Power frequency current long and short term trend filtering method, device and processing equipment

Abstract

The invention provides a power frequency current long and short term trend filtering method and device and processing equipment, and constructs a filter design comprising a long term trend module for long term control of accumulated residual errors and a short term trend module for short term estimation of linearization trend. A set of specific double-closed-loop control structure is built on the aspect of a long and short term trend framework, and further matching is performed from the aspect of a quantization formula, so that the limitation in the prior art can be effectively overcome, the current transient analysis in a complex scene can be completed with high precision, and the current transient analysis accuracy is improved. The method is suitable for real current observation of a power supply and transient analysis of power frequency current in specific application.

Description

Technical Field

[0001] The present application relates to the field of filtering, and specifically to a method, device and processing equipment for filtering long-term and short-term trends of power frequency current. Background Art

[0002] With the increasing complexity of power systems due to power electronics, industrial frequency power supplies containing pulse width modulation (PWM) harmonics are becoming widely used. Obtaining transient information about the voltage and current phases of power supplies is crucial for calculating instantaneous frequency and power. Although frequency domain harmonic analysis algorithms are widely adopted, determining the time window for variable-frequency interharmonic power supplies is difficult. Various short-circuit faults, large-capacity motors, the commissioning or removal of important equipment, and active islanding detection of power supplies can also affect the effectiveness of frequency domain analysis. Therefore, frequency domain analysis algorithms are not conducive to power supply transient analysis. Real-time and accurate mining of current transient information in real currents containing a large amount of modulated harmonics and random noise has become a core issue for algorithms such as current zero crossings and power allocation methods.

[0003] Among signal processing algorithms for long-term time series, mainstream neural networks can be used to analyze time series in power systems. However, in general, neural network training can only use historical data from equipment operation. During training, neural networks need to avoid overfitting and be unaffected by measurement errors. Due to the complexity of the neural network interface, it is difficult to find an optimal parameter combination. Professionals can only provide guidelines for effectively adjusting hyperparameters. When processing the current during abnormal operation of the power supply, the transient characteristics of the trained neural network are still unknown. In addition, the gating units and forgetting gates in the neural network structure will ignore some information in the time series, thereby losing some statistical features of the original data. In addition, since the length of the time series of similar algorithms is limited, the corresponding long-term observation capabilities are also limited.

[0004] Among the signal processing algorithms for short-term time series, frequency domain signal processing, statistical signal processing, and prediction theory algorithms are widely used in the processing of power frequency current. However, frequency domain indicators are usually used to design analog circuit component parameters or digital filter parameters. High-performance frequency domain filters usually only consider flat passbands, sharp transition bands, and highly suppressed stopbands. In random noise environments and nonlinear environments, the filtering ability of frequency domain methods is limited. Statistical theory can effectively deal with random noise and can also be applied to the processing of power frequency current signals. However, as long as it is statistical theory, it must involve statistical indicators, and statistical indicators must be related to the number of samples. For real-time time series, the number of samples must be within a limited short time range. In real-time processing, the distribution characteristics of the finite number of samples cannot represent the distribution characteristics of the global samples. In current signals that are closely related to time, parameters such as step size are very sensitive. Parameter settings may cause the algorithm to fail to converge or the convergence period to be too long. Therefore, adaptive filtering related to statistical theory is more widely used in real-time signal processing. The adaptive filtering algorithm is a feedback-driven closed-loop device that can adjust its own parameters through feedback information to achieve the best filtering effect. Commonly used adaptive filter algorithms include Least Mean Square (LMS) and Least Mean Square (LMS). Square (LMS), Recursive Least Squares (RLS), and Normalized Least Mean Square (NLMS) algorithms have the advantages of fast convergence, low complexity, and strong flexibility. However, input correlation can lead to interactions between adaptive filter parameters. The greater the filter dimension and input correlation, the greater the interaction between the overall filter parameters. Prediction theory can also be used to process power-frequency current signals. Prediction algorithms are generally based on prior knowledge and focus on local features. Locally linearized extended Kalman filters, featureless Kalman filters, and linearized particle filters are effective in most applications with small disturbances. However, these algorithms are affected by changes in environmental noise and complex parameter determination, and require the design of an initial test plan with a relatively small parameter set. These iterative methods increase the complexity of system design and verification.

[0005] The real-time processing of statistical algorithms focuses on the statistical characteristics of a finite set of samples, while the real-time processing of predictive algorithms is a guess. The quality of the guess depends on the tracking ability of the predicted value and the deviation between the predicted value and the true value. Model-free predictive filters get rid of the complexity of system design and complete the design based on stable signal tracking. Although this type of algorithm has the ability to improve accuracy and has fewer adjustment parameters, the algorithm will introduce estimation bias in the process of identifying motion and has limitations in the real-time processing of random signals. Summary of the Invention

[0006] The present application provides a method, device and processing equipment for filtering the long-term and short-term trends of industrial frequency current, constructs a filter design that includes a long-term trend module for long-term control of cumulative residuals and a short-term trend module for short-term estimation of linear trends, and builds a specific dual closed-loop control structure in the long-term and short-term trend framework. In addition, further supporting measures are made from the perspective of quantitative formulas, thereby effectively overcoming the limitations of existing technologies and completing current transient analysis in complex scenarios with high precision. It is suitable for real current observation of power supplies and transient analysis of industrial frequency current in specific applications.

