Concentrator with power-off intelligent recovery function

Abstract

The present application relates to the technical field of concentrator, and particularly relates to a concentrator with power-off intelligent recovery function, which comprises a data acquisition module, which is used for acquiring a denoised neighborhood power data point sequence corresponding to a power-off time stage of the concentrator; a prediction weight acquisition module, which is used for acquiring a prediction weight value of a denoised left neighborhood power data point sequence and a prediction weight value of a denoised right neighborhood power data point sequence; and a prediction module, which is used for obtaining a target prediction power data in the power-off time stage according to the prediction weight values of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence. The present application can improve the accuracy of power prediction in the power-off time stage, thereby improving the accuracy of power prediction interpolation in the power-off time stage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of concentrators, and in particular to a concentrator with intelligent power-off recovery function. BACKGROUND

[0002] The concentrator is an electric energy data centralized collection and communication management device installed at the lower end of the power distribution transformer or the building low-voltage incoming line, and is a bridge between the intelligent electric meter collection terminal and the master station system. At present, the concentrator will be powered off due to environmental and concentrator faults, and during the power-off period, the concentrator collection data will be missing. In order to ensure data integrity, maintain system cooperative operation, avoid misjudgment and misoperation, etc., it is usually necessary to interpolate the electric energy in the power-off stage, and the ARIMA prediction model is usually used to predict the electric energy in the power-off stage, and then the predicted data is interpolated into the corresponding position. The data used for prediction generally is the data before the power-off stage. However, the data used for prediction in the existing prediction interpolation method not only contains noise data, but also has the problem that the accuracy of the prediction result based on the data before the power-off stage is low, thereby resulting in low accuracy or reliability of the predicted and interpolated data in the power-off stage. Therefore, how to improve the accuracy or reliability of the prediction and interpolation has become a problem to be solved. SUMMARY

[0003] In order to solve the above problems, the present application provides a concentrator with intelligent power-off recovery function, and the technical scheme adopted is as follows:

[0004] One embodiment of the present application provides a concentrator with intelligent power-off recovery function, which comprises:

[0005] A data acquisition module is configured to acquire a denoised neighborhood electric energy data point sequence corresponding to a power-off time stage of the concentrator, wherein the denoised neighborhood electric energy data point sequence comprises a denoised left neighborhood electric energy data point sequence and a denoised right neighborhood electric energy data point sequence.

[0006] A prediction weight acquisition module is configured to obtain a prediction credibility of the denoised neighborhood electric energy data point sequence according to a trend component data point sequence and a periodic component data point sequence obtained by time series decomposition on the denoised neighborhood electric energy data point sequence, and a peak point on a frequency domain signal obtained by frequency domain conversion on the periodic component data point sequence, and obtain a prediction weight value of the denoised left neighborhood electric energy data point sequence and a prediction weight value of the denoised right neighborhood electric energy data point sequence according to the prediction credibility of the denoised left neighborhood electric energy data point sequence and the denoised right neighborhood electric energy data point sequence.

[0007] A prediction module is configured to obtain target predicted power data in the power-off time stage according to the prediction weight values of the denoised left-neighborhood power data point sequence and the denoised right-neighborhood power data point sequence.

[0008] Beneficial effects: The application comprises a data acquisition module configured to acquire a denoised neighborhood power data point sequence corresponding to a power-off time stage of a concentrator; a prediction weight acquisition module configured to obtain prediction reliability of the denoised neighborhood power data point sequence according to a trend component data point sequence and a periodic component data point sequence obtained by time series decomposition of the denoised neighborhood power data point sequence and a peak point on a frequency domain signal obtained by frequency domain conversion of the periodic component data point sequence, and obtain prediction weight values of the denoised left-neighborhood power data point sequence and the denoised right-neighborhood power data point sequence according to the prediction reliability of the denoised left-neighborhood power data point sequence and the denoised right-neighborhood power data point sequence; and a prediction module configured to obtain target predicted power data in the power-off time stage according to the prediction weight values of the denoised left-neighborhood power data point sequence and the denoised right-neighborhood power data point sequence. Based on the denoised left-neighborhood power data point sequence and the denoised right-neighborhood power data point sequence and the prediction weight values of the denoised left-neighborhood power data point sequence and the denoised right-neighborhood power data point sequence, the application can improve the accuracy of power prediction in the power-off time stage, thereby improving the accuracy of power prediction and interpolation in the power-off time stage. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, below, a brief introduction will be given to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings described below are only some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0010] Figure 1 The structural block diagram of the concentrator with power-off intelligent recovery function is shown in the figure. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, and not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the embodiments of the present application.

[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0013] This embodiment provides a concentrator with intelligent power failure recovery function, as detailed below:

[0014] like Figure 1 As shown, this embodiment provides a concentrator with intelligent power failure recovery function, comprising:

[0015] Data acquisition module 01 is used to acquire the sequence of denoised neighborhood power data points corresponding to the power outage period of the concentrator.

