An intelligent power distribution system based on microprocessor and sensor array

Abstract

The application discloses an intelligent power distribution system based on a microprocessor and a sensor array, and particularly relates to the technical field of intelligent power distribution, and by means of a parallel double-path control strategy, when the current rising rate exceeds a threshold value, a current-limiting clamp is started to suppress fault energy, and meanwhile, a time window provided by the clamp is utilized to complete waveform feature matching, so that quick response and accurate identification are taken into account, the inductive passing of normal motor starting and the selective quick removal of real short circuits are realized, mis-tripping and missed protection are avoided, and the reliability and engineering practicability of the intelligent power distribution box in a hybrid load scene are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent power distribution technology, and more specifically, to an intelligent power distribution system based on a microprocessor and a sensor array. Background Technology

[0002] In existing intelligent power distribution technologies based on microprocessors and sensor arrays, short-circuit faults are typically protected at the microsecond level by detecting the rate of change of current (di / dt). However, when the power distribution circuit is connected to impulsive loads such as motors and compressors, the di / dt generated during their normal startup process can reach 200~400A / μs, blurring the boundary with the current change rate of a short-circuit arc (typically above 500A / μs), making it unreliable to distinguish them solely by the slope threshold. To avoid misjudging normal startup, a load identification algorithm is needed, utilizing millisecond-level startup waveform characteristics for judgment; however, microsecond-level protection requires the microprocessor to make a cut-off decision within 50μs. There is a fundamental contradiction in the time scale between the two, namely, the data window required by the identification algorithm is much longer than the protection action window. This physical incompatibility leads existing intelligent power distribution methods to either sacrifice rapid protection capabilities, resulting in frequent false trips under impulsive loads, or abandon load identification, raising the current change rate threshold and missing protection for real short circuits, severely restricting the engineering practicality of intelligent distribution boxes in mixed load scenarios. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent power distribution system based on a microprocessor and a sensor array to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A smart power distribution system based on a microprocessor and sensor array includes a joint waveform sampling construction module, an impact feature extraction module, a matching and suspicion calculation module, a parallel clamping control module, a discrimination and action execution module, and a feature library self-learning module. The joint waveform sampling construction module is used to obtain the transient current and transient voltage of each branch and construct a joint current and voltage waveform description set; The impact feature extraction module is used to extract the time-varying curve of current rise rate and voltage drop recovery features based on the current-voltage joint waveform description set, calculate the curvature of the leading edge of the current waveform, and form an impact waveform feature vector. The matching and suspicion calculation module is used to match the impact waveform feature vector with the load start waveform feature library pre-stored in the microprocessor, output the matching confidence, and calculate the transient short circuit suspicion index based on the matching residual. The parallel clamping control module is used to trigger parallel dual-path control when the peak value of the time-varying curve of the current rise rate is greater than a preset peak value threshold. The first path drives the semiconductor switch into a current-limiting clamping mode to limit the current to a set multiple of the rated value. The second path uses the time window provided by the current-limiting clamping to continue to acquire the complete waveform and complete the waveform feature matching. The discrimination and action execution module is used to determine the source of the impact based on the matching confidence level and the transient short circuit suspicion index: if it is determined to be a normal motor start-up, it gradually exits the current limiting clamp and restores full conduction; if it is determined to be a real short circuit fault, it drives the semiconductor switch to be completely turned off within a set time after the current limiting clamp is activated. The feature library self-learning module is used to update the load startup waveform feature library online by taking the feature vectors of the impact waveforms that are determined to be normal startups but whose mismatch rate exceeds a preset threshold as new samples.

[0005] In a preferred embodiment, the joint waveform sampling construction module is specifically used to: collect the transient current and transient voltage of each branch; The transient current discrete sequence and transient voltage discrete sequence obtained after acquisition are divided into several half-wave segments with adjacent zero crossings as boundaries. Each half-wave segment corresponds to a complete waveform from the rising edge of the current to the next zero crossing interval. Linear interpolation normalization is performed on the number of discrete points within each half-wave segment to ensure that all half-wave segments have the same length, resulting in normalized current waveforms and normalized voltage waveforms. The effective values ​​of current and voltage for each half-wave segment are calculated, and the normalized current waveform, the normalized voltage waveform, the effective current value, and the effective voltage value are combined into a data tuple. The data tuples of all half-wave segments are arranged in chronological order to form the current-voltage joint waveform description set.

[0006] In a preferred embodiment, the impact feature extraction module is specifically used to: extract the normalized current waveform and the normalized voltage waveform of the k-th half-wave segment from the current-voltage joint waveform description set; take the interval from the zero-crossing point to the first peak point of the current waveform within the half-wave segment as the current rising edge interval, determine the start index and peak index of the current rising edge interval, and extract the current rising edge subsequence; continuously fit the current rising edge subsequence with a third-order spline function to obtain a continuous time function; obtain the time-varying curve of the current rising rate by taking the first derivative of the continuous time function, and record the maximum value of the time-varying curve of the current rising rate.