[0007] In a first aspect, the present application provides a method for filtering long-term and short-term trends of power frequency current, the method comprising:

[0008] Obtaining the initial power frequency current to be filtered;

[0009] The initial power frequency current is subjected to long-term and short-term trend filtering through a pre-configured filter. The filter, in terms of the long-term and short-term trend framework under the data fusion framework, includes: an outer loop structure and an inner loop structure in the dual closed-loop control structure, each including a long-term trend module for long-term control cumulative residuals and a short-term trend module for short-term estimation linearization trend; and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module; the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure;

[0010] The target power frequency current after filtering the initial power frequency current output by the filter is extracted.

[0011] In a second aspect, the present application provides a power frequency current long-term and short-term trend filtering device, the device comprising:

[0012] An acquisition unit, used for acquiring an initial power frequency current to be filtered;

[0013] The filtering unit is used to perform long-term and short-term trend filtering processing on the initial power frequency current through a pre-configured filter, wherein the filter has the following features in terms of the long-term and short-term trend framework under the data fusion framework: the outer loop structure and the inner loop structure in the dual closed-loop control structure respectively include a long-term trend module for long-term control cumulative residual and a short-term trend module for short-term estimation linearization trend, and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module, the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure;

[0014] The extraction unit is used to extract the target power frequency current output by the filter after the initial power frequency current is filtered.

[0015] In a third aspect, the present application provides a processing device including a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method provided in the first aspect of the present application is executed.

[0016] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a plurality of instructions suitable for loading by a processor to execute the method provided in the first aspect of the present application.

[0017] From the above content, it can be concluded that this application has the following beneficial effects:

[0018] Aiming at the goal of high-performance industrial frequency current filtering, this application constructs a filter design that includes a long-term trend module for long-term control of cumulative residuals and a short-term trend module for short-term estimation of linear trends, and builds a specific dual closed-loop control structure in the long-term and short-term trend framework. In addition, further supporting measures are made from the perspective of quantitative formulas, which can effectively overcome the limitations of existing technologies and complete current transient analysis in complex scenarios with high precision. It is suitable for real current observation of power supplies and transient analysis of industrial frequency currents in specific applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A flow chart of the long-term and short-term trend filtering method of power frequency current in this application;

[0021] Figure 2 This is a schematic diagram of the long-term and short-term trend framework of this application;

[0022] Figure 3 A schematic diagram of a scenario for the ideal trend filtering algorithm of this application;

[0023] Figure 4 For this application k sequence sum A time domain characteristic diagram of the local linearization of the curve;

[0024] Figure 5 A schematic diagram of a framework of the long-term and short-term trend filter of the model-free prediction filter of this application;

[0025] Figure 6 This is a structural diagram of the long-term and short-term trend filtering device for power frequency current of this application;

[0026] Figure 7 This is a structural diagram of the processing equipment for this application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0028] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. The naming or numbering of steps in this application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can be changed in the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0029] The division of modules in this application is a logical division. In actual application, there may be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between modules can be electrical or other similar forms, which are not limited in this application. Moreover, the modules or submodules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed into multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application.

[0030] Before introducing the long-term and short-term trend filtering method for power frequency current provided by this application, the background content involved in this application is first introduced.

[0031] The long-term and short-term trend filtering method, device and computer-readable storage medium of industrial frequency current provided in this application can be applied to processing equipment, constructing a filter design that includes a long-term trend module for long-term control cumulative residuals and a short-term trend module for short-term estimated linear trend, and building a specific dual closed-loop control structure in the long-term and short-term trend framework. In addition, further supporting measures are made from the perspective of quantitative formulas, thereby effectively overcoming the limitations of existing technologies and completing current transient analysis in complex scenarios with high precision. It is suitable for real current observation of power supplies and transient analysis of industrial frequency current in specific applications.

[0032] The power frequency current long-term and short-term trend filtering method mentioned in this application can be executed by a power frequency current long-term and short-term trend filtering device, or a server, physical host, or user equipment (UE) of different types of processing equipment that integrates the power frequency current long-term and short-term trend filtering device. Among them, the power frequency current long-term and short-term trend filtering device can be implemented in hardware or software. The UE can specifically be a terminal device such as a smartphone, tablet computer, laptop computer, desktop computer, or personal digital assistant (PDA). The processing device can be set up in a device cluster.

[0033] It should be understood that for the processing equipment that executes the filtering processing method in the time series analysis scenario provided by this application, or the processing equipment equipped with the application service corresponding to the filtering processing method in the time series analysis scenario provided by this application, in actual circumstances, it only needs to have the required data processing capabilities to meet the application requirements and carry out filtering processing on the industrial frequency current to be processed. Therefore, its specific equipment category and equipment deployment form are relatively flexible and can be configured according to actual conditions.

[0034] If further functional services such as real-time collection of industrial frequency current or result display are involved, it is obvious that further adaptive configuration of the processing equipment is required to have the corresponding data processing function.

[0035] Taking result display as an example, the processing device itself can be configured with a display screen (including a touch screen). In addition, the result display purpose can also be achieved through an external display screen device or other device with a display screen.

[0036] Next, the long-term and short-term trend filtering method of power frequency current provided by this application is introduced.

[0037] First, see Figure 1 , Figure 1A flow chart of the long-term and short-term trend filtering method for power frequency current provided by the present application is shown. The long-term and short-term trend filtering method for power frequency current provided by the present application may specifically include the following steps S101 to S103:

[0038] Step S101, obtaining an initial power frequency current to be filtered;

[0039] It can be understood that the present application is aimed at filtering the industrial frequency current. Therefore, when performing the corresponding filtering processing through the filter specially designed by the present application, it is necessary to obtain the industrial frequency current to be filtered (or industrial frequency current data).

[0040] For the convenience of explanation, the currently acquired power frequency current is recorded as the initial power frequency current.

[0041] In practical applications, the acquisition and processing of the initial power frequency current usually involves the extraction and processing of existing data, such as manual entry, local retrieval, remote retrieval, or direct reception of data forwarded by other devices, or real-time data collection and processing.