[0016] The purpose of this embodiment is to improve the accuracy of ARIMA model in predicting power consumption during concentrator outages. This embodiment mainly improves the accuracy of power consumption prediction during outages by denoising the data used for prediction and then analyzing the weights of the data before and after the outage in relation to the prediction. In other words, after denoising, a bidirectional prediction method is used to predict power consumption during outages, thereby improving the accuracy of power consumption prediction during outages and thus improving the accuracy and reliability of power consumption prediction interpolation during outages. Since the energy prediction method is the same for each concentrator during the power outage period in this embodiment, and for ease of understanding, this embodiment will subsequently describe the energy prediction process for any power outage period as an example. That is, all subsequent power outage periods are the same power outage period, and all subsequent concentrators are the same concentrator. In addition, predicting the energy during the power outage period is essentially predicting the actual energy metering data generated by the energy meter during the power outage period. Since a concentrator generally manages multiple energy meters, it is necessary to predict the actual energy metering data generated by all energy meters managed by the concentrator during the power outage period. However, since the process of predicting the actual energy metering data generated by different energy meters during the power outage period is consistent in this embodiment, for ease of understanding, this embodiment will subsequently describe the process of predicting the actual energy metering data generated by any energy meter Q managed by the concentrator during the power outage period as an example. The energy metering data will be referred to as energy data. That is, the energy data used in the subsequent prediction and the predicted energy data in this embodiment all belong to energy meter Q.

[0017] In this embodiment, the prediction of power data during the power outage period requires data from before and after the power outage. Therefore, this embodiment needs to first obtain the left neighbor power data point sequence corresponding to the left neighbor time stage and the right neighbor power data point sequence corresponding to the right neighbor time stage during the power outage period. The specific process for obtaining the left neighbor power data point sequence corresponding to the left neighbor time stage and the right neighbor power data point sequence corresponding to the right neighbor time stage during the power outage period is as follows:

[0018] Firstly, the left-neighbor time phase and the right-neighbor time phase of the power-off time phase are obtained, and the left-neighbor time phase is located in front of the power-off time phase and adjacent to the power-off time phase in time, and the right-neighbor time phase is located behind the power-off time phase and adjacent to the power-off time phase in time; in addition, the length of the left-neighbor time phase and the right-neighbor time phase is set according to the data length frequently used when the ARIMA model is used for prediction, and since the data length used for prediction is generally 3 to 5 times the length of the data to be predicted when the ARIMA model is used for prediction, for example, the data length used for prediction can be selected as 5 times the length of the data to be predicted, and then the time length of the left-neighbor time phase and the right-neighbor time phase in the embodiment is 5 times the length of the power-off time phase.

[0019] After obtaining the left-neighbor time phase and the right-neighbor time phase of the power-off time phase, all the electric energy data of the electric energy meter Q collected by the concentrator in the left-neighbor time phase is obtained, and all the electric energy data of the electric energy meter Q collected by the concentrator in the left-neighbor time phase is arranged in the order of collection, and the arrangement result is recorded as the left-neighbor electric energy sequence corresponding to the left-neighbor time phase, and all the electric energy data of the electric energy meter Q collected by the concentrator in the right-neighbor time phase is obtained, and all the electric energy data of the electric energy meter Q collected by the concentrator in the right-neighbor time phase is arranged in the order of collection, and the arrangement result is recorded as the right-neighbor electric energy sequence corresponding to the right-neighbor time phase; then a two-dimensional coordinate system is constructed, the horizontal axis of the two-dimensional coordinate system represents time, and the vertical axis represents electric energy data, then each electric energy data in the left-neighbor electric energy sequence and the right-neighbor electric energy sequence and the time when the corresponding electric energy data is collected are mapped into the two-dimensional coordinate system, to obtain the electric energy data points corresponding to each electric energy data in the left-neighbor electric energy sequence and the electric energy data points corresponding to each electric energy data in the right-neighbor electric energy sequence, and the time sequence sequence constructed by the electric energy data points corresponding to all the electric energy data in the left-neighbor electric energy sequence is recorded as the left-neighbor electric energy data point sequence corresponding to the left-neighbor time phase, and the time sequence sequence constructed by the electric energy data points corresponding to all the electric energy data in the right-neighbor electric energy sequence is recorded as the right-neighbor electric energy data point sequence corresponding to the right-neighbor time phase, and the left-neighbor electric energy data point sequence and the right-neighbor electric energy data point sequence both belong to the neighbor electric energy data point sequence.

[0020] Since the concentrator will generate noise data in the process of collecting data, and the existence of noise data will affect the prediction accuracy, it is necessary to carry out denoising processing before prediction to obtain the denoised neighborhood power data point sequence corresponding to the power-off time stage, so as to exclude the interference of noise on prediction; and since compared with the non-noise data points collected by the conventional concentrator, the noise data points usually have sudden jump or outlier characteristics, the sudden jump refers to the large difference between the data points and the adjacent data points, and the outlier refers to the large difference in the degree of change with the nearby data, therefore, the embodiment will analyze the sudden jump or outlier characteristics of the power data points in the neighborhood power data point sequence to screen out the noise data points in the neighborhood power data point sequence, and carry out noise removal to obtain the denoised neighborhood power data point sequence, and the specific process is as follows:

[0021] Firstly, according to the distance between the power data points in the left neighborhood power data point sequence and the adjacent power data points of the corresponding power data points, the noise data points in the left neighborhood power data point sequence are obtained, and the noise data points in the left neighborhood power data point sequence are replaced by interpolation, and the sequence after the noise data points are replaced is denoted as the denoised left neighborhood power data point sequence of the left neighborhood power data point sequence; and the denoised right neighborhood power data point sequence of the right neighborhood power data point sequence is obtained in the same way as the denoised left neighborhood power data point sequence of the left neighborhood power data point sequence, so the subsequent process of obtaining the denoised right neighborhood power data point sequence will not be described, and the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence both belong to the denoised neighborhood power data point sequence corresponding to the power-off time stage of the concentrator; in this embodiment, the implementer can select the method of replacing the noise data points in the left neighborhood power data point sequence by interpolation according to the actual situation, such as selecting the linear interpolation or moving average method to replace the noise data points in the neighborhood power data point sequence by interpolation.