[0007] In a preferred embodiment, the impact feature extraction module is further configured to: extract a voltage subsequence from the normalized voltage waveform that is within the same time interval as the current rising edge interval, calculate the minimum voltage value within the same time interval, and obtain the effective voltage value of the half-wave segment. The voltage drop depth is obtained; starting from the point of minimum voltage, the search proceeds backward to find the voltage value that first recovers to its minimum value. The index is used to calculate the recovery time; Based on the current rising edge subsequence, the curvature of the current waveform leading edge is calculated; The maximum value of the time-varying curve of the current rise rate, the voltage drop depth, the recovery time, and the curvature of the leading edge of the current waveform are combined to form the characteristic vector of the impact waveform. The process of obtaining the impact waveform feature vector is repeated for each half-wave segment in the current-voltage joint waveform description set to obtain the impact waveform feature vector sequence.

[0008] In a preferred embodiment, the matching and suspicion calculation module is specifically used to: extract the current impact waveform feature vector to be judged from the impact waveform feature vector sequence; the microprocessor has a pre-stored load start waveform feature library, including cluster center feature vectors of known normal load categories, and an accompanying allowable matching error covariance matrix of known normal load categories; Calculate the Mahalanobis distance between the impact waveform feature vector and the cluster center of the j-th class; take the minimum Mahalanobis distance among all known normal load categories and record the corresponding category index; The matching confidence and matching residuals are calculated based on the minimum Mahalanobis distance in the known normal load categories. Calculate the transient short-circuit suspicion index based on the matching residual; For each half-wave segment k in the current-voltage joint waveform description set, the process of obtaining the matching confidence and transient short-circuit suspicion index is repeated to obtain the matching confidence sequence and the transient short-circuit suspicion index sequence.

[0009] In a preferred embodiment, the first path drives the semiconductor switch into a current-limiting clamping mode to limit the current to a set multiple of the rated value. Specifically, the microprocessor sends a current-limiting clamping command to the driving circuit of the semiconductor switch connected in series with the branch. The driving circuit switches the operating mode of the semiconductor switch from full conduction to switch modulation mode, so that the branch current is limited to a set multiple of the rated current. In the current-limiting clamping mode, the microprocessor continuously monitors the branch current. If the branch current is greater than the product of the rated current and the set multiple, the microprocessor reduces the on-duty cycle of the semiconductor switch to precisely clamp the current within the neighborhood of the target value. At the same time, the microprocessor records the current-limiting clamping trigger time and the maximum current during the clamping period.

[0010] In a preferred embodiment, the second path utilizes the time window provided by the current-limiting clamp to continue acquiring the complete waveform and completing the waveform feature matching. Specifically, the microprocessor utilizes the time window provided by the current-limiting clamp mode to initiate the following sub-process: Continue to collect the transient current and transient voltage of the branch at the same sampling rate, and collect a complete half-wave segment from the current limiting clamp trigger moment to obtain the extended current discrete sequence and voltage discrete sequence. The extended current discrete sequence and voltage discrete sequence are spliced ​​together with the most recent unclamped half-wave segment stored in the current-voltage joint waveform description set before the trigger to form a complete impact waveform record. Based on the extraction logic of the impact waveform feature vector, the impact waveform feature vector is re-extracted from the complete impact waveform record, and it is matched with the load start waveform feature library to update the matching confidence and the transient short circuit suspicion index.

[0011] In a preferred embodiment, the discrimination and action execution module is specifically used for: the microprocessor reading the matching confidence score and the transient short-circuit suspicion index from shared memory; and setting a first trust threshold and a second suspicion threshold; the microprocessor executing the following discrimination logic: If the matching confidence level is greater than the first trust threshold and the transient short circuit suspicion index is less than the second suspicion threshold, then the current impact source is determined to be normal motor start-up; otherwise, it is determined to be a real short circuit fault. If the motor start-up is determined to be normal, the microprocessor controls the first path to gradually exit the current-limiting clamping mode. Specifically, the target current-limiting multiple of the semiconductor switch is increased from the current value in fixed steps. After each step, the microprocessor re-samples the branch current. If the current at three consecutive sampling points does not exceed the updated current-limiting target, the increment continues. If the current value reaches 1.0, the microprocessor sends a full-conduction command to the drive circuit to restore the semiconductor switch to the zero-dropout conduction state and clear the current-limiting clamping flag. If, at any time during the gradual exit process, the current is detected to exceed the current current-limiting target again and the duration exceeds [a certain threshold], the microprocessor will proceed with the exit. If the exit process is stopped immediately and the current limiting clamp mode is re-entered, the matching confidence is forcibly set to 0, and the process is transferred to the real short circuit fault determination process. If a genuine short-circuit fault is determined, the microprocessor starts a timer and sets a complete shutdown delay. During the complete shutdown delay, the first path continues to maintain the current-limiting clamping mode. When the timer reaches the complete shutdown delay, the microprocessor sends a complete shutdown command to the drive circuit, pulling the gate voltage of the semiconductor switch to a negative voltage, causing the branch current to drop to zero within 1μs. After complete shutdown, the microprocessor latches the fault state, records the matching confidence, the transient short-circuit suspicion index, and the impulse waveform feature vector of the last complete waveform before shutdown, and reports the genuine short-circuit fault event. During the fault latching period, the microprocessor prohibits any reclosing operation on the same branch until an external reset command is received.

[0012] In a preferred embodiment, the feature library self-learning module is specifically used for: the microprocessor maintaining a mismatch counter. The initial value is 0, and the most recent value is recorded. The microprocessor increments the mismatch counter for each event determined to be a normal startup. If the match confidence of an event determined to be a normal startup is less than a preset low confidence threshold, the microprocessor increments the mismatch counter. The characteristic vector of the impact waveform is stored in a temporary buffer, and the mismatch rate is defined. When the microprocessor completes the cumulative process After a normal startup is determined, the mismatch rate is calculated and compared with a preset threshold; if the mismatch rate is greater than the preset threshold, the online update of the load startup waveform feature library is triggered.

