequence components in a single sliding window are taken as a sample set, and the sample set logic is as follows:

[0048] The elements in the sample set include current mean value, voltage mean value, voltage main frequency amplitude, current main frequency amplitude, current high frequency energy ratio, voltage high frequency energy ratio, system load, zero sequence complex vector and zero sequence amplitude of the zero sequence complex vector;

[0049] As a preferred embodiment, the sample set is based on the following formula:

[0050]

[0051] Wherein, represents the sample set, represents the current mean value, represents the voltage mean value, represents the voltage main frequency amplitude, represents the current main frequency amplitude, represents the current high frequency energy ratio, represents the voltage high frequency energy ratio, represents the system load, represents the zero sequence complex vector, represents the zero sequence amplitude of the zero sequence complex vector;

[0052] As a preferred embodiment, the current mean value, the voltage mean value, the current fundamental frequency amplitude, the voltage fundamental frequency amplitude, the current high-frequency energy ratio, and the voltage high-frequency energy ratio all contain data of three-phase currents A, B, and C, and these features are common technical features in the field, and the specific calculation method is not described here.

[0053] A label is added to each sample set, which is the trip condition corresponding to the sliding window. When the label is trip, it is recorded as 1, and when the label is not trip, it is recorded as 0.

[0054] All sample sets and corresponding labels are summarized and normalized, and all sample sets are divided into a training set and a validation set in a ratio of 8:2.

[0055] Through preprocessing and multi-dimensional feature extraction of the three-phase voltage and current curves in the sliding window, the massive time series data are condensed into low-dimensional and information-rich sample sets, which not only retains the frequency domain and time domain nature of power signals, but also integrates zero sequence components to highlight the asymmetry of ground fault. This feature selection and label labeling mechanism effectively balances the distribution of positive and negative samples, reduces the overfitting risk of the model to noise or abnormal load, improves the robustness and diagnostic accuracy of the relay protector in the real power grid environment, and avoids the limitations of traditional methods that simply rely on zero sequence current. At the same time, it forms a complete sample generation pipeline by taking the output of the original data stream from the previous step, provides a standardized and normalized data set segmentation for model training, and bridges the gap between data acquisition and machine learning training in the overall logic, thereby ensuring knowledge inheritance from historical experience to predictive ability and improving the intelligence level of the entire anti-malfunction system.

[0056] Step 2: input the sample set as a feature, and the corresponding trip action as a feature output. Train in the gradient boosting tree model to obtain a fault prediction model. Obtain real-time power consumption data and extract features and zero sequence components to input into the fault prediction model. Obtain the judgment result of the fault prediction model for whether to trip. At the same time, the output layer of the fault prediction model generates the probability of the judgment result being trip. The confidence of the judgment result is obtained through the probability.

[0057] The step 2 includes the following contents:

[0058] The training set is input into the gradient boosting tree model for training to obtain a fault prediction model. The output is a judgment of whether to trip. At the same time, the output layer generates the probability of trip.

[0059] As a preferred embodiment, the number of trees is determined to be 200, the maximum depth of the tree is 5, the learning rate is 0.1, the subsampling rate is 0.8, and the minimum split gain is 0.1.

[0060] The electric data is collected in real time through the secondary circuit, and the features and zero sequence components are obtained according to the set sliding window, the features and zero sequence components are input into the fault prediction model, the judgment of whether to trip and the probability of tripping are obtained, and the judgment confidence is generated according to the probability of tripping, and the logic is as follows:

[0061] The first intermediate quantity is formed by multiplying the tripping probability by the tripping probability logarithmic function, the tripping probability logarithmic function is the tripping probability as the logarithm and the constant as the base, the second intermediate quantity is formed by multiplying the non-tripping probability by the non-tripping probability logarithmic function, the non-tripping probability logarithmic function is the non-tripping probability as the logarithm and the constant as the base, the non-tripping probability is equal to 1 minus the tripping probability, the first intermediate quantity, the second intermediate quantity and the first constant are added to equal the judgment confidence, the judgment confidence is higher when the tripping probability is closer to 1 and 0, and the judgment confidence is lower when the tripping probability is closer to 0.5.

[0062] As a preferred embodiment, the judgment confidence is based on the following formula:

[0063]

[0064] Wherein, is the judgment confidence, is the tripping probability.