[0022] In this embodiment, according to the distance between the power data points in the left neighborhood power data point sequence and the adjacent power data points of the corresponding power data points, the specific process of obtaining the noise data points in the left neighborhood power data point sequence is as follows: firstly, according to the adjacent power data points of each power data point in the left neighborhood power data point sequence, the neighborhood power distance of the power data points in the left neighborhood power data point sequence is obtained, and then according to the neighborhood power distance of each power data point in the left neighborhood power data point sequence, the noise index value of the power data points in the left neighborhood power data point sequence is obtained, and then the noise data points in the left neighborhood power data point sequence are obtained according to the noise index value.

[0023] In the embodiment, the specific process of obtaining the neighborhood energy distance of the power energy data point in the left neighborhood power energy data point sequence is as follows: for the gth power energy data point in the left neighborhood power energy data point sequence, if g is 1, the Euclidean distance between the gth power energy data point and the g+1th power energy data point in the left neighborhood power energy data point sequence is taken as the neighborhood energy distance of the gth power energy data point; if g is A, the Euclidean distance between the g-1th power energy data point and the gth power energy data point in the left neighborhood power energy data point sequence is taken as the neighborhood energy distance of the gth power energy data point; if g is neither 1 nor A, the sum of the Euclidean distance between the gth power energy data point and the g-1th power energy data point in the left neighborhood power energy data point sequence and the Euclidean distance between the gth power energy data point and the g+1th power energy data point in the left neighborhood power energy data point sequence is taken as the neighborhood energy distance of the gth power energy data point, that is, the sum of D1 and D2 is the neighborhood energy distance of the gth power energy data point, A is the total number of power energy data points in the left neighborhood power energy data point sequence, D1 is the Euclidean distance between the gth power energy data point and the g-1th power energy data point in the left neighborhood power energy data point sequence, and D2 is the Euclidean distance between the gth power energy data point and the g+1th power energy data point in the left neighborhood power energy data point sequence.

[0024] In the embodiment, the specific process of obtaining the noise index value of the power energy data point in the left neighborhood power energy data point sequence according to the neighborhood energy distance of each power energy data point in the left neighborhood power energy data point sequence is as follows: for the ath power energy data point in the left neighborhood power energy data point sequence, a is greater than 1 and less than A: first, the absolute value of the difference between the neighborhood energy distance of the a-1th power energy data point and the neighborhood energy distance of the ath power energy data point in the left neighborhood power energy data point sequence is calculated and recorded as a first difference value, the absolute value of the difference between the neighborhood energy distance of the a+1th power energy data point and the neighborhood energy distance of the ath power energy data point in the left neighborhood power energy data point sequence is calculated and recorded as a second difference value, then the sum of the first difference value and the second difference value is calculated, the sum of the first difference value and the second difference value is normalized, and the result of the normalization is recorded as a comprehensive difference value, the neighborhood energy distance of the ath power energy data point is normalized, and the result of the normalization is recorded as a normalized neighborhood energy distance, and the sum of the normalized neighborhood energy distance and the comprehensive difference value is taken as the noise index value of the ath power energy data point; and the specific calculation expression of the noise index value of the ath power energy data point is as follows:

[0025]

[0026] wherein, is the noise index value of the ath power energy data point, is a value between 0 and 1, Norm() is a normalization function, is the neighborhood energy distance of the a-th energy data point in the left neighborhood energy data point sequence, is the neighborhood energy distance of the a-1-th energy data point in the left neighborhood energy data point sequence, is the neighborhood energy distance of the a+1-th energy data point in the left neighborhood energy data point sequence. And when is greater, the more obvious the sudden jump feature of the a-th energy data point is, and when is greater, the more obvious the outlier feature of the a-th energy data point is, and when and is greater, is greater, so when is greater, the more likely the a-th energy data point is noise, and vice versa; in addition, the first and last energy data points in the left neighborhood energy data point sequence are not calculated for noise index values and noise judgment.

[0027] In the embodiment, the specific process of obtaining noise data points in the left neighborhood energy data point sequence according to the noise index value is as follows: for the a-th energy data point in the left neighborhood energy data point sequence, it is judged whether the noise index value of the a-th energy data point is greater than the preset noise threshold, if yes, it is determined that the a-th energy data point is a noise data point, and the a-th energy data point is recorded as a noise data point, otherwise, it is determined that the a-th energy data point is not a noise data point; and in specific application, the implementer can set the preset noise threshold according to the actual situation, experimental statistics and the value range of the noise index value, for example, in the embodiment, the data points with noise point features and the data points with normal fluctuation features are divided at both ends of the noise index value value range by normalization, so as to ensure the reliability of noise data point screening, the preset noise threshold can be set to 0.5.