[0042] In addition, in specific operations, the solution of the present application can be triggered in the form of a corresponding power frequency current filtering task, and the initial power frequency current can be directly carried in the task information, or the task information can also carry guidance on how to indirectly obtain the storage location or collection location of the power frequency current.

[0043] It is understandable that the specific source of the initial power frequency current is relatively flexible, and is usually the power system. Of course, corresponding to the flexible and changeable solution application requirements in actual situations, it can be any form of data source.

[0044] Step S102: Perform long-term and short-term trend filtering on the initial power frequency current through a pre-configured filter, wherein the filter has, in terms of the long-term and short-term trend framework under the data fusion framework, the following: the outer loop structure and the inner loop structure in the dual closed-loop control structure respectively include a long-term trend module for long-term control cumulative residual and a short-term trend module for short-term estimation linearization trend, and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module, the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure;

[0045] refer to Figure 2 The schematic diagram of a framework of the long-term and short-term trend framework of the present application is shown. The filter specifically involved in the present application is composed of two closed-loop control structures under the data fusion framework, that is, the long-term and short-term trend framework is a double closed-loop control structure, and the appearance structure includes a long-term trend module (i.e., a long-term trend module for long-term control of the accumulated residual) Figure 2 Long-Term1 in ) and a short-term trend module for short-term estimation of linearized trend (i.e. Figure 2 Similarly, the inner loop structure includes a long-term trend module that controls the cumulative residuals (i.e. Figure 2 Long-Term2 in ) and a short-term trend module for short-term estimation of linearized trend (i.e. Figure 2 Short-Term2 in ).

[0046] It can be seen that the inner loop structure is nested in the outer loop structure, and the closed-loop structures of the two are highly similar.

[0047] In terms of closed-loop structure design, specifically, in a closed-loop loop, the long-term error output by the long-term trend module directly compensates for the short-term deviation of the short-term trend module. In this application, considering that the long-term trend will introduce residuals at the current moment, the output selects a short-term trend with higher reliability. Under this, the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure.

[0048] The long-term trend module is specifically configured under the design goal of controlling the cumulative residual in the long term, and the short-term trend module is specifically configured under the design goal of estimating the linearized trend in the short term.

[0049] In this way, the present application achieves the design goal of a filter that constructs a long-term trend module for long-term control of cumulative residuals and a short-term trend module for short-term estimation of linear trends, and builds a specific dual-closed-loop control structure in the long-term and short-term trend framework, which can effectively overcome the limitations of the existing technology and complete the current transient analysis in complex scenarios with high precision. It is suitable for the real current observation of power supplies and transient analysis of power frequency current in specific applications.

[0050] Step S103 , extracting the target power frequency current output by the filter after filtering the initial power frequency current.

[0051] After the filter has completed normal operation and filtering of the initial power frequency current, it is easy to understand that the target power frequency current output by the filter after filtering the initial power frequency current can be extracted.

[0052] At this point, further data application processing may be performed in accordance with application requirements.

[0053] For example, the target power frequency current can be stored locally, stored remotely, displayed as a result, output a prompt indicating that filtering is complete, perform corresponding current observation processing, and perform corresponding transient analysis.

[0054] Obviously, the specific data application processing content can be flexibly adjusted according to the pre-configured and real-time configured data application strategies, and this application does not make too many restrictions.

[0055] It can be understood that in the above scheme description, the focus is on introducing the long-term and short-term trend framework constructed by this application in combination with the long-term trend module (or long-term trend module) equipped with a long-term trend feature processing algorithm and the short-term trend module (or short-term trend module) equipped with a short-term trend feature processing algorithm. The corresponding signal long / short-term trend feature extraction method is not unique. In the following scheme description, the specific signal long / short-term trend feature extraction method that can be adopted in the detailed operation is explained in more detail, and further specific support is made from the perspective of quantitative formulas.

[0056] Specifically, corresponding to the long-term filtering structure, the filter may also include the following configuration contents:

[0057] 1) In terms of the long-term trend of the statistical algorithm, the ideal estimated value tracking when the estimated value tracks the sampled value has the following characteristics:

[0058] The sum of the positive residual deviations is equal to the sum of the negative residual deviations;

[0059] Understandably, the predicted values ​​under time series tracking, although the least squares method considers the estimated value of the finite data set With the observed value z k The minimum deviation between Figure 3 A scenario diagram of the ideal trend filtering algorithm of the present application is shown. When the estimated value tracks the sampling value, the more desired ideal estimated value tracking is that the sum of the residuals has a 0 deviation characteristic (that is, the areas of expected region 1 and region 2 are equal), which is equivalent to the sum of the residual positive deviations being equal to the sum of the residual negative deviations.

[0060] Specifically, this feature is expressed as:

[0061] Define a finite sample set z in the time range [0,T], and the corresponding time series is {z0,…z k ,…,z T}, and satisfy the following formula:

[0062]

[0063] Among them, z k is the system output observation value at time k, is the estimated value of the system output at time k.

[0064] This formula also corresponds to the fact that the sum of the residuals above has the characteristic of 0 deviation, which is equivalent to the statement that the sum of the positive deviations of the residuals is equal to the sum of the negative deviations of the residuals.

[0065] 2) In terms of the long-term trend of the cumulative residual control, the characteristics are extended to infinitely long dynamic sequences collected in real time.

[0066] Understandably, how can we evaluate the real-time tracking capability of the predicted value? In signals containing random noise, the transient residual between the predicted value and the true value cannot be evaluated. However, long-term characteristics can still be evaluated based on the cumulative residual. Based on the aforementioned principle that the sum of the positive residual deviations is equal to the sum of the negative residual deviations, this principle can be extended to infinitely long dynamic sequences acquired in real time.

