Methods and systems for collecting and analyzing electricity consumption data from power distribution sensing equipment

Abstract

This invention relates to the field of distribution network monitoring technology, specifically to a method and system for acquiring and analyzing power consumption data from distribution sensing equipment. The method includes: firstly, acquiring the healthy current sequence before the fault point; dynamically determining the autoregressive order based on the local information entropy deviation; and suppressing high-frequency noise by weighting according to the information entropy gradient to obtain a forward-inferenced current compensation sequence; subsequently, acquiring the transient voltage phase angle of the phase-locked loop, calculating the cumulative drift residual, and performing strict phase correction through orthogonal rotation transformation; furthermore, introducing a hyperbolic tangent function to perform asymptotic saturation mapping on the phase angle gradient to obtain a damping weight limit correction amount to prevent divergence. This invention can avoid envelope collapse and ensure waveform phase continuity, improving the reliability of data acquisition and reconstruction during distribution network fault periods and power quality analysis.

Description

Technical Field

[0001] This invention relates to the field of power distribution network monitoring technology. More specifically, this invention relates to a method and system for collecting and analyzing electricity consumption data from power distribution sensing equipment. Background Technology

[0002] In the field of power distribution network monitoring, when extreme events such as voltage dips or short circuits occur in the power grid, resulting in long-span gaps in the data, it is necessary to perform time-domain extrapolation to fill in the missing time periods.

[0003] Existing technologies treat observed signals as generalized stationary random sequences, neglecting the transient attenuation DC component and high-frequency noise introduced by the sudden change in line impedance at the moment of a fault. This makes it easy for blind recursion of fixed parameters to overfit when fitting attenuation characteristics. The derived waveform exhibits envelope distortion in the early stage of the breakpoint, loses continuity, and thus undermines the accuracy of power quality analysis results.

[0004] There are two deep-seated defects in pure data-driven forward extrapolation: First, it cannot distinguish and suppress historical node information contaminated by high-frequency random noise, which leads to the coupling of noise with the real transient characteristics, and the prediction error is amplified in cascade over time; Second, increasing the virtual extrapolation step size will cause the predicted current phase angle to gradually decouple from the real frequency of the power grid, resulting in irreversible cumulative phase drift, causing waveform splicing jumps and power calculation failures.

[0005] If the voltage phase angle is introduced as a cross-dimensional correction reference, and the phase angle gradient output by the phase-locked loop is directly used to drive the rotation correction, the transient surge during the voltage recovery period can easily trigger the phase angle gradient explosion, leading to numerical divergence or even computational collapse.

[0006] Therefore, there is an urgent need for a method for collecting and reconstructing power consumption data in distribution networks that can dynamically isolate noise pollution, suppress phase drift, and defend against gradient explosion. Summary of the Invention

[0007] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.

[0008] In a first aspect, the present invention provides a method for collecting and analyzing electricity consumption data from power distribution sensing devices, the method comprising:

[0009] A method for collecting and analyzing electricity consumption data from power distribution sensing equipment includes:

[0010] When a data breakpoint in the distribution network is detected, acquire the healthy current timing data prior to the breakpoint.

[0011] The local information entropy deviation of the health current time series data is calculated using a sliding window. This deviation is used as a penalty factor, and combined with the deductive approximation residual, the adaptive regression order for each forward prediction is dynamically determined. Forward prediction is then performed based on historical nodes within this order range. Simultaneously, dynamic weights negatively correlated with the local information entropy gradient are used to suppress the contribution of noise-contaminated nodes, resulting in a forward-inferred current compensation sequence. Specifically, the difference between the overall average information entropy of the health current time series data and the statistical representative value of the local information entropy within the backtracking interval before the breakpoint is taken, weighted by a gain coefficient to form an information entropy deviation compensation factor, which is then superimposed into the weighted synthesis result.

[0012] The transient voltage phase angle sequence tracked by the phase-locked loop during the residual voltage period is obtained, and the transient voltage phase angle gradient is calculated. The absolute value of the voltage phase angle gradient is asymptotically saturated and mapped using the hyperbolic tangent function to obtain the transient damping constraint weights that are restricted within a preset safety boundary.

[0013] The forward-engineered current compensation sequence is subjected to power frequency bandpass filtering to extract the fundamental component, and the filtered sequence is subjected to Hilbert transform to extract the current phase angle gradient; the difference between the transient voltage phase angle gradient and the current phase angle gradient is calculated and discretely integrated, and the integral result is subjected to modulo 2π mapping to restrict its principal value to the interval (-π,π], to obtain the cumulative phase angle drift residual;

[0014] The cumulative phase angle drift residual is weighted using the transient damping constraint weights to obtain the effective rotation angle. An orthogonal rotation transformation is constructed based on the effective rotation angle to perform phase correction on the forward extrapolation current compensation sequence. The damping constraint is applied to the rotation angle parameter through orthogonal rotation mapping to maintain the constant magnitude characteristic during the rotation process. While suppressing surge-induced divergence, the rotation energy is conserved to obtain the phase alignment compensation sequence, thus completing the power consumption data analysis.

[0015] Preferably, the local information entropy deviation includes: setting a sliding monitoring window on the health current time series data, calculating the Shannon information entropy variation amplitude between adjacent data subsets within the window, and calculating the local information entropy deviation reflecting the anomaly of high-frequency domain components.