[0028] Therefore, the de-noised left neighborhood energy data point sequence and the de-noised right neighborhood energy data point sequence corresponding to the power-off time stage of the concentrator can be obtained by the above process.

[0029] The prediction weight obtaining module 02 is configured to obtain the prediction reliability of the denoised neighborhood power data point sequence according to the trend component data point sequence and the periodic component data point sequence obtained by performing time series decomposition on the denoised neighborhood power data point sequence and the peak point on the frequency domain signal obtained by performing frequency domain conversion on the periodic component data point sequence, and obtain the prediction weight value of the denoised left neighborhood power data point sequence and the prediction weight value of the denoised right neighborhood power data point sequence according to the prediction reliability of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence.

[0030] In order to improve the accuracy of prediction, the embodiment selects a bidirectional prediction mode to predict the power data in the power-off time stage, that is, the embodiment predicts the power data in the power-off time stage from two directions of before power-off and after power-off. Since the prediction reliabilities of the prediction results obtained by predicting the data before power-off and the data after power-off are different, the embodiment needs to analyze the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence and obtain the prediction reliability of the denoised left neighborhood power data point sequence and the prediction reliability of the denoised right neighborhood power data point sequence based on the analysis result, and determine the prediction weight value based on the obtained prediction reliabilities. The prediction weight value can reflect the contribution of the prediction results in different directions to the final prediction result, that is, the contribution of the prediction result with high prediction reliability is greater, so as to improve the prediction accuracy. It can be known from the above analysis that the embodiment will first obtain the prediction reliability of the denoised left neighborhood power data point sequence according to the trend component data point sequence and the periodic component data point sequence obtained by performing time series decomposition on the denoised left neighborhood power data point sequence and the peak point on the frequency domain signal obtained by performing frequency domain conversion on the periodic component data point sequence, and obtain the prediction reliability of the denoised right neighborhood power data point sequence according to the trend component data point sequence and the periodic component data point sequence obtained by performing time series decomposition on the denoised right neighborhood power data point sequence and the peak point on the frequency domain signal obtained by performing frequency domain conversion on the periodic component data point sequence. Since the method for obtaining the prediction reliability of the denoised right neighborhood power data point sequence is the same as the method for obtaining the prediction reliability of the denoised left neighborhood power data point sequence, the embodiment will only describe the process of obtaining the prediction reliability of the denoised left neighborhood power data point sequence. The specific process of obtaining the prediction reliability of the denoised left neighborhood power data point sequence is as follows.

[0031] Firstly, the STL time series decomposition is performed on the denoised left-neighborhood power data point sequence to obtain a trend component data point sequence and a periodic component data point sequence corresponding to the denoised left-neighborhood power data point sequence, and the trend component data point sequence corresponding to the denoised left-neighborhood power data point sequence is recorded as a first trend sequence, and the periodic component data point sequence corresponding to the denoised left-neighborhood power data point sequence is recorded as a first periodic sequence, and the STL time series decomposition on the denoised left-neighborhood power data point sequence is actually the STL time series decomposition on a time series sequence composed of the vertical coordinates of all the power data points in the denoised left-neighborhood power data point sequence, the vertical coordinate being the power data, the horizontal coordinate value of the hth trend component data point in the trend component data point sequence corresponding to the denoised left-neighborhood power data point sequence being the horizontal coordinate value of the hth power data point in the denoised left-neighborhood power data point sequence, the vertical coordinate value of the hth trend component data point being the trend component obtained by performing the STL time series decomposition on the vertical coordinate value of the hth power data point in the denoised left-neighborhood power data point sequence, the horizontal coordinate value of the hth periodic component data point in the periodic component data point sequence corresponding to the denoised left-neighborhood power data point sequence being the horizontal coordinate value of the hth power data point in the denoised left-neighborhood power data point sequence, and the vertical coordinate value of the hth periodic component data point being the periodic component obtained by performing the STL time series decomposition on the vertical coordinate value of the hth power data point in the denoised left-neighborhood power data point sequence, and the horizontal coordinates of all the points in the embodiment represent time; since the process of time series decomposition is known, the embodiment will not be described in detail.

[0032] After obtaining the first trend sequence, the inflection points in the first trend sequence are obtained, and the process of obtaining the inflection points is known; then, according to the distances between the trend component data points and the adjacent trend component data points in the first trend sequence and the inflection points in the first trend sequence, the trend item stability characteristic value of the denoised left-neighborhood power data point sequence is obtained, and the trend item stability characteristic value is a key index for determining the prediction reliability of the denoised left-neighborhood power data point sequence, and then the specific process of obtaining the trend item stability characteristic value of the denoised left-neighborhood power data point sequence is as follows: firstly, the neighborhood trend distance of each trend component data point in the first trend sequence is obtained according to the Euclidean distance between each trend component data point in the first trend sequence and the adjacent trend component data point corresponding to the trend component data point; then, the neighborhood slope difference value of each inflection point in the first trend sequence is obtained, and the trend item stability characteristic value of the denoised left-neighborhood power data point sequence is obtained according to the neighborhood trend distance of the trend component data point and the neighborhood slope difference value of the inflection point in the first trend sequence.