[0067] Specifically, the feature is extended to the infinite-length dynamic sequence collected in real time, that is, the long-term trend feature of the cumulative residual control possessed by the long-term trend module, which can be expressed by the following formula:

[0068] 1) Set IntPos k is the residual positive deviation integral at time k, IntNeg k is the residual negative deviation integral at time k, u 1k is the only input residual of the long-term trend module at time k, satisfying the following formula:

[0069]

[0070] 2) When the module estimate accurately tracks the signal, (the accumulated residual should) satisfy the following formula:

[0071] u 1k >0, IntPos k =u 1k +IntPos k ,

[0072] u 1k When ≤0, IntNeg k =-u 1k +IntNeg k ;

[0073] Right now:

[0074] If u 1k >0,

[0075] IntPos k =u 1k +IntPos k

[0076] Else,

[0077] IntNeg k =-u 1k +IntNeg k

[0078] 3) Accumulating values ​​can avoid calculations on limited data points and ensure the long-term performance of the algorithm. However, continuous accumulation of data can cause data overflow. Therefore, let N be the threshold (the specific value depends on the application scenario or the data overflow protection settings). When the value is greater than N, desaturation is used to prevent cumulative residual overflow using the following formula:

[0079] IntPos k >N or IntNeg k >N,

[0080] Right now:

[0081] If IntPos k >N||IntNeg k >N,

[0082]

[0083] 4) Set IntError k is the cumulative residual, satisfying the following formula:

[0084] IntError k =IntPos k -IntNeg k .

[0085] Under the above settings, the long-term trend characteristics of the prediction algorithm can be transformed from statistical characteristics to control IntError k Method with value 0, IntError k It is the only output of the long-term trend module.

[0086] Note: If real-time tracking is possible, IntError k tends to 0, new data will only introduce [u 1min u 1max ] range, N under stable control must be bounded.

[0087] On the other hand, corresponding to the short-term filtering structure, the filter can also include the following configuration contents:

[0088] In terms of the short-term trend of the prediction algorithm, the local linear algorithm of the estimated value is used to estimate the short-term trend characteristics.

[0089] Specifically, short-term methods have faster tracking capabilities, but cannot identify long-term cumulative deviations. Local linearization short-term methods can improve the real-time performance and reliability of time series tracking. Even when processing large amounts of test data, local linear models are widely used. However, in existing algorithms, the linear approximation of nonlinear systems has defects. When there are too few sampling points, the local linearization of the mean will introduce the influence of random noise. In a low signal-to-noise ratio environment, some methods that focus on input changes will introduce a large amount of random noise. Although the linearization method cannot obtain a high-precision solution, the calculation speed can be simple and fast. However, even more traditional linear algorithms still require more prior conditions in actual code implementation (such as the input function is a linear transformation function, the square criterion of the predicted smoothing function, the standard parameters, etc.). In order to suppress the prior conditions, multiple parameters, parameter sensitivity and other problems introduced by the local linear model, the local linear algorithm of the estimated value is used to estimate the short-term trend characteristics.

[0090] Furthermore, the local linear algorithm of the estimated value estimates the short-term trend characteristics, which can be specifically expressed by the following formula:

[0091] 1) Assume that the system output observation value z at time k k Relatively stable, and can be expressed as a fixed value z as follows c The sum of white noise ε and E(ε) with an expected sum of 0 is:

[0092] z k =z c +ε,E(ε)=0;

[0093] It is worth noting that at the current moment k, only the deviation described in the long-term trend of the cumulative residual control appears. The simplified analysis of the formula here mainly focuses on the trend of the fixed value.

[0094] 2) Assume that the estimated system output at time k is A fixed value z can be accurately approximated c ,have:

[0095]

[0096] Where K is the gain, The smaller the fluctuation (the more accurate), the smaller the gain K must be;

[0097] 3) For the estimation of the change value, it is assumed that the system output observation value z at any time can be locally linearized, z k and z k-1 (i.e., the system output observation value at time k-1) The system output increment observation value Δz in adjacent unit time that satisfies the linear relationship k-1 Equal, there are the following local linear approximations:

[0098] z k =z k-1 +Δz k-1 +ε,E(ε)=0;

[0099] 4) Assumptions (i.e., the estimated system output at time k-1) and (i.e., the estimated value of the system output increment at time k-1), which can accurately track z k-1 and Δz k-1 , then z k Can be Tracking, there are:

[0100]

[0101] Correspondingly, you can also combine Figure 4 The application shown k sequence sum A time domain characteristic diagram of the local linearization of the curve can be used for a more vivid understanding.

[0102] Next, from the above formula, we can know that and The value of short-term trend change can be estimated, and the deviation of short-term estimation can be compensated by two long-term trend modules respectively. Therefore, in this application scheme, a long-term and short-term trend module algorithm combining the long-term trend of the statistical algorithm and the short-term trend of the prediction algorithm is designed. Because the fusion filter can construct the most typical closed-loop control structure, combined with the previous formula There is the above Figure 2 The long-term and short-term trend module structure and long-term and short-term trend framework based on fusion filtering are shown.

[0103] And for Figure 2 , combined with the above formula part, we can learn:

[0104] In the inner ring structure, the long-term trend module is used to modify The long-term trend and short-term trend modules are used for linear approximation

[0105] In the outer ring structure, the long-term trend module is used to modify Long-term trend characteristics, short-term trend module is used for linear approximation

[0106] Correspondingly, in order to ensure real-time performance, the short-term trend module adopts the data fusion structure of the improved model-free prediction algorithm, and the numerical estimation uses the posterior estimation value as much as possible.