[0016] Preferably, the adaptive regression order includes: searching for a historical backtracking depth value that minimizes the sum of the cumulative residual degree and the information entropy deviation penalty of the current inference, as the adaptive regression order.

[0017] Preferably, the step of suppressing the contribution of noise pollution nodes based on dynamic weights negatively correlated with the local information entropy gradient includes: obtaining the local information entropy gradient of the interval where each backtracking node is located; and using a preset monotonically decreasing function to map and calculate the absolute value of the local information entropy gradient to obtain the dynamic weight coefficient corresponding to each backtracking node.

[0018] Preferably, obtaining the forward extrapolation current compensation sequence includes: introducing an information entropy deviation compensation factor; calculating the difference between the overall average information entropy of the healthy current time series data and the statistical representative value of the local information entropy of each complete window within the backtracking interval before the breakpoint, wherein the statistical representative value is the median or the truncated mean; multiplying the difference by a compensation gain coefficient to obtain the information entropy deviation compensation factor; and superimposing the information entropy deviation compensation factor into the weighted synthesis result of the dynamic weight coefficient to obtain the forward extrapolation current compensation sequence.

[0019] Preferably, the acquisition of the transient voltage phase angle sequence tracked by the phase-locked loop during the residual voltage period includes: extracting the transient voltage phase angle sequence by continuously outputting discrete phase tracking records by the underlying phase-locked loop during the residual voltage period.

[0020] Preferably, obtaining the cumulative phase drift residual includes: aligning the forward extrapolation current compensation sequence and the transient voltage phase angle sequence under the same timestamp scale; performing a first-order backward difference operation on the voltage phase angle time series to obtain the transient voltage phase angle gradient; extracting the current phase angle gradient implicit in the forward extrapolation current compensation sequence; calculating the difference between the transient voltage phase angle gradient and the current phase angle gradient, performing discrete integration over the entire breakpoint time domain, and performing modulo 2π mapping on the integration result to restrict its principal value to the interval (-π, π], thereby obtaining the cumulative phase drift residual.

[0021] Preferably, the phase correction with damping constraints includes: applying the transient damping constraint weights to the cumulative phase angle drift residuals to obtain an effective rotation angle; constructing an orthogonal rotation mapping using the effective rotation angle as a rotation parameter; extracting the orthogonal components of the forward extrapolation current compensation sequence; and performing multiplication and addition operations on the current amplitude of the forward extrapolation current compensation sequence and the orthogonal components with the cosine and sine values ​​of the effective rotation angle, respectively, to obtain a phase alignment compensation value.

[0022] Preferably, the step of performing asymptotic saturation mapping to obtain transient damping constraint weights confined within a preset safety boundary includes: normalizing the absolute value of the voltage phase angle gradient; inputting the normalized gradient value into the hyperbolic tangent saturation mapping function to obtain the transient damping constraint weights, and the value of the weights is confined within a preset safety boundary determined by the reference correction weight coefficient.

[0023] Secondly, this invention provides a system for collecting and analyzing electricity consumption data from power distribution sensing equipment, employing the following technical solution:

[0024] A system for acquiring and analyzing electricity consumption data from power distribution sensing equipment includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the method for acquiring and analyzing electricity consumption data from power distribution sensing equipment as described above.

[0025] The embodiments of the present invention have at least the following beneficial effects:

[0026] 1. The autoregression order is dynamically determined by the local information entropy deviation, and the negative correlation dynamic weights effectively isolate high-frequency random noise, avoiding overfitting of the fixed model under interference. This ensures that the forward prediction accurately fits the transient current decay characteristics, so that the reconstructed waveform maintains the continuity to the greatest extent in long discontinuities, and ensures the high reliability of power quality analysis after data filling.

[0027] 2. By introducing the real voltage phase angle as a cross-dimensional reference, the cumulative phase angle drift residual is calculated. The deduced current is forced to correct the deviation through orthogonal rotation transformation, which reduces the limitations of pure data blind inference. The reconstructed waveform is anchored to the real power grid frequency in real time, which reduces the possibility of phase angle misalignment and jump at the recovery critical point, ensures the accuracy of subsequent power factor and harmonic analysis, and prevents the failure of derived calculation indicators due to phase cumulative drift.

[0028] 3. By employing the hyperbolic tangent function to perform asymptotic saturation mapping on the phase gradient, a constrained transient damping weight is generated. When transient surges cause gradient spikes, the maximum value can be automatically constrained within a preset safety boundary, helping to suppress the risk of numerical divergence and reduce the possibility of multiplier explosion. In the stationary region, the hyperbolic tangent function is approximately linear, and the correction accuracy is not significantly affected. Therefore, it helps improve numerical stability under extreme transient impacts while maintaining correction sensitivity under stable operating conditions, thus contributing to the robustness of the output power consumption data. Attached Figure Description

[0029] Figure 1 The flowchart illustrates the steps of the method for collecting and analyzing power consumption data from power distribution sensing devices in this invention. Detailed Implementation

[0030] 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, not all, of the embodiments of the present invention. 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.

[0031] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] S1: When a data breakpoint in the distribution network is detected, acquire the healthy current timing data before the breakpoint.