[0033] In the embodiment, the specific obtaining process of the neighborhood trend distance of the trend component data point is as follows: for the kth trend component data point in the first trend sequence, if k is 1, the Euclidean distance between the kth trend component data point and the k+1th trend component data point in the first trend sequence is taken as the neighborhood trend distance of the kth trend component data point; if k is K0, the Euclidean distance between the k-1th trend component data point and the kth trend component data point in the first trend sequence is taken as the neighborhood trend distance of the kth trend component data point; if k is neither 1 nor K0, the sum of the Euclidean distance between the kth trend component data point and the k-1th trend component data point in the first trend sequence and the Euclidean distance between the kth trend component data point and the k+1th trend component data point in the first trend sequence is taken as the neighborhood trend distance of the kth trend component data point, that is, the sum of D3 and D4 is the neighborhood trend distance of the kth trend component data point, K0 is the total number of trend component data points in the first trend sequence, D3 is the Euclidean distance between the kth trend component data point and the k-1th trend component data point in the first trend sequence, and D4 is the Euclidean distance between the kth trend component data point and the k+1th trend component data point in the first trend sequence.

[0034] In the embodiment, the specific obtaining process of the neighborhood slope difference value of each inflection point in the first trend sequence is as follows: for any inflection point in the first trend sequence, in the first trend sequence, the trend component data point located on the left side of the inflection point and adjacent to the inflection point is obtained and recorded as the left adjacent trend component data point of the inflection point, the trend component data point located on the right side of the inflection point and adjacent to the inflection point is obtained and recorded as the right adjacent trend component data point of the inflection point, the slope between the left adjacent trend component data point and the inflection point is calculated and recorded as the first slope of the inflection point, the slope between the right adjacent trend component data point and the inflection point is calculated and recorded as the second slope of the inflection point, and the absolute value of the difference between the first slope and the second slope is calculated and taken as the neighborhood slope difference value of the inflection point; the calculation method of the slope between any two points is a known technology.

[0035] In the embodiment, according to the neighborhood trend distance of the trend component data points in the first trend sequence and the neighborhood slope difference value of the inflection points, the specific process of obtaining the trend item stability characteristic value of the denoised left neighborhood electric energy data point sequence is as follows: the mean value of the neighborhood trend distance of all the trend component data points in the first trend sequence is calculated and recorded as the neighborhood trend distance mean value, the accumulation result of the neighborhood slope difference value of all the inflection points in the first trend sequence is calculated and recorded as the comprehensive neighborhood slope difference value, the neighborhood trend distance mean value is multiplied by the comprehensive neighborhood slope difference value, and the multiplication result is negatively correlated, and the mapping result is recorded as the trend item stability characteristic value of the denoised left neighborhood electric energy data point sequence; and the calculation expression of the trend item stability characteristic value of the denoised left neighborhood electric energy data point sequence is as follows:

[0036]

[0037] wherein R1 is the trend item stability characteristic value of the denoised left neighborhood electric energy data point sequence, is the neighborhood trend distance of the kth trend component data point in the first trend sequence, K0 is the total number of the trend component data points in the first trend sequence, and K1 is the number of the inflection points in the first trend sequence, is the first slope of the ith inflection point in the first trend sequence, is the second slope of the ith inflection point in the first trend sequence, and exp() is an exponential function with constant e as the base, which is used for negative correlation mapping. When is smaller, it indicates that the trend item of the denoised left neighborhood electric energy data point sequence is more stable as a whole. is smaller, it indicates that the curve composed of the data points in the first trend sequence of the denoised left neighborhood electric energy data point sequence is smoother. When the trend item of the denoised left neighborhood electric energy data point sequence is more stable as a whole and the curve composed of the data points in the first trend sequence of the denoised left neighborhood electric energy data point sequence is smoother, it indicates that the continuous similarity feature between the distribution trend of the electric energy data in the left neighborhood time stage and the distribution trend of the electric energy data in the power-off time stage is more obvious. When the continuous similarity feature is more obvious, it indicates that the regularity stability of the electric energy data in the left neighborhood time stage is higher, that is, the regularity stability of the denoised left neighborhood electric energy data point sequence is higher. When the regularity stability of the denoised left neighborhood electric energy data point sequence is higher, the credibility of the prediction result obtained based on the denoised left neighborhood electric energy data point sequence is higher. Therefore, when is smaller, that is, R1 is larger, it indicates that the credibility of the prediction result obtained based on the denoised left neighborhood electric energy data point sequence is higher. Conversely, when R1 is smaller, it indicates that the credibility of the prediction result obtained based on the denoised left neighborhood electric energy data point sequence is lower.