[0013] In a preferred embodiment, the feature library self-learning module is further used for: Retrieve all from the temporary cache area The feature vectors of the mismatched impact waveforms are used to obtain the set of mismatched feature vectors; An unsupervised clustering algorithm is used to divide the mismatched feature vector set into... A new cluster; For each new cluster, calculate its cluster center and covariance matrix; The cluster centers and covariance matrix of the new clusters are added to the load start waveform feature library; at the same time, the microprocessor performs an aging and elimination mechanism on the existing categories. After the update is completed, the mismatch counter is reset to zero, the temporary cache is cleared, and the timestamp of this update is recorded. The microprocessor sends a load start waveform feature library update report to the cloud or operation and maintenance terminal, which includes the number of new categories, the number of samples for each new category, and the number of old categories that have been eliminated.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention aims to solve the time scale contradiction between microsecond-level short-circuit protection and millisecond-level load identification in existing intelligent power distribution technology. By adopting a parallel dual-path control strategy, when the current rise rate exceeds the threshold, current limiting clamping is first activated to suppress fault energy. At the same time, the time window provided by the clamping is used to complete waveform feature matching, thereby balancing fast response and accurate identification. This achieves seamless passage of normal motor startup and selective and rapid disconnection of real short circuits, avoiding false tripping and leakage protection, and significantly improving the reliability and engineering practicality of intelligent power distribution boxes in mixed load scenarios. Attached Figure Description

[0015] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart of the system according to an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example: The present invention provides, as follows Figure 1 The intelligent power distribution system shown includes a joint waveform sampling construction module, an impact feature extraction module, a matching and suspicion calculation module, a parallel clamping control module, a discrimination and action execution module, and a feature library self-learning module. The joint waveform sampling construction module is used to obtain the transient current and transient voltage of each branch and construct a joint current and voltage waveform description set; The impact feature extraction module is used to extract the time-varying curve of current rise rate and voltage drop recovery features based on the current-voltage joint waveform description set, calculate the curvature of the leading edge of the current waveform, and form an impact waveform feature vector. The matching and suspicion calculation module is used to match the impact waveform feature vector with the load start waveform feature library pre-stored in the microprocessor, output the matching confidence, and calculate the transient short circuit suspicion index based on the matching residual. The parallel clamping control module is used to trigger parallel dual-path control when the peak value of the time-varying curve of the current rise rate is greater than a preset peak value threshold. The first path drives the semiconductor switch into a current-limiting clamping mode to limit the current to a set multiple of the rated value. The second path uses the time window provided by the current-limiting clamping to continue to acquire the complete waveform and complete the waveform feature matching. The discrimination and action execution module is used to determine the source of the impact based on the matching confidence level and the transient short circuit suspicion index: if it is determined to be a normal motor start-up, it gradually exits the current limiting clamp and restores full conduction; if it is determined to be a real short circuit fault, it drives the semiconductor switch to be completely turned off within a set time after the current limiting clamp is activated. The feature library self-learning module is used to update the load startup waveform feature library online by taking the feature vectors of the impact waveforms that are determined to be normal startups but whose mismatch rate exceeds a preset threshold as new samples.

[0018] In this embodiment of the invention, the process of obtaining the transient current and transient voltage of each branch and constructing a joint current-voltage waveform description set is as follows: Collect the transient current of each branch. With the transient voltage Sampling rate Set to at least 200kHz and not less than 20 times the expected maximum current rise time width of the branch, and use a multi-channel synchronous analog-to-digital converter to sample each branch in parallel, so that the sampling time offset between the current channel and the voltage channel of the same branch is less than 1μs. Discrete sequence of transient current obtained after acquisition With transient voltage discrete sequence The continuous waveform is divided into several half-wave segments with adjacent zero-crossing points as boundaries. Each half-wave segment corresponds to a complete waveform from the rising edge of the current to the next zero-crossing interval. The zero-crossing point refers to the point where the amplitude of an AC voltage or current waveform changes from positive to negative or from negative to positive and crosses zero. In a complete cycle of a power frequency AC current, the waveform will cross zero twice (rising zero-crossing point and falling zero-crossing point). The adjacent zero-crossing points refer to two consecutive zero-crossing points in time, such as a rising zero-crossing point and its immediate following falling zero-crossing point, or a falling zero-crossing point and its immediate following rising zero-crossing point. Using the adjacent zero-crossing points as boundaries, the continuous waveform can be divided into several independent half-wave segments. Each half-wave segment corresponds to a complete interval of the waveform starting from a zero-crossing point, passing through the peak value, and returning to the next zero-crossing point. The number of discrete points within each half-wave segment is linearly interpolated and normalized to ensure that all half-wave segments have the same length. The normalized current waveform is obtained. With normalized voltage waveform ,in This refers to the half-wave segment number. ; Calculate the effective current value for each half-wave segment. With voltage RMS value and the normalized current waveform The normalized voltage waveform The effective value of the current and the effective value of the voltage Combined into a single data tuple; the data tuples of all the half-wave segments are arranged in chronological order to form the current-voltage joint waveform description set. ,in The description set represents the total number of continuously sampled half-wave segments and is used for the extraction of impact waveform feature vectors in subsequent steps.