[0033] During the continuous online monitoring of power distribution sensing equipment in the power distribution network, the underlying analog-to-digital conversion channel synchronously samples the current signal and obtains discrete current scalar observation values ​​in a time-series flow.

[0034] Because extreme power quality events such as voltage dips or transient short circuits may occur in distribution network lines, the current sensing link will be subjected to strong electromagnetic transient impacts at the moment of the fault. At this time, the sampling interruption detection logic of the analog-to-digital conversion channel will identify the discontinuity break or illegal check code flag in the continuous sampling points.

[0035] When a valid data breakpoint is determined to be a non-instantaneous pulse interference, the real-time data writing operation of the current time axis is immediately frozen. Using the breakpoint occurrence time as the cutoff boundary, a time sliding window backtracking extraction is performed on the historical data buffer to extract a set of clean and statistically stable current time-domain discrete points before the fault anomaly occurs, and these points are encapsulated as healthy current time-series data before the breakpoint.

[0036] Specifically, the selection and extraction of healthy current time series data is based on the power frequency cycle characteristics of the distribution network to set back depth constraints. It requires that the extracted historical observation point sequence be continuous in time and cover a sampling span of no less than 5 complete steady-state cycles to ensure that the data segment fully contains the real load current fundamental waveform and the normal power frequency fundamental envelope pattern. Here, 5 is an empirical value that can be adjusted by the implementer according to the specific implementation scenario.

[0037] Perform point-by-point validity checks on the current scalar values ​​in the buffer, remove isolated abnormal spikes caused by link jitter, and finally obtain healthy current time-series data arranged at equal intervals in the time domain. , where subscript This represents a discrete sampling index that increments in the historical direction with the breakpoint as the origin. Each element in this sequence is the instantaneous amplitude of a scalar current in amperes. The sampling interval is fixed as a constant determined by the underlying analog-to-digital converter clock, in seconds (s). During the acquisition of health current timing data, the sensing device synchronously stores the voltage phase angle sequence output by the phase-locked loop. The empirical value is 128, which can be adjusted by the implementer according to the specific implementation scenario.

[0038] Among them, the point-by-point validity verification can be implemented using the median filtering algorithm. The analog-to-digital conversion channel sampling interruption detection logic and the point-by-point validity verification technology using median filtering are well-known technical contents and will not be described in detail here.

[0039] S2: Calculate the local information entropy deviation of the healthy current time series data using a sliding window. Dynamically determine the adaptive regression order of each forward prediction using the local information entropy deviation as a penalty factor. Perform forward prediction based on historical nodes within the order range. At the same time, suppress the contribution of noise pollution nodes according to dynamic weights that are negatively correlated with the local information entropy gradient, and obtain the forward extrapolation current compensation sequence.

[0040] Because the impact of a fault in the distribution network often injects high-frequency transient noise components into the current link, these noise components are superimposed on the originally stable load current, which significantly enhances the randomness of the signal in some intervals of the health time series data.

[0041] If these contaminated historical points are still treated the same as uncontaminated points and used for forward extrapolation, the prediction model will overfit at noisy locations, causing the extrapolated current waveform envelope to become rapidly distorted at the breakpoint initiation stage.

[0042] Therefore, it is necessary to analyze the local information structure of the health current time series data to identify the degree of noise pollution and its variation characteristics on the time axis, so as to implement dynamic isolation in subsequent simulations.

[0043] exist Set a time-domain sliding monitoring window, with a window length of [missing information]. Set as There are discrete sampling points, and the window sliding step size is... Each sampling point is located at each microstep position of the window sliding. Extract the current amplitude subsequence within the window, divide it into two equal-length subsets, and calculate the Shannon information entropy for each subset. The window length is... These are empirical values ​​and can be adjusted by the implementer based on the specific implementation scenario.

[0044] When calculating the entropy value, an equal-width binning method based on the current amplitude range is used to divide the current values ​​within the subset into... For each interval, the frequency of sampling points within each interval is counted and converted into probability. Then, the information entropy value of each interval is calculated according to Shannon's definition of information entropy. These are empirical values ​​and can be adjusted by the implementer based on the specific implementation scenario.

[0045] The absolute value of the difference between the information entropy of the two subsets is taken as the local information entropy deviation at that window position. .

[0046] Because high-frequency noise disrupts the local deterministic structure of the current waveform and increases the randomness of the amplitude distribution, The higher the value, the more severe the high-frequency domain component anomaly contamination of the signal segment within the window.

[0047] Thus, a sequence of local information entropy deviations of the same length as the health current time series data is obtained by scanning a sliding window.

[0048] Since blindly recursing with a fixed order can lead to an increase in prediction residuals due to over-inclusion of distant contaminated nodes or failure to fully utilize recent reliable nodes, it is necessary to dynamically determine the number of historical backtracking nodes used for this prediction for the first forward prediction point needed after the breakpoint, i.e., the adaptive regression order.

[0049] Therefore, the local information entropy deviation calculated from the most recent sliding window is used. As a core penalty factor, it is introduced into the order optimization process.

[0050] For each candidate order Its value range is strictly constrained to Within an integer space, to ensure short-term relevance and economy of computation, where The value range is based on experience and can be adjusted by the implementer according to the specific implementation scenario.