[0065] Specifically reflects the quantitative measure of the judgment uncertainty based on the tripping probability , which means that the reliability of the relay protection decision is evaluated through the concept of information entropy, and the fuzziness problem of ground fault detection in the real power environment is reflected, such as the probability boundary uncertainty caused by transient noise or load fluctuation, thereby avoiding the absolute risk of binary judgment; the independent variable is the tripping probability, which captures the confidence level of the fault prediction model output, represents the probabilistic expression of the system to abnormal events under complex power grid conditions, and highlights the gradual fault characteristics ignored by the traditional threshold method. and are closely related, because the formula is essentially derived from the binary entropy function , wherein directly converts uncertainty into confidence, has a symmetrical inverted U-shaped curve relationship with , that is, when deviates from 0.5 (close to the extreme value 0 or 1), the entropy decreases, resulting in increases, which reflects the low uncertainty and high confidence of high probability judgment, and maximizes the entropy, resulting in Minimization, reflecting the low reliability of decision ambiguity; in the size change relationship, with Increasing from 0 to 0.5, Monotonically decreasing to the minimum value, and then symmetrically increasing from 0.5 to 1, this nonlinear dynamic ensures that the confidence is lifted with the enhancement of probability certainty, thereby guiding the conservative adjustment of the protection threshold in practical applications, and strengthening the robustness of the anti-misoperation.

[0066] By inputting the aggregated sample set into the gradient boosting tree model for training and applying it to the power consumption data collected by the secondary circuit in real time, a complete data processing closed loop from offline learning to online prediction is constructed, which not only captures the complex interaction between features through the nonlinear decision boundary of the tree model, but also quantifies the uncertainty through probability output and confidence calculation; this mechanism significantly improves the response speed and accuracy of the relay protector to transient faults, reduces the missed detection under ground fault or the refusal to operate under normal state, ensures the continuity and safety of the power system, and avoids the rigidity of the traditional rule method; at the same time, it connects the training verification set of the previous step as the fuel for model optimization, and feeds the prediction judgment and confidence output directly to the comparison module, which is the core of the intelligent reasoning engine in the overall scheme, realizes the dynamic transformation of data from historical accumulation to real-time decision, and deepens the prediction depth and interpretability of anti-misoperation.

[0067] Step 3: input the real-time power consumption data into the relay protector and obtain the judgment result of the relay protector on whether to trip, compare the judgment result of the relay protector with the judgment result of the fault prediction model, if the two judgment results are the same, directly execute, if different, obtain the protection threshold of the relay protector, adjust the protection threshold through the confidence, generate the judgment result of the relay protector on whether to trip again according to the adjusted protection threshold, and execute according to the result;

[0068] The step 3 includes the following contents:

[0069] Step 301: analyze the power consumption data of the secondary circuit by the relay protector, and obtain the judgment result of the relay protector, the judgment of the relay protector includes tripping and not tripping;

[0070] Compare the output of the fault prediction model with the judgment result of the relay protector, if the results are the same, directly execute the judgment result of the relay protector.

[0071] Through independent analysis of the secondary circuit power data by the relay protector and direct comparison with the fault prediction model output, efficient execution path of the system in the consistency scene is ensured, which not only verifies the reliability of the model prediction, but also avoids the waste of resources caused by redundant calculation; this rapid consistency checking mechanism maintains the core authority of the relay protector, and at the same time, with the help of the model, it filters the noise interference of the edge case, improves the optimization of the overall response delay and the stability of the power grid operation, and prevents the system-level risk caused by a single judgment source; at the same time, it takes the real-time prediction result of the previous step as the comparison benchmark and provides a clear trigger signal for the difference branch, bridging the interface between prediction and execution in the data processing process, thereby strengthening the smooth transition from model output to actual operation, and enhancing the practicality and fault tolerance of the anti-malfunction scheme.