[0107] set up is the posterior estimate output by the short-term trend module in the outer loop structure, It is the posterior estimate of the output of the short-term trend module in the inner loop structure.

[0108] In the closed loop structure of the long-term trend module, the solution to the algebraic loop problem uses the output delayed data and Right now, It is the delayed data output by the long-term trend module in the outer loop structure. It is the delayed data output by the long-term trend module in the inner loop structure.

[0109] In this case, for the outer ring structure, the following equation is satisfied:

[0110]

[0111] For the inner ring structure, the following equation is satisfied:

[0112]

[0113] For both, h0 and h1 can be involved, where h0 is based on the processing perception Δz of the moving maximum window k|k-1 , h1 based on the processing perception of the largest moving window

[0114] Furthermore, the steady-state model only involves K kmin , h, but the fixed h is not suitable for the changing noise environment, so this application uses the receptive area to estimate h. The addition of the receptive area is sensitive to outliers, but the addition of the receptive area makes the filtering algorithm have only one setting parameter K kmin , and the dynamic perception h can cope with real-time current in variable noise environments.

[0115] Among them, in complex scenarios, excessive pursuit of trend linearity can easily cause the current prediction value to lag, which is inconsistent with other algorithms. Perturbation is the key to the normal operation of the algorithm of this application. In a variable noise environment, K kmin The choice of suggests keeping a little disturbance.

[0116] Next, the corresponding long-term and short-term trend module framework is Figure 2 Based on this, you can continue to refer to Figure 5 The schematic diagram of a framework of the long-term and short-term trend filter of the model-free prediction filter of the present application is shown. The short-term trend module in the framework includes a Moving Maximum Window (i.e., a moving maximum window, shown as Moving Max), a model-free prediction module, and a moving maximum window for designing a receptive area. n is the number of data in the moving maximum window, and h is the number of z in the moving maximum window. k and The maximum absolute value of the residual error between the two, corresponding to the following formula:

[0117]

[0118] In order to obtain a relatively stable h, a steady-state model without model prediction is used. prediction, h0 and h1 perceive Δz respectively k|k-1 ,

[0119] like Figure 5 As shown, the short-term trend module in the outer ring structure includes Self-prediction module and Forecast module, the short-term trend module in the inner ring structure includes Self-prediction module and Prediction module.

[0120] These four prediction modules or four model-free prediction steady-state models perform different functions, but can adopt a common structure. From the overall point of view, they include four input parameters (z in 、h in , K kmin 、h min ) and two output parameters (y1, y2), that is, there is an input signal z in , input residual h in , set constant K kmin , input the minimum residual value h min , posterior estimate y1, and the deviation y2 between the measured value and the posterior estimate.

[0121] When estimating the steady-state h0 and h1, the value of n should not be too large, otherwise it will affect the real-time performance and be unfavorable for the processing of changing random noise. However, when the value of n is too small, transient perception fails when the residual is actually perceived. In the real current processing, the algorithm adds a constant h min It is necessary, and the algorithm is min The choice of is not sensitive, so h min It can be selected according to actual situation.

[0122] The input parameters of the self-prediction module are h0, h0, K kmin 、h min , y1 output is y2 is not output, this module is used to self-identify the stability (It should be noted here that Figure 5 As shown, the module has two h0 inputs, which are specifically for self-prediction processing services);

[0123] The input parameter of the prediction module is z k 、 K kmin , y1 output is y2 output is Δz k|k-1 , this module is used to estimate the stable And output Δz k|k-1 To further estimate

[0124] The input parameters of the self-prediction module are h1, K kmin 、h min , y1 output is y2 is not output, this module is used to self-identify the stability Selecting h1 for the second module input will reduce real-time performance, so select a faster response However, due to the rapid identification of the first module, h min No numerical limit is set at this time;

[0125] The input parameter of the prediction module is Δz k|k-1 、 K kmin , y1 output is y2 is not output, this module is used to estimate the stable And output Δz k|k-1 Further estimates

[0126] Next, let Q k is the adaptive process noise covariance matrix at time k, R k is the approximate measurement noise covariance matrix at time k, To input the prior estimate, for The corresponding posterior estimate, K k is the self-iteration gain at time k, P k is the variance matrix of the approximate estimation error at time k;

[0127] The four modules repeatedly execute the following formula:

[0128] 1) Stable estimate forecast:

[0129]

[0130] 2) Prior error covariance update:

[0131] P k =P k-1 +Q k-1 ;

[0132] 3) Model-free prediction gain update:

[0133] K k =P k / (P k +R k-1 );

[0134] 4) Update the estimated value of the posterior:

[0135]

[0136] 5) Posterior error covariance update:

[0137] P k =(IK k )P k ;

[0138] 6) Adaptive process noise covariance matrix update:

[0139] hour,

[0140] When, if

[0141] Right now:

[0142]

[0143] 7) K k Even if the module does not trigger, the design should maintain this feature:

[0144] K k <K kmin When K k =K kmin ;

[0145] Right now:

[0146] If K k <K kmin ,

[0147] K k =K kmin

[0148] 8) If yes The self-prediction module or the In the self-prediction module, y1 satisfies the following relationship:

[0149] hour,

[0150] When y1=h min ;

[0151] If it is Prediction module or the In the prediction module, y1 satisfies the following relationship:

[0152]

[0153] 9)

[0154] 10) Update the covariance matrix of measurement noise:

[0155] R k =h in(k) / K kmin ;

[0156] 11) Prior estimate update:

[0157]

[0158] In this way, when the four prediction modules are running in real time, the corresponding formula The output obtained is the following:

[0159]

[0160] Finally, for the above solution content, in general, for the goal of high-performance power frequency current filtering, this application constructs a filter design that includes a long-term trend module for long-term control of cumulative residuals and a short-term trend module for short-term estimation of linear trends, and builds a specific dual closed-loop control structure in the long-term and short-term trend framework. In addition, further supporting measures are made from the perspective of quantitative formulas, which can effectively overcome the limitations of existing technologies and complete current transient analysis in complex scenarios with high precision. It is suitable for real current observation of power supplies and transient analysis of power frequency currents in specific applications.