[0051] Use the latest Using 100 health data points as the fitting validation segment, The model performs backward prediction on the validation segment and calculates the root mean square value of the approximate residual. The unit is ampere (A), where, These are empirical values, which can be adjusted by the implementer based on the specific implementation scenario.

[0052] Simultaneously, a local information entropy fluctuation penalty term is constructed, specifically in the form of... ,in As a penalty intensity coefficient, in order to ensure the consistency of physical dimensions in the calculation process, this coefficient is assigned the unit dimension of ampere (A), with an empirical value of 1.5 amperes, which can be adjusted by the implementer according to the specific implementation scenario.

[0053] The total cost function is then... .in This can be interpreted as an equivalent reduced current residual penalty due to information entropy deviation, and its dimensions are... Similarly, the two can be directly added together to jointly drive order optimization. The search makes... Minimized The value is determined as the adaptive regression order for this forward prediction. .

[0054] Among them, the search method for order optimization adopts the grid search method, and implementers can choose other search methods according to the specific implementation scenario.

[0055] This process causes the penalty term to compress the effective order value when the local information entropy deviation is large, avoiding the introduction of too many distant historical points contaminated by noise, thereby achieving adaptation to the noise intensity at the order level.

[0056] Determine the order Then, historical nodes within this order range are used to predict the current time. Current compensation value Perform weighted synthesis. Then, combine the obtained healthy current time-series data. Reindex by timeline, Indicates the number of points before the breakpoint The current amplitude at each sampling point, of which ,Right now Among them, the first sampling point after the breakpoint is taken as .

[0057] Each historical node Corresponding weight coefficients It is not a fixed value, but is dynamically calculated based on the information entropy gradient of the local interval where the historical node is located.

[0058] First, calculate the local information entropy gradient at each historical node, which is defined as the local information entropy deviation of the current window. The mathematical definition of the local information entropy gradient is the difference between the local information entropy deviation of the adjacent previous window and the local information entropy deviation of the previous window: = This refers to the difference in local information entropy between the current sliding window and the adjacent previous window.

[0059] Because when the local information entropy around a node surges abnormally, it means that the node is in a burst of noise and its measurement is unreliable, so the initial weight parameters of each backtracking node are first calculated using an exponential decay method: ,in The baseline weighting factor is empirically set to 1.0 and is dimensionless. This is the attenuation sensitivity coefficient, empirically taken as 0.5, and is dimensionless. Divide by... The term ensures that the further the historical node is from the current time, the lower its baseline contribution weight when it is not affected by noise, which is consistent with the prior assumption of decreasing time correlation. and It can be adjusted by the implementer according to the specific implementation scenario.

[0060] To prevent uncontrollable scaling of the overall amplitude of the reconstructed waveform caused by directly adding the weights of each node, a normalization mapping is performed on the obtained initial weight parameters to calculate the dynamic weight coefficients for formal use: .

[0061] Through this normalization operation, when the local information entropy of the interval where a backtracking node is located surges abnormally, its corresponding normalization weight... The noise will be significantly suppressed, thereby reasonably transferring and suppressing the contribution of the noise pollution node in the weighted synthesis, achieving the stripping away of unreliable historical information and maintaining overall energy conservation.

[0062] The aforementioned dynamic weighting mechanism corrects the local waveform distortion caused by noise contamination by reallocating the contribution ratios of nodes. This mechanism applies to the relative weights of each historical node, reasonably suppressing the contribution of nodes contaminated by noise, while correspondingly enhancing the contribution of trusted nodes.

[0063] Based on this, an information entropy deviation compensation factor is introduced. This factor is used to correct the weighted synthesis baseline drift caused by the overall information structure shift of the health data segment. This factor is independent of the aforementioned dynamic weighting mechanism in terms of its function. The dynamic weighting adjusts the relative contribution ratio of each node, while the compensation factor applies an overall baseline shift to the weighted synthesis result. Together, they correct the prediction output from two dimensions: local structure ratio and global baseline bias.

[0064] To ensure that the compensation factor reflects only the overall information structure shift of the health data, rather than instantaneous noise disturbances, the factor is set to the average information entropy of the entire health current time series data. The median of the local information entropy of each complete window within the backtracking interval before the breakpoint The difference is then multiplied by the compensation gain coefficient. ,Right now .

[0065] in, Before the breakpoint The median value of the local information entropy of a complete sliding window. An empirical value of 5 is chosen to match the coverage span of the steady-state cycle, and can be adjusted by the implementer according to the specific implementation scenario.

[0066] Using the median instead of the instantaneous value of the last window before the breakpoint can effectively avoid the interference of transient noise disturbances in the interval near the breakpoint on the compensation term, so that the compensation factor only reflects the systematic energy shift of the data segment in a statistical sense.

[0067] The empirical value is The unit is ampere (A), which can be adjusted by the implementer according to the specific implementation scenario. This is the average value obtained by calculating the entropy of all windows in the statistical health data segment.

[0068] If the statistical median value is higher than the baseline entropy, it indicates that the overall randomness of the signal in this data segment is increased due to persistent interference. The weighted synthesis result tends to have a systematically low baseline, and the compensation factor provides a positive correction; the opposite is also true.