[0038] After the trend item stable characteristic value is obtained, Fourier transform is performed on the first periodic sequence, and a frequency domain signal obtained by the Fourier transform is recorded as a first frequency domain signal. Then, a frequency corresponding to a peak point with the largest amplitude on the first frequency domain signal is obtained and recorded as a main frequency. An inverse of the main frequency is obtained and used as a first main period. Then, the first periodic sequence is divided by using the first main period to obtain each subsequence on the first periodic sequence. The purpose of dividing the first periodic sequence is to analyze the periodic item stability, and the periodic item stability is a key index for determining the prediction reliability of the denoising left-neighborhood power data point sequence. In addition, an example of dividing the first periodic sequence by using the first main period to obtain each subsequence on the first periodic sequence is as follows: if the time length of the first main period is T, then the periodic component data points in the first periodic sequence are sequentially connected in chronological order, and the connected curve is recorded as a first periodic curve. Starting from a starting periodic component data point on the first periodic curve, the first periodic curve is non-overlappingly divided by using the time length T to obtain all subcurve segments on the first periodic curve. A sequence composed of all periodic component data points on each subcurve segment in chronological order is recorded as a subsequence corresponding to the corresponding subcurve segment, and is also recorded as each subsequence on the first periodic sequence. That is, a subsequence corresponding to the jth subcurve segment on the first periodic curve is the jth subsequence on the first periodic sequence. In addition, except for the last subsequence on the first periodic sequence, the time length of the other subsequences is T, and the time length of the last division time period on the first periodic sequence is less than or equal to T. The time length of the subsequence refers to the time interval between the starting point of the subcurve segment corresponding to the subsequence and the last point of the subcurve segment corresponding to the subsequence, which is also the horizontal coordinate difference.

[0039] Then, the periodic item stable characteristic value of the denoising left-neighborhood power data point sequence is obtained according to the periodic component difference between the periodic component data points with the same position in the adjacent subsequences on the first periodic sequence. The specific calculation process of the periodic item stable characteristic value of the denoising left-neighborhood power data point sequence is as follows:

[0040] First, the neighborhood periodic component difference values of each periodic component data point in each subsequence on the first periodic sequence are obtained. Then, for any subsequence, the cumulative sum of the neighborhood periodic component difference values corresponding to all periodic component data points in the subsequence is recorded as the comprehensive neighborhood periodic component difference value of the subsequence. The mean value of the comprehensive neighborhood periodic component difference values of all subsequences on the first periodic sequence is calculated and recorded as a comprehensive mean value. The comprehensive mean value is negatively correlated and mapped, and the mapping result is recorded as the periodic item stable characteristic value corresponding to the denoising left-neighborhood power data point sequence.

[0041] In the embodiment, the specific acquisition process of the neighborhood periodic component difference value of each periodic component data point in each subsequence on the first periodic sequence is as follows: for the fth periodic component data point in the dth subsequence on the first periodic sequence, if d is 1, the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d+1th subsequence on the first periodic sequence is taken as the neighborhood periodic component difference value of the fth periodic component data point in the dth subsequence; if d is D, the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d-1th subsequence on the first periodic sequence is taken as the neighborhood periodic component difference value of the fth periodic component data point in the dth subsequence; if d is neither 1 nor D, the sum of the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d+1th subsequence and the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d-1th subsequence is taken as the neighborhood periodic component difference value of the fth periodic component data point in the dth subsequence, that is, if d is neither 1 nor D, the neighborhood periodic component difference value of the fth periodic component data point in the dth subsequence is the sum of F1 and F2, F1 is the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d+1th subsequence, F2 is the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d-1th subsequence, the horizontal coordinate of the periodic component data point is time and the vertical coordinate is periodic component, and D is the number of subsequences on the first periodic sequence; and if the time length corresponding to the last subsequence on the first periodic sequence is less than the first main period, the dth subsequence is the second last subsequence on the first periodic sequence, and the number of periodic component data points in the last subsequence on the first periodic sequence is less than f, the neighborhood periodic component difference value of the fth periodic component data point in the dth subsequence is also the absolute value of the vertical coordinate difference between the fth periodic component data point in the dth subsequence and the fth periodic component data point in the d-1th subsequence on the first periodic sequence.

[0042] In addition, the specific calculation expression of the periodic term stability characteristic value corresponding to the de-noised left neighborhood electric energy data point sequence is as follows:

[0043]

[0044] wherein R2 is the periodic term stability characteristic value corresponding to the de-noised left neighborhood electric energy data point sequence, D is the number of subsequences on the first periodic sequence, Let exp() be the difference value of the comprehensive neighborhood periodic components of the d-th subsequence in the first periodic sequence. Here, exp() performs negative correlation mapping. Since the more similar adjacent subsequences (i.e., adjacent sub-curve segments) are, the higher the regularity and stability of the denoised left-neighbor energy data point sequence. And since the higher the regularity and stability of the denoised left-neighbor energy data point sequence, the higher the reliability of the prediction results obtained based on the denoised left-neighbor energy data point sequence. Furthermore, since when... The smaller the value, the higher the similarity between adjacent subsequences in the first periodic sequence. The smaller the value of R2, the larger the value of R2. Therefore, in this embodiment, when R2 is larger, it indicates that the similarity between adjacent subsequences on the first periodic sequence is higher, the regularity and stability of the denoised left neighbor power data point sequence are higher, and the reliability of the prediction result obtained based on the denoised left neighbor power data point sequence is higher. Conversely, when R2 is smaller, it indicates that the reliability of the prediction result obtained based on the denoised left neighbor power data point sequence is lower.

[0045] After obtaining the stable eigenvalues ​​of the periodic and trend terms of the denoised left-neighbor power data point sequence, the sum of the stable eigenvalues ​​of the periodic and trend terms of the denoised left-neighbor power data point sequence is used as the prediction reliability of the denoised left-neighbor power data point sequence. That is, the larger the stable eigenvalues ​​of the periodic and trend terms, or the higher the prediction reliability of the denoised left-neighbor power data point sequence, the higher the reliability or confidence of the predicted power during the power outage period obtained based on the denoised left-neighbor power data point sequence. Conversely, the smaller the stable eigenvalues ​​of the periodic and trend terms, the lower the reliability or confidence of the predicted power during the power outage period obtained based on the denoised left-neighbor power data point sequence.