[0161] The above is an introduction to the long-term and short-term trend filtering method of the power frequency current provided by this application. In order to facilitate better implementation of the long-term and short-term trend filtering method of the power frequency current provided by this application, this application also provides a long-term and short-term trend filtering device for the power frequency current from the perspective of functional modules.

[0162] See Figure 6 , Figure 6 This is a schematic diagram of the structure of the power frequency current long-term and short-term trend filtering device of the present application. In the present application, the power frequency current long-term and short-term trend filtering device 600 may specifically include the following structure:

[0163] An acquisition unit 601 is configured to acquire an initial power frequency current to be filtered;

[0164] The filtering unit 602 is configured to perform long-term and short-term trend filtering on the initial power frequency current through a pre-configured filter, wherein the filter has the following features in terms of a long-term and short-term trend framework under the data fusion framework: the outer loop structure and the inner loop structure in the dual closed-loop control structure respectively include a long-term trend module for long-term control cumulative residuals and a short-term trend module for short-term estimation linearization trend, and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module, the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure;

[0165] The extraction unit 603 is configured to extract the target power frequency current output by the filter after filtering the initial power frequency current.

[0166] In an exemplary embodiment, corresponding to the long-term filtering structure, the filter further includes the following configuration contents:

[0167] In terms of the long-term trend of statistical algorithms, the ideal estimated value tracking when the estimated value tracks the sampled value has the following characteristics:

[0168] The sum of the positive residual deviations is equal to the sum of the negative residual deviations;

[0169] In terms of long-term trending of accumulated residual control, the feature is extended to infinitely long dynamic sequences acquired in real time.

[0170] In another exemplary embodiment, the characteristics are specifically expressed as:

[0171] Define a finite sample set z in the time range [0,T], and the corresponding time series is {z0,…z k ,…,z T}, and satisfy the following formula:

[0172]

[0173] Among them, z k is the system output observation value at time k, is the estimated value of the system output at time k.

[0174] In yet another exemplary embodiment, the characteristic is extended to an infinite length dynamic sequence acquired in real time, as expressed by the following formula:

[0175] Set IntPos k is the residual positive deviation integral at time k, IntNeg k is the residual negative deviation integral at time k, u 1k is the only input residual of the long-term trend module at time k, satisfying the following formula:

[0176]

[0177] When the module estimate accurately tracks the signal, the following equation is satisfied:

[0178] u 1k >0, IntPos k =u 1k +IntPos k ,

[0179] u 1k When ≤0, IntNeg k =-u 1k +IntNeg k ;

[0180] Let N be the threshold. When the value is greater than N, desaturation is used to avoid cumulative residual overflow through the following formula:

[0181] IntPos k >N or IntNeg k >N,

[0182] Let IntError k is the cumulative residual, satisfying the following formula:

[0183] IntError k =IntPos k -IntNeg k .

[0184] In another exemplary embodiment, corresponding to the short-term filtering structure, the filter further includes the following configuration contents:

[0185] In terms of the short-term trend of the prediction algorithm, the local linear algorithm of the estimated value is used to estimate the short-term trend characteristics.

[0186] In yet another exemplary embodiment, the local linear algorithm of the estimated value estimates the short-term trend characteristic, which is expressed as follows:

[0187] Assume that the system output observation value z at time k k Relatively stable, and can be expressed as a fixed value z as follows c The sum of white noise ε and E(ε) with an expected sum of 0 is:

[0188] z k =z c +ε,E(ε)=0;

[0189] Assume that the estimated system output at time k is Approximate fixed value z c ,have:

[0190]

[0191] Where K is the gain;

[0192] Assuming that the system output observation z is locally linearizable, z k and z k-1 The incremental observation value Δz of the system output in adjacent unit time that satisfies the linear relationship k-1 Equal, there are:

[0193] z k =z k-1 +Δz k-1 +ε,E(ε)=0;

[0194] Assumptions and Able to accurately track z k-1 and Δz k-1 ,have:

[0195]

[0196] h0 is based on the processing perception of Δz based on the moving maximum window k|k-1 , h1 based on the processing perception of the largest moving window

[0197] ,set up is the posterior estimate output by the short-term trend module in the outer loop structure, is the posterior estimate output by the short-term trend module in the inner loop structure, It is the delayed data output by the long-term trend module in the outer loop structure. It is the delayed data output by the long-term trend module in the inner loop structure;

[0198] For the outer ring structure, the following equation is satisfied:

[0199]

[0200] For the inner ring structure, the following equation is satisfied:

[0201]

[0202] In another exemplary embodiment, for a dual closed-loop control structure, the filter further includes the following configuration contents:

[0203] Let n be the number of data in the moving maximum window, and h be the number of data in the moving maximum window. k and The maximum absolute value of the residual error between is:

[0204]

[0205] The short-term trend module in the outer ring structure includes Prediction module and Forecast module, the short-term trend module in the inner ring structure includes Prediction module and Prediction module;

[0206] Assume that the input signal z in , input residual h in , set constant K kmin , input the minimum residual value h min , posterior estimate y1, deviation y2 between the measured value and the posterior estimate;