[0072] Step 302: If the output of the fault prediction model is different from the judgment result of the relay protector, the protection threshold set in the relay protector is obtained, the protection threshold represents the tripping condition set by the system during design to ensure the safety of the equipment, and the protection threshold of the relay protector is adjusted by judging the confidence, and the logic is as follows:

[0073] When the judgment result of the fault prediction model is tripping, the judgment result is calibrated as a constant 1, and when the judgment result of the fault prediction model is not tripping, the judgment result is calibrated as a constant 0, the third intermediate quantity is obtained by subtracting the constant 0.5 from the judgment result, the third intermediate quantity controls the adjustment direction of the threshold value, and when the judgment result is tripping, the protection threshold is lowered, and when the judgment result is not tripping, the protection threshold is raised, the fourth intermediate quantity is obtained by multiplying the third intermediate quantity and the confidence of the judgment result of the fault prediction model, the adjusted protection threshold is obtained by taking the negative of the intermediate quantity and multiplying it with the protection threshold before adjustment, and the adjustment amplitude of the threshold value is controlled by the fourth intermediate quantity, and the higher the confidence, the greater the adjustment amplitude;

[0074] As a preferred embodiment, the formula for adjusting the protection threshold is as follows:

[0075]

[0076] Wherein, is the adjusted protection threshold, is the protection threshold before adjustment, is the adjustment amplitude, is the judgment of the fault prediction model, when the judgment is tripping, , when the judgment is not tripping,

[0077] ​Specifically reflects the dynamic optimization of the relay protection under the assistance of the fault prediction model, which means that the adaptive fine-tuning of the threshold value is realized through the weighted correction of the confidence and the judgment deviation, and the boundary fuzzy problem of the ground fault detection in the real power environment is embodied, such as the threshold sensitivity disorder caused by high load interference or transient signal, thereby avoiding the over-conservative or aggressive response of the static threshold under variable working conditions. As a benchmark protection threshold, it represents the safety bottom line of system design, and captures the rigid requirements of the inherent protection logic of the device, The adjustment range quantifies the sensitivity of the correction, reflecting the control requirements of the threshold drift in engineering practice, The confidence of the judgment highlights the influence of model uncertainty on the reliability of the decision, and in a high-noise environment such as a mountainous power grid, It can further increase the determinability of ground fault identification, and As a binary judgment flag (trip is 1, no trip is 0), it reflects the conflict scenario between the prediction output and the traditional judgment. Closely related to these independent variables, the original threshold T is taken as an anchor point through a multiplicative structure, and a linear correction term is introduced to incorporate model insights, where the influence of has a directional stretching relationship: and positive amplification enhances the correction strength, while sign determines the adjustment direction ( negative to lower the threshold, promoting tripping; positive to raise the threshold, suppressing misoperation), which overall embodies the synergistic effect that the higher the confidence and the greater the deviation, the more significant the adjustment; in terms of size change relationship, when and increase, the correction term tends to negative value, making smaller than (threshold lowering, conducive to sensitive response), and vice versa , larger than (threshold rising, enhancing conservatism), the increase of linearly amplifies the adjustment range, and the whole dynamic ensures that monotonically deviates from with the confidence and the deviation of the judgment, this asymmetric response mechanism balances the safety margin and the risk of misoperation in practical application, and strengthens the accuracy and adaptability of misoperation prevention.

[0078] Meanwhile, a protection threshold adjustment upper limit is set, and the adjustment upper limit represents a relative change amount of the adjusted protection threshold relative to the protection threshold before adjustment.

[0079] The relay protection device uses the adjusted protection threshold to make a judgment on the power consumption data again, and performs a trip or no trip operation according to the judgment result.

[0080] By means of the confidence-driven protection threshold dynamic correction and upper limit constraint mechanism, the quantitative insights of the fault prediction model are integrated into the traditional threshold logic of the relay protection device, which not only realizes the targeted optimization of the difference judgment, but also prevents the extreme deviation caused by the threshold drift; such threshold fine-tuning based on probability confidence significantly improves the adaptability of the relay protection device to uncertain fault scenarios, such as pseudo-grounding signals under high load, reduces the misoperation rate and improves the accuracy of tripping, thereby ensuring the long-term safe and economic operation of the power equipment and avoiding the failure of the static threshold under variable working conditions; at the same time, it connects the inconsistent comparison in the previous step as the activation condition and transmits the adjustment upper limit to the state evaluation optimization, forming a feedback loop from prediction deviation to threshold iteration in the data processing process, thereby deepening the robustness and continuous learning potential of the scheme and comprehensively strengthening the dynamic defense capability of preventing misoperation of the relay protection device.