[0069] Finally, the forward-derived current compensation value .in, The total number of sampling points corresponding to the duration of the breakpoint is calculated by the breakpoint detection module based on the fixed sampling interval of the analog-to-digital converter and the breakpoint time span. Then, by performing the above steps point by point, all sampling points to be predicted within the breakpoint interval are processed. The deductions are carried out sequentially, and all of them are deduced. Combine them in chronological order to output the forward extrapolation current compensation sequence.

[0070] Each element in the sequence is a scalar instantaneous current amplitude in amperes (A). The process of obtaining it dynamically adjusts the order and weights according to the local noise distribution, so that the interference of high-frequency random noise is effectively blocked and the deterministic characteristics of the transient decay of the current are preserved.

[0071] S3: Obtain the transient voltage phase angle sequence tracked by the phase-locked loop during the residual voltage period, calculate the difference between the transient voltage phase angle gradient and the current phase angle gradient of the forward extrapolation current compensation sequence, and perform discrete integration to obtain the cumulative phase angle drift residual.

[0072] Since the forward-deductive current compensation sequence is obtained entirely by the autoregressive extrapolation of the current time series, when the breakpoint span is large, the phase angle of the predicted current will gradually lose synchronization with the actual frequency of the power grid due to the inertia of mathematical recursion, resulting in irreversible cumulative phase shift.

[0073] This offset stems from the fact that the power frequency of the grid is not absolutely constant during the residual voltage period, but rather there are tiny dynamic fluctuations caused by generator inertia and load switching. Pure current autoregressive models cannot detect this drift in the true frequency.

[0074] Therefore, an external phase reference is introduced into the current derivation process to measure and calibrate the phase of the derivation current.

[0075] Through the distribution network sensing terminal, with the local residual voltage signal as input, its internal voltage-controlled oscillator operates under closed-loop control of a phase comparator and a loop filter.

[0076] The residual voltage period refers to the time period from the moment the fault occurs until the voltage recovers to the protection device's operating threshold or the voltage recovers to a preset proportion, such as 90%, of the nominal voltage. During this period, even if the voltage amplitude drops significantly or the waveform is distorted, the phase-locked loop can still maintain continuous tracking and locking of the fundamental voltage phase angle within its design bandwidth and capture range, and output a discrete transient voltage phase angle sequence.

[0077] If the voltage amplitude drops below the minimum operating voltage of the sensing device, causing the phase-locked loop to lose lock, it enters the holding mode and extrapolates using the last effective phase angle gradient before the lockout.

[0078] The phase-locked loop outputs its instantaneous phase alignment result with a fixed discrete time step, forming a transient voltage phase angle sequence that is equidistant from the time axis of the breakpoint interval, denoted as . , where index Indicates the start time from the breakpoint The beginning of the first Each sampling point is measured in radians (rad).

[0079] This voltage phase angle sequence reflects the frequency evolution of the distribution network during fault transients, providing a rigid anchoring reference across dimensions for correction.

[0080] After obtaining the transient voltage phase angle sequence, a first-order backward difference operation is performed on it to calculate the phase angle increment between every two adjacent sampling points, thus obtaining the transient voltage phase angle gradient sequence. The unit is radians per sampling step (rad / sample).

[0081] Meanwhile, the implicit current phase angle is calculated from the time-series signal of the forward-derived current compensation sequence.

[0082] First, a zero-phase power frequency bandpass filter is performed on the forward extrapolation current compensation sequence to extract the fundamental frequency component. Since transient DC attenuation components and high-frequency noise may remain in the extrapolated signal, directly performing a Hilbert transform would lead to distortion in instantaneous frequency estimation. Narrowband filtering can suppress these interferences, ensuring the signal meets the single-component condition without introducing additional phase shift. Subsequently, a Hilbert transform is performed on the entire filtered fundamental frequency sequence to extract the orthogonal axis virtual current components at each time step. Then, the instantaneous phase is extracted through four-quadrant arctangent operation, and the phase values ​​of adjacent points are subjected to first-order difference to obtain the forward-derived current phase angle correlated gradient sequence. Among them, the technique of extracting instantaneous phase using Hilbert transform is a well-known technique and will not be described in detail here.

[0083] To calculate the degree of phase mismatch caused by pure data-driven extrapolation, the instantaneous difference between the two gradient sequences at each sampling point is calculated. This difference is used as the instantaneous phase angle drift rate of the current microstep, with the unit being radians per sampling step (rad / sample).

[0084] The frequency shifts of the voltage and current phase angles at the same node during transient processes are both governed by the grid synchronization frequency, and their phase angle gradient changes are essentially from the same source. It should be noted that the impedance change at the moment of a fault will cause a step jump in the power angle between the fundamental voltage and current. This step is manifested as an instantaneous constant bias in the absolute difference of the phase angle, which is naturally eliminated in the subsequent first-order differential operation.

[0085] The conclusion that the power angle step is eliminated by differential is based on the premise that the impedance angle remains constant or only undergoes a negligible slow drift after the fault.

[0086] In actual distribution network transient processes, if situations such as arc reignition or system deparallelization occur, causing the impedance angle to undergo continuous low-frequency modulation, the difference between the voltage phase angle gradient and the current phase angle gradient will not only include the phase mismatch rate extrapolated from pure data, but may also be superimposed with the slowly varying residual component introduced by the dynamic change of the impedance angle.