[0207] The input parameters of the self-prediction module are h0, h0, K kmin 、h min , y1 output is y2 is not output;

[0208] The input parameter of the prediction module is z k 、 K kmin , y1 output is y2 output is Δz k|k-1 ;

[0209] The input parameters of the self-prediction module are h1, K kmin 、h min , y1 output is y2 is not output;

[0210] The input parameter of the prediction module is Δz k|k-1 、 K kmin , y1 output is y2 is not output;

[0211] Let Q k is the adaptive process noise covariance matrix at time k, R k is the approximate measurement noise covariance matrix at time k, To input the prior estimate, for The corresponding posterior estimate, K k is the self-iteration gain at time k, P k is the variance matrix of the approximate estimation error at time k;

[0212] The four modules repeatedly execute the following formula:

[0213] 1)

[0214] 2)P k =P k-1 +Q k-1 ;

[0215] 3)K k =P k / (P k +R k-1 );

[0216] 4)

[0217] 5)P k =(IK k )P k ;

[0218] 6) hour,

[0219] When, if

[0220] 7) K k <K kmin When K k =K kmin ;

[0221] 8) If yes Self-prediction module or In the self-prediction module, y1 satisfies the following relationship:

[0222] hour,

[0223] When y1=h min ,

[0224] in the case of Prediction module or In the prediction module, y1 satisfies the following relationship:

[0225]

[0226] 9)

[0227] 10)R k =h in / K kmin ;

[0228] 11)

[0229] When the four prediction modules are running in real time, the output is as follows:

[0230]

[0231] This application also provides a processing device from the perspective of hardware structure, see Figure 7 , Figure 7 The schematic diagram of the structure of the processing device of the present application is shown. Specifically, the processing device of the present application may include a processor 701, a memory 702 and an input / output device 703. The processor 701 is used to execute the computer program stored in the memory 702 to implement the following Figure 1 The steps of the method for filtering the long-term and short-term trends of power frequency current in the corresponding embodiment; or, when the processor 701 is used to execute the computer program stored in the memory 702, the following is implemented: Figure 6 The memory 702 is used to store the functions of each unit in the corresponding embodiment. Figure 1 The computer program required by the long-term and short-term trend filtering method of the power frequency current in the corresponding embodiment.

[0232] For example, the computer program may be divided into one or more modules / units, one or more of which are stored in the memory 702 and executed by the processor 701 to complete the present application. One or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in a computer device.

[0233] The processing device may include, but is not limited to, a processor 701, a memory 702, and an input / output device 703. Those skilled in the art will appreciate that the illustrations are merely examples of processing devices and do not limit the processing device. The processing device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the processing device may also include a network access device, a bus, etc., and the processor 701, the memory 702, the input / output device 703, etc. may be connected via a bus.

[0234] The processor 701 may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the processing device and connects various parts of the entire device using various interfaces and lines.

[0235] The memory 702 can be used to store computer programs and / or modules. The processor 701 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 702 and accessing the data stored in the memory 702. The memory 702 may mainly include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, etc.; the data storage area may store data created based on the use of the processing device, etc. In addition, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0236] When the processor 701 is used to execute the computer program stored in the memory 702, it can specifically implement the following functions:

[0237] Obtaining the initial power frequency current to be filtered;

[0238] The initial power frequency current is subjected to long-term and short-term trend filtering through a pre-configured filter. The filter, in terms of the long-term and short-term trend framework under the data fusion framework, includes: an outer loop structure and an inner loop structure in the dual closed-loop control structure, each including a long-term trend module for long-term control cumulative residuals and a short-term trend module for short-term estimation linearization trend; and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module; the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure;

[0239] The target power frequency current after filtering the initial power frequency current output by the filter is extracted.

[0240] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the power frequency current long-term and short-term trend filtering device, processing equipment and corresponding units described above can refer to the following. Figure 1 The description of the long-term and short-term trend filtering method of the power frequency current in the corresponding embodiment will not be repeated here.

[0241] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0242] To this end, the present application provides a computer-readable storage medium, which stores a plurality of instructions, which can be loaded by a processor to execute the present application as follows: Figure 1 The steps of the power frequency current long-term and short-term trend filtering method in the corresponding embodiment, the specific operation can be referred to as follows Figure 1 The description of the long-term and short-term trend filtering method of the power frequency current in the corresponding embodiment will not be repeated here.

[0243] The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0244] Due to the instructions stored in the computer readable storage medium, the present application can be executed as follows: Figure 1 The steps of the power frequency current long-term and short-term trend filtering method in the corresponding embodiment, therefore, the present application can be realized as follows Figure 1 The beneficial effects that can be achieved by the long-term and short-term trend filtering method for power frequency current in the corresponding embodiment are detailed in the previous description and will not be repeated here.

[0245] The above is a detailed introduction to the long-term and short-term trend filtering method, device, processing equipment and computer-readable storage medium of the industrial frequency current provided by this application. Specific examples are used in this article to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the core idea of ​​this application; at the same time, for technical personnel in this field, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this application.

Claims

1. A method for filtering long-term and short-term trends of power frequency current, characterized in that: The method comprises: Obtaining the initial power frequency current to be filtered; The initial power frequency current is subjected to long-term and short-term trend filtering processing by a pre-configured filter, wherein, in terms of the long-term and short-term trend framework of the filter under the data fusion framework, the outer loop structure and the inner loop structure in the dual closed-loop control structure respectively include a long-term trend module for long-term control cumulative residual and a short-term trend module for short-term estimation linearization trend, and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module, the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure; A target power frequency current is extracted after the initial power frequency current is filtered and output by the filter.