[0081] Step 4: Set a protection threshold adjustment upper limit when adjusting the protection threshold, obtain the historical usage record of the relay protection device and generate a state evaluation index, and adjust the protection threshold adjustment upper limit through the state evaluation index.

[0082] The step 4 includes the following contents:

[0083] Obtain the historical usage record of the relay protection device, and the usage record includes the running years, the rated service life, the number of trips performed, and the number of judgment errors, the number of judgment errors includes the number of times that should have tripped but did not trip and the number of times that should not have tripped but did trip, obtain the state evaluation index of the relay protection device, and the logic is as follows:

[0084] Form a fifth intermediate quantity by multiplying the ratio of the running years to the rated service life by the number of judgment errors, the more the running years or the number of judgment errors, the larger the fifth intermediate quantity, and the smaller the rated service life, the larger the fifth intermediate quantity, form a sixth intermediate quantity by taking the logarithm function of the ratio of the number of trips performed to the running years, the ratio of the number of trips performed to the running years as the logarithm, and the constant as the base, the larger the number of trips performed, the larger the sixth intermediate quantity, and the smaller the running years, the larger the sixth intermediate quantity, and the state evaluation index is obtained by assigning weights to the fifth intermediate quantity and the sixth intermediate quantity and summing them up.

[0085] As a preferred embodiment, the formula based on which the state evaluation index is obtained is as follows:

[0086]

[0087] wherein, represents a state evaluation index, and represents a weight, , , is a running year, represents a rated service life, is a number of times of tripping, is a number of times of misjudgment;

[0088] specifically reflects a comprehensive score of multi-dimensional health degradation of a relay protector, and its meaning is that the quantitative evaluation of equipment reliability level is integrated by weighting time aging and operation abnormality, and the accumulated fault hidden problem under long-term operation in a real power environment is embodied, such as component aging caused by prolonged running year or logic fatigue caused by frequent tripping, thereby avoiding the blind area of instant judgment which is away from historical data; the independent variables and as complementary weights capture the engineering trade-off of misjudgment priority, and reflect the amplified concern for misjudgment risk in safety design The ratio of reflects the aging progress, and highlights the reality of time-dependent performance degradation, as a misjudgment count reflects the accuracy defect of the protection mechanism, and the normalized logarithmic term quantifies the tripping density, representing the nonlinear stress of stress accumulation under high load working conditions. These independent variables are closely related, and the influence of the independent variable on is positively correlated: The linear sensitivity of accelerates the exponential effect of misjudgment contribution, and the logarithmic moderation mechanism inhibits the excessive punishment of extreme tripping, and the whole reflects the interactive amplification effect of aging and operation abnormality; in terms of size change relationship, as or increases, shows a linearly accelerated increase highlighting the degradation threshold, the increase of makes the logarithmic term gradually rise to saturation to ensure stability, and the increase of further stretches the misjudgment dominant curve, and the whole dynamic makes

[0089] monotonically climb to the threshold trigger point as the independent variables accumulate, and this progressive evaluation guides the dynamic tightening of the upper limit of the protection threshold adjustment in practical application, and strengthens the preventability and sustainability of the anti-misoperation.The state evaluation index threshold is set, and when the state evaluation index does not exceed the state evaluation index threshold, the state evaluation index does not adjust the protection threshold adjustment upper limit, and when the state evaluation index does not exceed the state evaluation index threshold, the state evaluation index adjusts the protection threshold adjustment upper limit, and the logic is as follows:

[0090] The product of the state evaluation index and the protection threshold upper limit before adjustment is taken as an adjustment factor, and the adjustment factor is added to the protection threshold upper limit before adjustment to obtain the protection threshold upper limit after adjustment. The larger the protection threshold upper limit before adjustment is, the larger the protection threshold upper limit after adjustment is, and the larger the state evaluation index is, the larger the protection threshold after adjustment is.

[0091] As a preferred embodiment, the adjustment is based on the following formula:

[0092]

[0093] wherein, represents the protection threshold adjustment upper limit after adjustment, represents the protection threshold adjustment upper limit before adjustment, is a weight, .