[0046] After obtaining the prediction confidence of the denoised left-neighbor power data point sequence and the prediction confidence of the denoised right-neighbor power data point sequence, the prediction weight value is obtained by summing and normalizing. The higher the prediction confidence of the denoised left-neighbor power data point sequence, the higher the prediction weight value of the denoised left-neighbor power data point sequence obtained by normalization. The higher the prediction weight value of the denoised neighbor power data point sequence, the higher the contribution or participation of the predicted power of the power outage time stage obtained based on the denoised neighbor power data point sequence in the final determination of the predicted power of the power outage time stage. The specific process of obtaining the prediction weight values ​​of the denoised left-neighbor power data point sequence and the denoised right-neighbor power data point sequence based on the prediction confidence of the denoised left-neighbor power data point sequence and the denoised right-neighbor power data point sequence is as follows:

[0047] The sum of the prediction confidence of the de-noised left-neighborhood power data point sequence and the prediction confidence of the de-noised right-neighborhood power data point sequence is calculated and recorded as a comprehensive prediction confidence, and the ratio of the prediction confidence of the de-noised left-neighborhood power data point sequence to the comprehensive prediction confidence is calculated and recorded as a prediction weight value of the de-noised left-neighborhood power data point sequence, and the ratio of the prediction confidence of the de-noised right-neighborhood power data point sequence to the comprehensive prediction confidence is calculated and recorded as a prediction weight value of the de-noised right-neighborhood power data point sequence.

[0048] Therefore, the prediction weight value of the de-noised left-neighborhood power data point sequence and the prediction weight value of the de-noised right-neighborhood power data point sequence can be obtained through the above process.

[0049] The prediction module 03 is configured to obtain target predicted power data in the power-off time stage according to the prediction weight values of the de-noised left-neighborhood power data point sequence and the de-noised right-neighborhood power data point sequence.

[0050] After obtaining the prediction weight value of the de-noised left-neighborhood power data point sequence and the prediction weight value of the de-noised right-neighborhood power data point sequence, the power in the power-off time stage is predicted by using the ARIMA prediction algorithm according to the de-noised left-neighborhood power data point sequence, and the obtained prediction data is recorded as first predicted power data, and the power in the power-off time stage is predicted by using the ARIMA prediction algorithm according to the de-noised right-neighborhood power data point sequence, and the obtained prediction data is recorded as second predicted power data, that is, the first predicted power data is predicted based on the de-noised left-neighborhood power data point sequence, and the second predicted power data is predicted based on the de-noised right-neighborhood power data point sequence; since the process of using the ARIMA prediction algorithm for forward or reverse data prediction under the premise of known data used for prediction is a known technology, the embodiment will not be described in detail.

[0051] Then, the first predicted power data and the second predicted power data are weighted and fused according to the prediction weight value of the de-noised left-neighborhood power data point sequence and the prediction weight value of the de-noised right-neighborhood power data point sequence to obtain target predicted power data in the power-off time stage, and the specific process is as follows: for any power-off time in the power-off time stage, the product of the first predicted power data at the power-off time and the prediction weight value of the de-noised left-neighborhood power data point sequence is recorded as first weighted data, the product of the second predicted power data at the power-off time and the prediction weight value of the de-noised right-neighborhood power data point sequence is recorded as second weighted data, and the sum of the first weighted data and the second weighted data is recorded as target predicted power data at the power-off time.

[0052] Therefore, the embodiment can obtain the target predicted power data at each power-off time in the power-off time stage by the above manner, and finally inserts the target predicted power data at each power-off time into the corresponding power-off time position, so as to obtain the power data in the power-off time stage.

[0053] Up to now, the embodiment completes the prediction and interpolation of the power data in the power-off time stage.

[0054] In summary, the embodiment comprises a data acquisition module configured to acquire a denoised neighborhood power data point sequence corresponding to the power-off time stage of the concentrator; a prediction weight acquisition module configured to obtain a prediction credibility of the denoised neighborhood power data point sequence according to a trend component data point sequence and a periodic component data point sequence obtained by time series decomposition on the denoised neighborhood power data point sequence, and a peak point on a frequency domain signal obtained by frequency domain conversion on the periodic component data point sequence, and obtain a prediction weight value of the denoised left neighborhood power data point sequence and a prediction weight value of the denoised right neighborhood power data point sequence according to the prediction credibility of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence; and a prediction module configured to obtain target predicted power data in the power-off time stage according to the prediction weight value of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence. The embodiment can improve the accuracy of power prediction in the power-off time stage based on the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence, and the prediction weight value of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence, so as to improve the accuracy of power prediction and interpolation in the power-off time stage.