[0019] In this embodiment of the invention, the process of extracting the time-varying curve of the current rise rate and the voltage drop recovery feature based on the current-voltage joint waveform description set, calculating the curvature of the current waveform leading edge, and forming the impact waveform feature vector is as follows: The normalized current waveform of the k-th half-wave segment is extracted from the current-voltage joint waveform description set. With the normalized voltage waveform ,in The normalized discrete-time index is used; the interval from the zero-crossing point to the first peak point of the current waveform within the half-wave segment is taken as the current rising edge interval, and the starting index of the current rising edge interval is determined. and peak index Extract the current rising edge subsequence ,in , The current rising edge subsequence is continuously fitted using a third-order spline function to obtain a continuous-time function. , ,in , For sampling interval; The time-varying curve of the current rise rate is obtained by taking the first derivative. And record the maximum value of the time-varying curve of the current rise rate. ; The current rising edge subsequence is continuously fitted using a third-order spline function to obtain a continuous-time function, specifically: Suppose that the current rising edge subsequence includes There are discrete sampling points, and the discrete index of each point is: The corresponding time coordinate is ,in The sampling interval is denoted as . ; in the interval Constructing a third-order spline function The function is required to satisfy the following condition: in each subinterval superior, a cubic polynomial ,in , , , The coefficients are undetermined; the function has continuous first and second derivatives over the entire interval, i.e. and for It holds true; simultaneously, using natural boundary conditions, the second derivatives at both ends are set to zero, i.e. and By solving the system of linear equations consisting of the continuity and boundary conditions described above, the coefficients on all subintervals can be obtained. , , , By piecing together the cubic polynomials of each subinterval, we obtain the polynomial defined in... Continuous-time function on The function passes precisely through the discrete sampling points. Furthermore, adjacent subintervals are smoothly connected, which can be used to subsequently calculate the first and second derivatives; Simultaneously, from the normalized voltage waveform Extract the voltage subsequence within the same time interval as the current rising edge interval. Calculate the minimum voltage value within the same time interval. And obtain the effective voltage value of the half-wave segment. The voltage drop depth is obtained. Starting from the point of minimum voltage, the search proceeds backward to find the voltage value that initially recovers to its minimum value. index Calculate recovery time ,in The index corresponding to the minimum voltage value; Based on the current rising edge subsequence Calculate the curvature of the leading edge of the current waveform The formula is: in From The maximum value extracted from the second derivative. The maximum value of the time-varying curve of the current rise rate; The maximum value of the time-varying curve of the current rise rate The voltage drop depth The recovery time and the curvature of the leading edge of the current waveform Combined, the impact waveform feature vector is formed: ; For each half-wave segment of the current-voltage joint waveform description set The process of obtaining the characteristic vectors of the repeated impact waveform is repeated to obtain the sequence of characteristic vectors of the impact waveform. This is used for subsequent matching with the pre-stored feature library.

[0020] In this embodiment of the invention, the process of matching the impact waveform feature vector with the pre-stored load start-up waveform feature library in the microprocessor, outputting the matching confidence score, and calculating the transient short-circuit suspicion index based on the matching residual is as follows: From the shock waveform feature vector sequence Extract the feature vector of the current impact waveform to be determined. The microprocessor pre-stores the load start waveform feature library. ,in For the first Given the cluster center feature vectors of normal loads (including motors, compressors, transformers, etc.), each... With Same dimensions And it includes the allowable matching error covariance matrix for this type of load. ,in This represents the typical maximum value of the time-varying current rise rate curve during the startup process of a type j normal load (such as a motor or compressor). This represents the typical voltage drop depth of the bus during startup of a type j normal load. After starting up with a normal load of type j, the voltage recovers from its minimum value to... Typical recovery time required, The typical curvature of the leading edge of the j-th type of normal load current waveform; Calculate the eigenvector of the impact waveform Mahalanobis distance between the cluster centers of the j-th cluster and the cluster center The formula is: ,in Let be the inverse matrix of the allowed matching error covariance matrix; take all Minimum value of middle Mahalanobis distance And record the corresponding category index. ; the matching confidence Defined as: ,in The closer the value is This indicates that the current waveform closely matches a certain type of known normal load; Calculate the matching residual , that is, the difference vector between the current feature vector and the best matching cluster center; Calculate the transient short-circuit suspicion index based on the matching residual. The formula is: ,in and The preset weighting coefficients satisfy... , , The first item The first item reflects the overall degree of waveform mismatch, and the second item is the relative excess ratio of the current rise rate exceeding the typical value of this type of load (if it does not exceed the typical value, then take 0). For each half-wave segment k in the current-voltage joint waveform description set, the process of obtaining the matching confidence score and transient short-circuit suspicion index is repeated to obtain the matching confidence score sequence. With transient short-circuit suspected index sequence This is for subsequent steps to make dual-path decisions.