[0094] By mining the historical usage records of the relay protector, a state evaluation index is generated and the protection threshold adjustment upper limit is finely adjusted accordingly. In the data processing logic of the scheme, a closed-loop terminator of long-term feedback and short-term optimization is constructed, which not only quantifies the multi-factor degradation effects of device aging, trip frequency and error accumulation, but also realizes the balance between conservative and aggressive adjustment through threshold triggering mechanism. This history-driven state evaluation significantly prolongs the service life of the relay protector, reduces the maintenance demand induced by sudden failures, and adapts to the personalized needs of power grid load evolution, avoiding the failure risk of universal threshold on aging devices, improving the sustainability and cost-effectiveness of the overall system. At the same time, it takes the initial adjustment upper limit of the previous step as a dynamic benchmark and strengthens the knowledge cycle from real-time data to historical experience in the whole process, thereby consolidating the evolution and self-healing ability of the anti-malfunction scheme, ensuring the stable evolution of power protection in the whole life cycle.

[0095] The above formulas are dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation. The preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0096] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art can be aware that units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on the specific application and design constraints of the technical solutions.

[0097] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, and can be located in one place or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiments according to actual needs.

[0098] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method of preventing maloperation of a relay protection, characterized by, The specific steps include: Step 1: Obtain historical power consumption data, the power consumption data including current curve and voltage curve, set a sliding window and a sampling rate in the power consumption data, obtain features and zero sequence components from the power consumption data in the sliding window, form a sample set from the features and the zero sequence components, and associate the sample set with whether a relay protector performs a tripping action, the tripping action being a correct action after correction; Step 2: input the sample set as a feature input and the corresponding tripping action as a feature output, train in a gradient boosting tree model to obtain a fault prediction model, obtain real-time power consumption data and extract features and zero sequence components to input into the fault prediction model, obtain a judgment result of the fault prediction model for whether to trip, and simultaneously generate a probability of the judgment result being tripping through an output layer of the fault prediction model, and obtain a confidence of the judgment result through the probability; Step 3: input the real-time power consumption data into the relay protector and obtain a judgment result of the relay protector for whether to trip, compare the judgment result of the relay protector with the judgment result of the fault prediction model, if the two judgment results are the same, directly execute, if different, obtain a protection threshold of the relay protector, adjust the protection threshold through the confidence, generate a judgment result of the relay protector for whether to trip again according to the adjusted protection threshold, and execute according to the result; Step 4: set a protection threshold adjustment upper limit when adjusting the protection threshold, obtain a historical use record of the relay protector and generate a state evaluation index, the use record including a running year number, a rated use life, a tripping execution number and a judgment error number, and adjust the protection threshold adjustment upper limit through the state evaluation index.

2. The method of claim 1, wherein: All historical power consumption data is obtained through a secondary circuit, the power consumption data including voltage curves and current curves of A, B and C three phases, the same sliding window and sampling rate being set in the voltage curves and the current curves, and the start and end points of each sliding window being the same in the voltage curves and the current curves of A, B and C three phases, the length of the sliding window being greater than 5 phase change periods, and the sampling rate of each sliding window being greater than 1 kHz, and the zero sequence components corresponding to each sliding window being obtained through a symmetrical component method, the zero sequence components including a zero sequence complex vector and a zero sequence amplitude.

3. The method of claim 2, wherein: The current curves and the voltage curves of each sliding window are preprocessed, features are extracted from the voltage curves and the current curves of A, B and C three phases in each sliding window, the features including current mean value, voltage mean value, current main frequency amplitude, voltage main frequency amplitude, current high frequency energy ratio, voltage high frequency energy ratio and system load; The features and the zero sequence components in a single sliding window form a sample set, and the sample set logic is as follows: The elements in the sample set include current mean value, voltage mean value, voltage main frequency amplitude, current main frequency amplitude, current high frequency energy ratio, voltage high frequency energy ratio, system load, zero sequence complex vector and zero sequence amplitude of the zero sequence complex vector; A label is added to each sample set, the label being a tripping condition corresponding to the sliding window, when the label is tripping, recorded as 1, and when the label is not tripping, recorded as 0. All sample sets and corresponding labels are aggregated and normalized, and all sample sets are divided into a training set and a validation set in an 8:2 ratio.

4. The method of claim 3, wherein: The training set is input into a gradient boosting tree model for training to obtain a fault prediction model, which outputs a judgment on whether to trip and a probability of tripping.