[0087] To address this, the gradient difference is bounded by the hyperbolic tangent asymptotic saturation mapping in subsequent step S5, which limits the magnitude of the potential slowly varying residual within a preset safety boundary, thus preventing excessive correction caused by dynamic changes in the impedance angle.

[0088] Under the assumption that the grid synchronization frequency dominates, the difference between the voltage phase angle gradient and the current phase angle gradient mainly reflects the phase mismatch rate of pure data extrapolation relative to the grid reference; the superimposed potential impedance angle dynamic residual components are uniformly incorporated into the constraint range by the damping mechanism of step S5.

[0089] The cumulative phase drift residual obtained by integrating the gradient difference, under the constraints, characterizes the degree of dynamic phase deviation of the derived waveform.

[0090] This instantaneous deviation accumulates continuously in the breakpoint time domain as the prediction step size increases, forming a total phase misalignment that evolves over time.

[0091] Therefore, from the start time of the breakpoint To the current prediction point Discretely accumulate and integrate all instantaneous phase drift rates over the entire interval to obtain the total cumulative phase drift.

[0092] To use this drift amount for orthogonal rotation transformation, the sine / cosine functions satisfy... , Based on the periodic equivalence property, a modulo 2π mapping is performed on the integral result to extract the equivalent base angle that has the same trigonometric function mapping effect as the cumulative drift, and its principal value is limited to the interval (-π,π] to obtain the cumulative phase angle drift residual used to drive the rotation.

[0093] It should be noted that this modulo 2π mapping operation only guarantees the equivalence of the trigonometric function mapping value and the original drift amount in terms of rotation effect, and does not retain absolute phase accumulation information beyond a single period.

[0094] Therefore, using the residual for orthogonal rotation transformation is complete and rigorous; however, if subsequent applications require extracting absolute cumulative information of phase monotonically increasing with time, the original integral value before the modulo 2π mapping needs to be saved separately. The mapped residual is used only for the rotation transformation stage, while the original integral value is used for trend analysis. The two work together to avoid information loss.

[0095] This mapping operation avoids the loss of floating-point precision caused by the continuous growth of accumulated drift and ensures the mathematical correctness of the rotation transformation.

[0096] The residual fully encompasses the cascaded phase error introduced by neglecting the dynamic changes in the actual frequency of the power grid, and can be used for orthogonal rotation transformation.

[0097] S4: Based on the accumulated phase angle drift residual, construct an orthogonal rotation transformation to perform coordinate system rotation phase correction on the current amplitude of the forward extrapolation current compensation sequence.

[0098] Since the forward-derived current compensation sequence relies solely on the time extrapolation of the current itself, when the breakpoint span is long, the current phase angle will gradually desynchronize with the actual frequency of the power grid during the residual voltage period. The resulting cumulative phase drift has been calculated as the cumulative phase angle drift residual in step S3. .

[0099] The residual represents the value up to the current prediction time. This refers to the overall angular offset of the extrapolated waveform relative to the actual voltage reference of the distribution network. If this current sequence is used directly for subsequent power quality analysis without correction, this systematic phase skew will cause the power factor calculation and the analysis of the fundamental phase-independent terms to fail.

[0100] Therefore, it is necessary to rotate the coordinate system of each current amplitude in the forward-deduced current compensation sequence based on the residual, and forcibly resynchronize the freely diverging current vector to the rotating reference system determined by the grid frequency.

[0101] After obtaining the cumulative phase angle drift residual sequence Then, it is compared with the forward-derived current compensation sequence. By sampling index Strict one-to-one correspondence sorting and alignment, index The first prediction point after the breakpoint The incrementing process begins. Since the accumulated phase drift residual characterizes the overall angular displacement misalignment of the pure data extrapolation waveform relative to the actual voltage reference of the distribution network, it is necessary to perform coordinate system rotation phase correction on each current amplitude in the forward extrapolation current compensation sequence to resynchronize the freely diverging current vector to a rotating reference frame determined by the grid frequency. This phase correction will be implemented in step S5 in conjunction with transient damping constraint weights to prevent calculation divergence caused by surges.

[0102] S5: The absolute value of the voltage phase angle gradient is asymptotically saturated and mapped using the hyperbolic tangent function to generate transient damping constraint weights. Based on these weights, the accumulated phase angle drift residuals are subjected to bounded rotation correction to obtain the phase alignment compensation sequence.

[0103] To implement the phase correction described in S4, before constructing the orthogonal rotation transformation, this step first uses the hyperbolic tangent function to perform an asymptotic saturation mapping on the absolute value of the voltage phase angle gradient, generating transient damping constraint weights to constrain the effective range of the rotation angle, thereby preventing numerical divergence caused by gradient explosion while achieving phase alignment.

[0104] When calculating the voltage phase angle gradient, the distribution network is often accompanied by transient surges caused by circuit breaker closing or fault point clearing at the moment of voltage recovery. This process causes the node voltage phase angle to undergo a near discontinuous step change within a very short time scale.

[0105] This extreme discontinuity at this level directly maps to the numerical calculation of the phase angle gradient, making the voltage phase angle gradient obtained by performing a first-order backward difference on the transient voltage phase angle sequence based on the phase-locked loop output... absolute value Explosive growth has occurred.