[0021] In this embodiment of the invention, when the peak value of the time-varying curve of the current rise rate is greater than a preset peak threshold, the microprocessor triggers parallel dual-path control: the first path drives the semiconductor switch into a current-limiting clamping mode to limit the current to a set multiple of the rated value; the second path utilizes the time window provided by the current-limiting clamping to continue acquiring the complete waveform and completing the waveform feature matching process as follows: When the peak value of the time-varying curve of the current rise rate exceeds a preset peak threshold, the microprocessor triggers parallel dual-path control: the first path drives the semiconductor switch into a current-limiting clamping mode, limiting the current to a set multiple of the rated value; the second path utilizes the time window provided by the current-limiting clamping to continue acquiring the complete waveform and completing the waveform feature matching, specifically: The microprocessor monitors the time-varying curve of the current rise rate in real time. Extracted current peak value The peak value is then compared with a pre-stored peak threshold. The range of values ​​is The specific value depends on the parasitic inductance of the line where the distribution box is located. With rated voltage According to the formula Pre-calibrated; when the peak value of the time-varying curve of the current rise rate exceeds a preset peak threshold, the microprocessor immediately triggers the parallel dual-path control: First path: The microprocessor sends a current-limiting clamping command to the driving circuit of the semiconductor switch (the semiconductor switch is a SiCMOSFET or GaN HEMT) connected in series with the branch. The driving circuit switches the operating mode of the semiconductor switch from full conduction to the linear region or switching modulation mode, thereby limiting the current in the branch. Limited to rated current The set multiple The set multiple The range of values ​​is The specific values ​​are pre-configured based on the short-term overload tolerance capability of the load at the downstream end of the branch; in current-limiting clamping mode, the microprocessor continuously monitors the branch current. If the current is greater than Then, further reduce the duty cycle or channel opening of the semiconductor switch to precisely clamp the current within the neighborhood of the target value, and simultaneously record the current-limiting clamping trigger time. and the maximum current during clamping ; Second path: The microprocessor utilizes the time window provided by the current limiting clamping mode (the duration of the time window is preset by the microprocessor, and the value range is...). (its length is sufficient to cover the time required for subsequent waveform acquisition and matching calculations), and the following sub-process is initiated: Continue at the same sampling rate Collect the transient current of the branch. With transient voltage From the current limiting clamp trigger time At least one complete half-wave segment is acquired from the beginning to obtain the extended discrete current sequence. With voltage discrete sequence ,in Corresponding time interval ; The extended current discrete sequence and voltage discrete sequence are combined with the current-voltage joint waveform description set. The most recent unclamped half-wave segment stored before the trigger is spliced ​​together to form a complete impact waveform record. ; Based on the extraction logic of the impact waveform feature vector, from the complete impact waveform record The shock waveform feature vector is re-extracted in the middle. And match the waveform features with the load start waveform feature library, and update the matching confidence and the transient short circuit suspicion index; The first path and the second path are executed in parallel by two independent hardware threads of the microprocessor, and time synchronization between them is achieved through status flags in shared memory; after the second path completes waveform feature matching calculation, the updated matching confidence and transient short-circuit suspicion index are written into the shared memory; if within the time window If the second path fails to complete waveform feature matching (e.g., due to timeout caused by computational resource conflicts), the microprocessor forcibly sets the matching confidence to zero and sets the transient short-circuit suspicion index to a preset maximum value. (The maximum value of the transient short-circuit suspicion index) (The value is 10.0) to implement a conservative short-circuit protection strategy.

[0022] In this embodiment of the invention, the source of the impact is determined based on the matching confidence level and the transient short-circuit suspicion index: if it is determined to be a normal motor start-up, the current limiting clamp is gradually released and full conduction is restored; if it is determined to be a real short-circuit fault, the process of driving the semiconductor switch to be completely turned off within a set time after the current limiting clamp is activated is as follows: The microprocessor reads the matching confidence score and the transient short-circuit suspicion index from shared memory; and sets a first trust threshold. With the second suspicion threshold The first trust threshold The range of values ​​is The second suspicion threshold The range of values ​​is The microprocessor executes the following discrimination logic: If the matching confidence level is greater than the first trust threshold and the transient short circuit suspicion index is less than the second suspicion threshold, then the current impact source is determined to be normal motor start-up; otherwise, it is determined to be a real short circuit fault. Scenario 1: If the motor is determined to be starting normally The microprocessor controls the first path to gradually exit the current-limiting clamping mode, specifically by: reducing the target current-limiting multiple of the semiconductor switch. From the current value According to fixed step size Increment, the step size The range of values ​​is The time interval between each step is The The range of values ​​is After each step is executed, the microprocessor re-acquires the branch current. If the current at three consecutive sampling points does not exceed the updated current limit target If the current value is zero, then continue to increment; if the current value is zero... When the current reaches 1.0, the microprocessor sends a full-conduction command to the driving circuit to restore the semiconductor switch to the zero-dropout conduction state and clear the current-limiting clamp flag; if the current is detected to exceed the current-limiting target again at any time during the gradual exit process and the duration exceeds [a certain value], [the microprocessor will then proceed accordingly]. If the exit process is stopped immediately and the current limiting clamp mode is re-entered, the matching confidence is forcibly set to 0, and the process is transferred to the real short circuit fault determination process. Scenario 2: Determined to be a genuine short-circuit fault The microprocessor starts a timer and sets a complete shutdown delay. The The range of values ​​is The The specific values ​​are pre-calibrated based on the safe operating area of ​​the semiconductor switch and the parasitic inductance of the line; during the complete turn-off delay During the specified time, the first path continues to maintain current-limiting clamping mode to limit the short-circuit current amplitude and prevent damage to the preceding circuitry; when the timer reaches the complete turn-off delay... At that time, the microprocessor sends a complete turn-off command to the driving circuit, pulling the gate voltage of the semiconductor switch to a negative voltage (for SiC MOSFET, the gate voltage is pulled to -4 V to -6 V; for GaN HEMT, the gate voltage is pulled to -5 V to -8 V), causing the branch current to drop to zero within 1 μs; after complete turn-off, the microprocessor latches the fault state and records the matching confidence, the transient short-circuit suspicion index, and the impulse waveform feature vector of the last complete waveform before turn-off. The microprocessor is prohibited from performing any reclosing operation on the same branch during the fault latching period until an external reset command is received or the fault is manually cleared. The discrimination logic and execution process are completed by a high-priority interrupt service routine of the microprocessor, ensuring that the total delay from reading the matching confidence and transient short-circuit suspicion index to execution exit or shutdown does not exceed 10μs.