5. The method of claim 4, wherein: Real-time power consumption data is collected through the secondary circuit, and features and zero sequence components are obtained according to a set sliding window. The features and zero sequence components are input into the fault prediction model to obtain a judgment on whether to trip and a probability of tripping. A judgment confidence is generated based on the probability of tripping, and the logic is as follows: A first intermediate quantity is formed by multiplying the probability of tripping by a probability of tripping logarithmic function, which takes the probability of tripping as the logarithm and a constant as the base. A second intermediate quantity is formed by multiplying the probability of not tripping by a probability of not tripping logarithmic function, which takes the probability of not tripping as the logarithm and a constant as the base. The probability of not tripping is equal to 1 minus the probability of tripping. The sum of the first intermediate quantity, the second intermediate quantity, and a first constant equals the judgment confidence. The closer the probability of tripping is to 1 and 0, the higher the judgment confidence. The closer the probability of tripping is to 0.5, the lower the judgment confidence.

6. The method of claim 5, wherein: The power consumption data of the secondary circuit is analyzed by a relay protector, and the judgment result of the relay protector is obtained. The judgment of the relay protector includes tripping and not tripping. The output of the fault prediction model is compared with the judgment result of the relay protector. If the results are the same, the judgment result of the relay protector is directly executed.

7. The method of claim 6, wherein: If the output of the fault prediction model is not the same as the judgment result of the relay protector, a protection threshold set in the relay protector is obtained. The protection threshold represents the tripping condition set by the system during design to ensure the safety of equipment. The protection threshold of the relay protector is adjusted based on the judgment confidence, and the logic is as follows: When the judgment result of the fault prediction model is tripping, the judgment result is calibrated to a constant 1. When the judgment result of the fault prediction model is not tripping, the judgment result is calibrated to a constant 0. A third intermediate quantity is obtained by subtracting a constant 0.5 from the judgment result. The third intermediate quantity controls the adjustment direction of the threshold. When the judgment result is tripping, the protection threshold is lowered. When the judgment result is not tripping, the protection threshold is increased. A fourth intermediate quantity is obtained by multiplying the third intermediate quantity and the confidence of the judgment result of the fault prediction model. The adjusted protection threshold is obtained by multiplying the negative of the intermediate quantity and the protection threshold before adjustment. The fourth intermediate quantity controls the adjustment amplitude of the threshold. The higher the confidence, the larger the adjustment amplitude. At the same time, an upper limit of the protection threshold adjustment is set, which represents the relative change of the adjusted protection threshold relative to the protection threshold before adjustment.

8. The method of claim 7, wherein: The relay protector uses the adjusted protection threshold to make a judgment on the power consumption data again, and executes tripping or not tripping operations based on the judgment result.

9. The method of claim 8, wherein: Obtain the historical use record of the relay protector, the use record includes the running years, the rated service life, the number of tripping and the number of misjudgment, the number of misjudgment includes the number of times that should trip but not trip and the number of times that should not trip but trip, obtain the state evaluation index of the relay protector, the logic is as follows: The fifth intermediate quantity is formed by the ratio of the running years to the rated service life multiplied by the number of misjudgment, the more the running years or the number of misjudgment, the larger the fifth intermediate quantity, the smaller the rated service life, the larger the fifth intermediate quantity, the sixth intermediate quantity is formed by the logarithmic function of the ratio of the number of tripping to the running years, the ratio of the number of tripping to the running years as the logarithm, the constant as the base, the larger the number of tripping, the larger the sixth intermediate quantity, the smaller the running years, the larger the sixth intermediate quantity, the state evaluation index is obtained by summing the fifth intermediate quantity and the sixth intermediate quantity respectively after assigning weights; Set the state evaluation index threshold value, when the state evaluation index does not exceed the state evaluation index threshold value, the state evaluation index does not adjust the upper limit of the protection threshold value, when the state evaluation index does not exceed the state evaluation index threshold value, the state evaluation index adjusts the upper limit of the protection threshold value, the logic is as follows: The product of the state evaluation index and the upper limit of the protection threshold value before adjustment is taken as the adjustment factor, the adjustment factor is added to the upper limit of the protection threshold value before adjustment to obtain the upper limit of the protection threshold value after adjustment, the larger the upper limit of the protection threshold value before adjustment, the larger the upper limit of the protection threshold value after adjustment, the larger the state evaluation index, the larger the protection threshold value.

Images