[0106] If this surge in absolute gradient value is directly imported into the calculation link of the correction for orthogonal rotation transformation without restriction, this abnormally large value will force the accumulation of phase drift residuals. Distortion occurs, and through Item and The multiplicative amplification effect of the term causes the phase alignment compensation value, which should have remained continuous, to be affected. At the surge node, it tends towards infinity or produces non-numerical results, thus causing a chain reaction of collapses in the entire reconstructed dataset under this extreme condition.

[0107] To address the computational divergence problem caused by the explosion of voltage phase angle gradients, before involving the orthogonal rotation correction of the voltage phase angle gradient, we first... A benchmarking normalization process is performed to compress its numerical range to a dimensionless input domain that matches the saturation characteristics of the hyperbolic tangent function.

[0108] Normalization processing by Divide by the expected amplitude benchmark calculated within the health time series statistical interval. Since the health data segment and the residual voltage period belong to the same continuous monitoring period, and the system was operating in steady-state condition before the fault, this benchmark statistical value can represent the normal amplitude level of the voltage phase angle gradient under non-surge conditions. It is suitable for normalized mapping during the residual voltage period, and the normalized absolute value of the voltage phase angle gradient is obtained. .

[0109] in, This is the root mean square value of the absolute value of the voltage phase angle gradient within multiple power frequency cycles corresponding to the window of the healthy current time series data before the breakpoint, in radians per sampling step (rad / sample), normalized result. It is dimensionless.

[0110] This normalization operation maps the phase gradients under different transient intensities to a comparable relative scale, ensuring that the nonlinear compression characteristics of the saturation mapping always act on the same reference domain.

[0111] Then, Input an inverse saturation mapping model based on the hyperbolic tangent function to generate transient damping constraint weights. Its specific relation is:

[0112]

[0113] In the formula, It is a hyperbolic tangent nonlinear mapping kernel; The baseline correction weighting coefficient represents the maximum allowable gain multiple of the distribution network under normal and stable operating conditions. It is empirically set to 1.0 and is dimensionless. This is a nonlinear extreme response factor used to adjust the asymptotic convergence stiffness of the saturation curve as the gradient increases. An empirical value of 3.0 is used; it is dimensionless. and It can be adjusted by the implementer according to the specific implementation scenario.

[0114] When the voltage phase angle gradient is within the normal fluctuation range, i.e. When approaching zero, The function output is close to zero. The term approaches 1.0, at which point... Basically maintained at the same level as Equally high weighting levels ensure lossless execution of correction calculations; however, when surges cause... It increases sharply and exceeds the response factor When the defined soft threshold is reached, The function rapidly enters the deep saturation region and asymptotically approaches 1.0, leading to... The term approaches zero, making It is forcibly suppressed to an extremely low level, effectively inhibiting the risk of multiplier amplification caused by gradient explosion.

[0115] Obtain the transient damping constraint weights Then, it is applied to the accumulated phase angle drift residual to constrain the effective rotation angle in the orthogonal rotation transformation, rather than directly compressing the current amplitude. Specifically, the transient damping constraint weight is combined with the accumulated phase angle drift residual to construct a damped orthogonal rotation transformation.

[0116] ΔΘdamped(k)=Wdamp(k)·ΔΘdrift(k)

[0117]

[0118] In the formula, For the first The forward-looking current amplitude at each sampling point is expressed in amperes (A). These are the orthogonal components obtained through the Hilbert transform, in amperes (A). This represents the cumulative phase drift residual up to this point, expressed in radians (rad). The transient damping constraint weights are dimensionless. The effective rotation angle after damping constraint, in radians (rad).

[0119] Applying damping weights to the rotation angle rather than the final magnitude: the rotation matrix itself satisfies... The identity relation strictly maintains the magnitude of the vector at any rotation angle, thus mathematically guaranteeing the energy conservation characteristic of the phase correction process and avoiding the envelope distortion problem caused by direct amplitude compression.

[0120] Under normal operating conditions Maintaining a high position, approaching 1.0, effective rotation angle Approximately equal to the original accumulated phase drift residual, the correction equation essentially performs coordinate system rotation mapping without attenuation; under the surge-induced gradient explosion condition, Suppressed to an extremely low level, effective rotation angle Confined to a safe area, cut off from the cause The phase jump and divergence path caused by distortion through trigonometric function mapping.

[0121] Therefore, while maintaining normal correction accuracy, the risk of cascading error amplification under extreme shocks is reduced. This is achieved by adjusting all corrected phase alignment compensation values ​​within the breakpoint interval. Arranged in chronological order, the phase alignment compensation sequence is obtained, thus completing the power consumption data analysis.

[0122] This invention also discloses a system for collecting and analyzing electricity consumption data from power distribution sensing devices, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method for collecting and analyzing electricity consumption data from power distribution sensing devices of this invention is implemented.