[0023] In this embodiment of the invention, the process of updating the load startup waveform feature library online by using the impact waveform feature vectors that are determined to be normal startups but have a mismatch rate exceeding a preset threshold as new samples is as follows: The microprocessor maintains a mismatch counter. The initial value is 0, and the most recent value is recorded. The matching confidence level for each event that is determined to be a normal start-up. ,in The range of values ​​is For the first If an event is determined to be a normal startup event, and its matching confidence level is less than the preset low confidence threshold, then... When, the range of values ​​is The microprocessor then increments the mismatch counter. ,Right now The characteristic vector of the impact waveform is then stored in a temporary buffer. ; Define the mismatch rate for: ; When the microprocessor completes the cumulative process After the first normal startup determination, the false matching rate is calculated. and compare it with a preset threshold The comparison, the The range of values ​​is If the mismatch rate exceeds a preset threshold, the online update of the load start waveform feature library is triggered. The online update specifically refers to: The first step is to retrieve data from the temporary cache. Extract all The mismatched impact waveform feature vectors are used to obtain the mismatched feature vector set. ; The second step is to apply an unsupervised clustering algorithm to the set of mismatched feature vectors (the unsupervised clustering algorithm is density-based DBSCAN or K-means++, where K-means++ has a certain number of classes). The profile coefficient is automatically determined, and its value range is [value range missing]. The mismatch feature vector set is divided into... A new cluster; The third step is to process each new cluster. Calculate its cluster centers With covariance matrix The formula is: in For the first The number of samples contained in a new cluster. This is the nth mismatch feature vector belonging to this cluster; The fourth step is to determine the cluster centers of the new clusters. With covariance matrix The feature is added to the load-initiating waveform feature library; simultaneously, to prevent the feature library from expanding indefinitely, the microprocessor performs an aging and elimination mechanism for existing categories: for each existing category... Record the timestamp of its most recent successful match. If the current time and The difference exceeds the preset elimination cycle. (the aforementioned) If the value range is 30 days to 90 days, then this type and its covariance matrix are deleted from the load start waveform feature library. Fifth, after the update is complete, reset the mismatch counter. The system is reset to zero, the temporary cache is cleared, and the timestamp of this update is recorded. The microprocessor sends a load start waveform feature library update report to the cloud or operation and maintenance terminal, which includes the number of new categories, the number of samples for each new category, and the number of old categories that have been removed.

[0024] The online update process is performed during the microprocessor's idle period (e.g., when the load current is less than 5% of the rated value and the continuous duration exceeds 10 seconds) to avoid interfering with the real-time protection task.

[0025] This invention fundamentally resolves the time scale contradiction between microsecond-level short-circuit protection and millisecond-level load identification through parallel dual-path control and current-limiting clamping technology. Specifically, when the peak current rise rate exceeds the threshold, the first path does not immediately cut off the circuit but drives the semiconductor switch into current-limiting clamping mode, limiting the current to a fixed multiple of the rated value. This curbs the spread of fault energy within microseconds, providing a millisecond-level time window for the identification algorithm. Simultaneously, the second path utilizes this window to continue acquiring the complete waveform and performing feature matching, enabling the previously mutually exclusive fast response and accurate identification to be achieved in parallel. This reliably distinguishes between normal motor startup and actual short-circuit faults. For normal startup, full conduction is gradually restored after current-limiting clamping, avoiding the short-circuit protection failures or over-protection issues caused by blindly raising the threshold in traditional solutions. The invention addresses the problem of frequent false tripping due to blind and rapid disconnection. For genuine short circuits, it completely shuts off the semiconductor switch within 50-100μs after the current-limiting clamp is activated, maintaining a protection speed superior to traditional mechanical circuit breakers. By introducing multi-dimensional features such as the curvature of the current waveform leading edge, voltage drop depth, and recovery time, and combining Mahalanobis distance matching and transient short-circuit suspicion index calculation, the robustness of load identification is significantly improved. Furthermore, the online feature library update mechanism based on the false matching rate enables the system to adaptively learn the waveform characteristics of unknown or novel impulsive loads, resulting in a continuous reduction in the false alarm rate over long-term operation. While maintaining microsecond-level fast protection capabilities, this invention achieves seamless passage of impulsive loads and selective disconnection of genuine short circuits, greatly enhancing the engineering practicality and power supply reliability of intelligent distribution boxes in mixed load scenarios.

[0026] This invention aims to solve the time scale contradiction between microsecond-level short-circuit protection and millisecond-level load identification in existing intelligent power distribution technology. By adopting a parallel dual-path control strategy, when the current rise rate exceeds the threshold, current limiting clamping is first activated to suppress fault energy. At the same time, the time window provided by the clamping is used to complete waveform feature matching, thereby balancing fast response and accurate identification. This achieves seamless passage of normal motor startup and selective and rapid disconnection of real short circuits, avoiding false tripping and leakage protection, and significantly improving the reliability and engineering practicality of intelligent power distribution boxes in mixed load scenarios.