2. The method according to claim 1, characterized in that Corresponding to the long-term filtering structure, the filter also includes the following configuration content: In terms of the long-term trend of statistical algorithms, the ideal estimated value tracking when the estimated value tracks the sampled value has the following characteristics: The sum of the positive residual deviations is equal to the sum of the negative residual deviations; In terms of the long-term trend of the accumulated residual control, the described properties are extended to infinitely long dynamic sequences acquired in real time.

3. The method according to claim 2, characterized in that The characteristics are specifically expressed as follows: Define a finite sample set z in the time range [0,T], and the corresponding time series is {z0,…z k ,…,z T }, and satisfy the following formula: Among them, z k is the system output observation value at time k, is the estimated value of the system output at time k.

4. The method according to claim 3, characterized in that The extension of the above characteristics to infinite length dynamic sequences acquired in real time is expressed by the following formula: Set IntPos k is the residual positive deviation integral at time k, IntNeg k is the residual negative deviation integral at time k, u 1k is the only input residual of the long-term trend module at time k, satisfying the following formula: When the module estimate accurately tracks the signal, the following equation is satisfied: u 1k When > 0, IntPos k = u 1k + IntPos k , u 1k When ≤ 0, IntNeg k = -u 1k + IntNeg k ; Let N be the threshold. When the value is greater than N, desaturation is used to avoid cumulative residual overflow through the following formula: IntPos k >N or IntNeg k >N, Let IntError k is the cumulative residual, satisfying the following formula: IntError k =IntPos k -IntNeg k 。 5. The method according to claim 1, characterized in that Corresponding to the short-term filtering structure, the filter also includes the following configuration contents: In terms of the short-term trend of the prediction algorithm, the local linear algorithm of the estimated value is used to estimate the short-term trend characteristics.

6. The method according to claim 5, characterized in that The local linear algorithm of the estimated value estimates the short-term trend characteristics, which is expressed as follows: Assume that the system output observation value z at time k k Relatively stable, and can be expressed as a fixed value z as follows c The sum of white noise ε and E(ε) with an expected sum of 0 is: z k =z c +ε,E(ε)=0; Assume that the estimated system output at time k is Approximate the fixed value z c ,have: Where K is the gain; Assuming that the system output observation z is locally linearizable, z k and z k-1 The incremental observation value Δz of the system output in adjacent unit time that satisfies the linear relationship k-1 Equal, there are: z k =z k-1 +Δz k-1 +ε,E(ε)=0; Assumptions and Able to accurately track z k-1 and Δz k-1 ,have:

7. The method according to claim 6, characterized in that set up is the posterior estimate output by the short-term trend module in the outer loop structure, is the posterior estimate output by the short-term trend module in the inner loop structure, It is the delayed data output by the long-term trend module in the outer loop structure. It is the delayed data output by the long-term trend module in the inner loop structure; For the outer ring structure, the following formula is satisfied: For the inner ring structure, the following formula is satisfied: h0 is based on the processing perception of Δz based on the moving maximum window k|k-1 , h1 is based on the processing perception of the moving maximum window Let n be the number of data in the moving maximum window, h be the number of data in the moving maximum window. k and The maximum absolute value of the residual error between is: The short-term trend module in the outer ring structure includes Prediction module and Prediction module, the short-term trend module in the inner loop structure includes Prediction module and Prediction module; Assume that the input signal z in , input residual h in , set constant K kmin , input the minimum residual value h min , posterior estimate y1, deviation y2 between the measured value and the posterior estimate; described The input parameters of the self-prediction module are h0, h0, K kmin 、h min , y1 output is y2 is not output; described The input parameter of the prediction module is z k 、 K kmin , y1 output is y2 output is Δz k|k-1 ; described The input parameters of the self-prediction module are h1, K kmin 、h min , y1 output is y2 is not output; described The input parameter of the prediction module is Δz k|k-1 、 K kmin , y1 output is y2 is not output; Let Q k is the adaptive process noise covariance matrix at time k, R k is the approximate measurement noise covariance matrix at time k, To input the prior estimate, for The corresponding posterior estimate, K k is the self-iteration gain at time k, P k is the variance matrix of the approximate estimation error at time k; The four modules repeatedly execute the following formula: 1) 2)P k =P k-1 +Q k-1 ; 3)K k =P k / (P k +R k-1 ); 4) 5)P k =(I-K k )P k ; 6) hour, When, if 7) K k <K kmin When K k =K kmin ; 8) If it is the The self-prediction module or the In the self-prediction module, y1 satisfies the following relationship: hour, When y1=h min , If it is Prediction module or the In the prediction module, y1 satisfies the following relationship: 9) 10)R k =h in(k) / K kmin ; 11) When the four prediction modules are running in real time, the output is as follows:

8. A power frequency current long-term and short-term trend filtering device, characterized in that: The device comprises: An acquisition unit, used for acquiring an initial power frequency current to be filtered; A filtering unit is configured to perform long-term and short-term trend filtering processing on the initial power frequency current through a pre-configured filter, wherein, in terms of the long-term and short-term trend framework of the filter under the data fusion framework, the outer loop structure and the inner loop structure in the dual closed-loop control structure respectively include a long-term trend module for long-term control cumulative residual and a short-term trend module for short-term estimation linearization trend, and in a closed-loop loop, the long-term error output by the long-term trend module directly compensates the short-term deviation of the short-term trend module, the output of the short-term trend module in the outer loop structure serves as the input of the inner loop structure, and the sum of the outputs of the two short-term trend modules serves as the output of the dual closed-loop control structure; An extraction unit is used to extract the target power frequency current output by the filter after filtering the initial power frequency current.

9. A processing device, characterized in that The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method according to any one of claims 1 to 7 is executed.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 7.

Images