[0123] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0124] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for collecting and analyzing electricity consumption data from power distribution sensing equipment, characterized in that, include: When a data breakpoint in the distribution network is detected, acquire the healthy current timing data prior to the breakpoint. The local information entropy deviation of the health current time series data is calculated using a sliding window. This deviation is used as a penalty factor, and combined with the deductive approximation residual, the adaptive regression order for each forward prediction is dynamically determined. Forward prediction is then performed based on historical nodes within this order range. Simultaneously, dynamic weights negatively correlated with the local information entropy gradient are used to suppress the contribution of noise-contaminated nodes, resulting in a forward-inferred current compensation sequence. Specifically, the difference between the overall average information entropy of the health current time series data and the statistical representative value of the local information entropy within the backtracking interval before the breakpoint is taken, weighted by a gain coefficient to form an information entropy deviation compensation factor, which is then superimposed into the weighted synthesis result. The transient voltage phase angle sequence tracked by the phase-locked loop during the residual voltage period is obtained, and the transient voltage phase angle gradient is calculated. The absolute value of the voltage phase angle gradient is asymptotically saturated and mapped using the hyperbolic tangent function to obtain the transient damping constraint weights that are restricted within a preset safety boundary. The forward-engineered current compensation sequence is subjected to power frequency bandpass filtering to extract the fundamental component, and the filtered sequence is subjected to Hilbert transform to extract the current phase angle gradient; the difference between the transient voltage phase angle gradient and the current phase angle gradient is calculated and discretely integrated, and the integral result is subjected to modulo 2π mapping to restrict its principal value to the interval (-π,π], to obtain the cumulative phase angle drift residual; The cumulative phase angle drift residual is weighted using the transient damping constraint weights to obtain the effective rotation angle. An orthogonal rotation transformation is constructed based on the effective rotation angle to perform phase correction on the forward extrapolation current compensation sequence. The damping constraint is applied to the rotation angle parameter through orthogonal rotation mapping to maintain the constant magnitude characteristic during the rotation process. While suppressing surge-induced divergence, the rotation energy is conserved to obtain the phase alignment compensation sequence, thus completing the power consumption data analysis.

2. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The local information entropy deviation includes: setting a sliding monitoring window on the health current time series data, calculating the Shannon information entropy variation amplitude between adjacent data subsets within the window, and calculating the local information entropy deviation reflecting the anomaly of high-frequency domain components.

3. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The adaptive regression order includes: a historical backtracking depth value that minimizes the sum of the cumulative residual degree and the information entropy deviation penalty in the current inference, which is used as the adaptive regression order.

4. The method for collecting and analyzing electricity consumption data from power distribution sensing equipment according to claim 1, characterized in that, The step of suppressing the contribution of noise-contaminated nodes based on dynamic weights negatively correlated with the local information entropy gradient includes: obtaining the local information entropy gradient of the interval where each backtracking node is located; and using a preset monotonically decreasing function to map and calculate the absolute value of the local information entropy gradient to obtain the dynamic weight coefficient corresponding to each backtracking node.

5. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The obtained forward extrapolation current compensation sequence includes: An information entropy deviation compensation factor is introduced; the difference between the overall average information entropy of the health current time series data and the statistical representative value of the local information entropy of each complete window in the backtracking interval before the breakpoint is calculated, wherein the statistical representative value is the median or the truncated mean; Multiply the difference by the compensation gain coefficient to obtain the information entropy deviation compensation factor; The information entropy deviation compensation factor is superimposed on the weighted synthesis result of the dynamic weight coefficient to obtain the forward extrapolation current compensation sequence.

6. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The acquisition of the transient voltage phase angle sequence tracked by the phase-locked loop during the residual voltage period includes: extracting the transient voltage phase angle sequence by continuously outputting discrete phase tracking records by the underlying phase-locked loop during the residual voltage period.

7. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The method for obtaining the cumulative phase drift residual includes: Align the forward extrapolation current compensation sequence with the transient voltage phase angle sequence under the same timestamp scale; Perform a first-order backward differential operation on the voltage phase angle timing sequence to obtain the transient voltage phase angle gradient; extract the current phase angle gradient implied in the forward extrapolation current compensation sequence; The difference between the transient voltage phase angle gradient and the current phase angle gradient is calculated. Discrete integration is performed over the entire breakpoint time domain, and the integral result is mapped to modulo 2π to restrict its principal value to the interval (-π,π], thus obtaining the cumulative phase angle drift residual.

8. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The phase correction with damping constraint includes: The effective rotation angle is obtained by weighting the cumulative phase angle drift residual using the transient damping constraint weights. Using the effective rotation angle as the rotation parameter, an orthogonal rotation mapping is constructed, and the orthogonal components of the forward extrapolation current compensation sequence are extracted. The current amplitude of the forward extrapolation current compensation sequence and the orthogonal components are multiplied and added with the cosine and sine values ​​of the effective rotation angle, respectively, to obtain the phase alignment compensation value.

9. The method for collecting and analyzing electricity consumption data for power distribution sensing equipment according to claim 1, characterized in that, The process of performing asymptotic saturation mapping to obtain transient damping constraint weights that are confined within a preset safety boundary includes: normalizing the absolute value of the voltage phase angle gradient; inputting the normalized gradient value into the hyperbolic tangent saturation mapping function to obtain the transient damping constraint weights, and the value of the weights is confined within a preset safety boundary determined by the reference correction weight coefficient.

10. A system for acquiring and analyzing electricity consumption data from power distribution sensing equipment, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the method for acquiring and analyzing power consumption data for power distribution sensing devices according to any one of claims 1-9.

Images