[0027] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0028] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0029] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0030] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0031] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent power distribution system based on a microprocessor and sensor array, characterized in that: It includes a joint waveform sampling construction module, an impact feature extraction module, a matching and suspect calculation module, a parallel clamping control module, a discrimination and action execution module, and a feature library self-learning module; The joint waveform sampling construction module is used to obtain the transient current and transient voltage of each branch and construct a joint current and voltage waveform description set; The impact feature extraction module is used to extract the time-varying curve of current rise rate and voltage drop recovery features based on the current-voltage joint waveform description set, calculate the curvature of the leading edge of the current waveform, and form an impact waveform feature vector. The matching and suspicion calculation module is used to match the impact waveform feature vector with the load start waveform feature library pre-stored in the microprocessor, output the matching confidence, and calculate the transient short circuit suspicion index based on the matching residual. The parallel clamping control module is used to trigger parallel dual-path control when the peak value of the time-varying curve of the current rise rate is greater than a preset peak value threshold. The first path drives the semiconductor switch into a current-limiting clamping mode to limit the current to a set multiple of the rated value. The second path uses the time window provided by the current-limiting clamping to continue to acquire the complete waveform and complete the waveform feature matching. The discrimination and action execution module is used to determine the source of the impact based on the matching confidence level and the transient short circuit suspicion index: if it is determined to be a normal motor start-up, it gradually exits the current limiting clamp and restores full conduction; If the fault is determined to be a real short circuit, the semiconductor switch will be driven to turn off completely within a set time after the current limiting clamp is activated. The feature library self-learning module is used to update the load startup waveform feature library online by taking the feature vectors of the impact waveforms that are determined to be normal startups but whose mismatch rate exceeds a preset threshold as new samples.

2. The intelligent power distribution system based on a microprocessor and sensor array according to claim 1, characterized in that: The joint waveform sampling construction module is specifically used to: collect the transient current and transient voltage of each branch; The transient current discrete sequence and transient voltage discrete sequence obtained after acquisition are divided into several half-wave segments with adjacent zero crossings as boundaries. Each half-wave segment corresponds to a complete waveform from the rising edge of the current to the next zero crossing interval. Linear interpolation normalization is performed on the number of discrete points in each half-wave segment to make all half-wave segments have the same length, resulting in normalized current waveforms and normalized voltage waveforms. Calculate the RMS current and RMS voltage values ​​for each half-wave segment, and combine the normalized current waveform, the normalized voltage waveform, the RMS current value, and the RMS voltage value into a data tuple; arrange the data tuples of all half-wave segments in chronological order to form the current-voltage joint waveform description set.

3. The intelligent power distribution system based on a microprocessor and sensor array according to claim 2, characterized in that: The impact feature extraction module is specifically used for: extracting the normalized current waveform and the normalized voltage waveform of the k-th half-wave segment from the current-voltage joint waveform description set; taking the interval from the zero-crossing point to the first peak point of the current waveform within the half-wave segment as the current rising edge interval, determining the starting index and peak index of the current rising edge interval, and extracting the current rising edge subsequence; continuously fitting the current rising edge subsequence with a third-order spline function to obtain a continuous time function; obtaining the time-varying curve of the current rising rate by taking the first derivative of the continuous time function, and recording the maximum value of the time-varying curve of the current rising rate.

4. The intelligent power distribution system based on a microprocessor and sensor array according to claim 3, characterized in that: The impact feature extraction module is further configured to: extract voltage subsequences within the same time interval as the current rising edge interval from the normalized voltage waveform, calculate the minimum voltage value within the same time interval, and obtain the effective voltage value of the half-wave segment. The voltage drop depth is obtained; starting from the point of minimum voltage, the search proceeds backward to find the voltage value that first recovers to its minimum value. The index is used to calculate the recovery time; Based on the current rising edge subsequence, the curvature of the current waveform leading edge is calculated; The maximum value of the time-varying curve of the current rise rate, the voltage drop depth, the recovery time, and the curvature of the leading edge of the current waveform are combined to form the characteristic vector of the impact waveform. The process of obtaining the impact waveform feature vector is repeated for each half-wave segment in the current-voltage joint waveform description set to obtain the impact waveform feature vector sequence.

5. The intelligent power distribution system based on a microprocessor and sensor array according to claim 1, characterized in that: The matching and suspicion calculation module is specifically used to: extract the current impact waveform feature vector to be judged from the impact waveform feature vector sequence; the microprocessor has a pre-stored load start waveform feature library, including cluster center feature vectors of known normal load categories, and an accompanying allowable matching error covariance matrix of known normal load categories; Calculate the Mahalanobis distance between the impact waveform feature vector and the cluster center of the j-th class; take the minimum Mahalanobis distance among all known normal load categories and record the corresponding category index; The matching confidence and matching residuals are calculated based on the minimum Mahalanobis distance in the known normal load categories. Calculate the transient short-circuit suspicion index based on the matching residual; For each half-wave segment k in the current-voltage joint waveform description set, the process of obtaining the matching confidence and transient short-circuit suspicion index is repeated to obtain the matching confidence sequence and the transient short-circuit suspicion index sequence.

6. The intelligent power distribution system based on a microprocessor and sensor array according to claim 1, characterized in that: The first path drives the semiconductor switch into a current-limiting clamping mode, limiting the current to a set multiple of the rated value. Specifically, the microprocessor sends a current-limiting clamping command to the driving circuit of the semiconductor switch connected in series with the branch. The driving circuit switches the operating mode of the semiconductor switch from full conduction to switch modulation mode, so that the branch current is limited to a set multiple of the rated current. In the current-limiting clamping mode, the microprocessor continuously monitors the branch current. If the branch current is greater than the product of the rated current and the set multiple, the microprocessor reduces the conduction duty cycle of the semiconductor switch to precisely clamp the current within the neighborhood of the target value. At the same time, the microprocessor records the current-limiting clamping trigger time and the maximum current during the clamping period.

7. The intelligent power distribution system based on a microprocessor and sensor array according to claim 1, characterized in that: The second path utilizes the time window provided by the current limiting clamp to continue acquiring the complete waveform and completing the waveform feature matching. Specifically, the microprocessor uses the time window provided by the current limiting clamp mode to initiate the following sub-processes: Continue to collect the transient current and transient voltage of the branch at the same sampling rate, and collect a complete half-wave segment from the current limiting clamp trigger moment to obtain the extended current discrete sequence and voltage discrete sequence. The extended current discrete sequence and voltage discrete sequence are spliced ​​together with the most recent unclamped half-wave segment stored in the current-voltage joint waveform description set before the trigger to form a complete impact waveform record. Based on the extraction logic of the impact waveform feature vector, the impact waveform feature vector is re-extracted from the complete impact waveform record, and it is matched with the load start waveform feature library to update the matching confidence and the transient short circuit suspicion index.

8. The intelligent power distribution system based on a microprocessor and sensor array according to claim 1, characterized in that: The discrimination and action execution module is specifically used for: the microprocessor reading the matching confidence score and the transient short-circuit suspicion index from shared memory; setting a first trust threshold and a second suspicion threshold; and the microprocessor executing the following discrimination logic: If the matching confidence level is greater than the first trust threshold and the transient short circuit suspicion index is less than the second suspicion threshold, then the current impact source is determined to be normal motor start-up; otherwise, it is determined to be a real short circuit fault. If the motor is determined to be starting normally, the microprocessor controls the first path to gradually exit the current limiting clamping mode, specifically by increasing the target current limiting multiple of the semiconductor switch from the current value in fixed steps. After each step is executed, the microprocessor re-collects the branch current. If the current at three consecutive sampling points does not exceed the updated current limit target, it continues to increase. If the current value reaches 1.0, the microprocessor sends a full conduction command to the driving circuit to restore the semiconductor switch to the zero-dropout conduction state and clear the current limiting clamp flag; If, at any point during the gradual withdrawal process, the current is detected to exceed the current limiting target again and the duration exceeds [a certain value], [the following applies]. If the exit process is stopped immediately and the current limiting clamp mode is re-entered, the matching confidence is forcibly set to 0, and the process is transferred to the real short circuit fault determination process. If a genuine short-circuit fault is determined, the microprocessor starts a timer and sets a complete shutdown delay. During the complete shutdown delay, the first path continues to maintain the current-limiting clamping mode. When the timer reaches the complete shutdown delay, the microprocessor sends a complete shutdown command to the drive circuit, pulling the gate voltage of the semiconductor switch to a negative voltage, causing the branch current to drop to zero within 1μs. After complete shutdown, the microprocessor latches the fault state, records the matching confidence, the transient short-circuit suspicion index, and the impulse waveform feature vector of the last complete waveform before shutdown, and reports the genuine short-circuit fault event. During the fault latching period, the microprocessor prohibits any reclosing operation on the same branch until an external reset command is received.

9. The intelligent power distribution system based on a microprocessor and sensor array according to claim 1, characterized in that: The feature library self-learning module is specifically used for: the microprocessor to maintain a mismatch counter. The initial value is 0, and the most recent value is recorded. The matching confidence level for each event that is determined to be a normal start-up; For events determined to be normal startups, if their matching confidence level is less than a preset low confidence threshold, the microprocessor increments the mismatch counter. And store the feature vector of the impact waveform in a temporary buffer; And define the false match rate; When the microprocessor completes the cumulative process After a normal startup is determined, the false match rate is calculated and compared with a preset threshold. If the mismatch rate exceeds a preset threshold, the online update of the load start waveform feature library is triggered.

10. The intelligent power distribution system based on a microprocessor and sensor array according to claim 9, characterized in that: The feature library self-learning module is also specifically used for: Retrieve all from the temporary cache area The feature vectors of the mismatched impact waveforms are used to obtain the set of mismatched feature vectors; An unsupervised clustering algorithm is used to divide the mismatched feature vector set into... A new cluster; For each new cluster, calculate its cluster center and covariance matrix; The cluster centers and covariance matrix of the new clusters are added to the load start waveform feature library; at the same time, the microprocessor performs an aging and elimination mechanism on the existing categories. After the update is complete, the mismatch counter is reset to zero, the temporary cache is cleared, and the timestamp of this update is recorded. The microprocessor sends a load start waveform feature library update report to the cloud or operation and maintenance terminal, which includes the number of newly added categories, the number of samples for each new category, and the number of old categories that have been removed.

Images