[0055] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit the same; even though the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements on some technical features thereof; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A concentrator with power outage intelligent recovery function, characterized in that, The concentrator with the power-off intelligent recovery function comprises: a data acquisition module, configured to acquire a denoised neighborhood power data point sequence corresponding to a power-off time phase of the concentrator, wherein the denoised neighborhood power data point sequence comprises a denoised left neighborhood power data point sequence and a denoised right neighborhood power data point sequence; a prediction weight acquisition module, configured to acquire a prediction credibility of the denoised neighborhood power data point sequence according to a trend component data point sequence and a periodic component data point sequence obtained by time series decomposition on the denoised neighborhood power data point sequence and a peak point on a frequency domain signal obtained by frequency domain conversion on the periodic component data point sequence, and acquire a prediction weight value of the denoised left neighborhood power data point sequence and a prediction weight value of the denoised right neighborhood power data point sequence according to the prediction credibility of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence; a prediction module, configured to acquire a target predicted power data in the power-off time phase according to the prediction weight value of the denoised left neighborhood power data point sequence and the denoised right neighborhood power data point sequence; a prediction credibility acquisition method, comprising: denoting a trend component data point sequence and a periodic component data point sequence obtained by time series decomposition on the denoised left neighborhood power data point sequence as a first trend sequence and a first periodic sequence respectively, denoting a frequency domain signal obtained by Fourier transform on the first periodic sequence as a first frequency domain signal, denoting a frequency of a peak point with the largest amplitude on the first frequency domain signal as a main frequency, taking an inverse of the main frequency as a first main period, and dividing the first periodic sequence by using the first main period to obtain each subsequence on the first periodic sequence; acquiring a trend item stability characteristic value of the denoised left neighborhood power data point sequence according to a distance between a trend component data point and a neighboring trend component data point in the first trend sequence and an inflection point in the first trend sequence, acquiring a periodic item stability characteristic value of the denoised left neighborhood power data point sequence according to a periodic component difference between periodic component data points with the same position in adjacent subsequences on the first periodic sequence, and taking a sum of the trend item stability characteristic value and the periodic item stability characteristic value as the prediction credibility of the denoised left neighborhood power data point sequence; the prediction credibility acquisition method of the denoised right neighborhood power data point sequence is the same as the prediction credibility acquisition method of the denoised left neighborhood power data point sequence.

2. The concentrator with power down intelligent recovery function of claim 1, wherein, a denoised neighborhood power data point sequence acquisition method corresponding to a power-off time phase, comprising: acquiring a left neighborhood power data point sequence corresponding to a left neighborhood time phase and a right neighborhood power data point sequence corresponding to a right neighborhood time phase of the power-off time phase; According to the distance between the power data points in the left neighborhood power data point sequence and the adjacent power data points corresponding to the power data points, noise data points in the left neighborhood power data point sequence are obtained, and the noise data points in the left neighborhood power data point sequence are replaced by interpolation to obtain a denoised left neighborhood power data point sequence; the method for obtaining a denoised right neighborhood power data point sequence is the same as the method for obtaining the denoised left neighborhood power data point sequence, and the denoised right neighborhood power data point sequence and the denoised left neighborhood power data point sequence both belong to a denoised neighborhood power data point sequence.

3. The concentrator with power down intelligent recovery function of claim 2, wherein, The method for obtaining the noise data points in the left neighborhood power data point sequence comprises: obtaining the neighborhood power distance of each power data point in the left neighborhood power data point sequence, wherein the neighborhood power distance of the power data point is the Euclidean distance between the power data point and the adjacent power data points corresponding to the power data point; for the a-th power data point in the left neighborhood power data point sequence, a is greater than 1 and less than A, A is the total number of power data points in the left neighborhood power data point sequence: the absolute value of the difference between the neighborhood power distance of the a-1-th power data point in the left neighborhood power data point sequence and the neighborhood power distance of the a-th power data point is added to the absolute value of the difference between the neighborhood power distance of the a+1-th power data point in the left neighborhood power data point sequence and the neighborhood power distance of the a-th power data point, and then the result is normalized, and the result is denoted as a comprehensive difference value; the sum of the normalized value of the neighborhood power distance of the a-th power data point and the comprehensive difference value is taken as the noise index value of the a-th power data point, and if the noise index value is greater than a preset noise threshold, the a-th power data point is recorded as a noise data point.

4. The concentrator with power down intelligent recovery function of claim 1, wherein, The method for obtaining the trend item stable characteristic value of the denoised left neighborhood power data point sequence comprises: taking the Euclidean distance between each trend component data point in the first trend sequence and the adjacent trend component data points corresponding to the trend component data point as the neighborhood trend distance of the corresponding trend component data point; obtaining the neighborhood slope difference value of each inflection point in the first trend sequence, wherein the neighborhood slope difference value of any inflection point in the first trend sequence is the absolute value of the difference between the first slope and the second slope of the inflection point, the first slope of the inflection point is the slope between the left adjacent trend component data point of the inflection point and the inflection point, and the second slope of the inflection point is the slope between the right adjacent trend component data point of the inflection point and the inflection point; obtaining the trend item stable characteristic value of the denoised left neighborhood power data point sequence according to the neighborhood trend distance of the trend component data point and the neighborhood slope difference value of the inflection point in the first trend sequence.

5. A concentrator with power down intelligent recovery functionality as recited in claim 4, wherein, The method for obtaining the trend item stable characteristic value of the denoised left neighborhood power data point sequence according to the neighborhood trend distance of the trend component data point and the neighborhood slope difference value of the inflection point in the first trend sequence comprises: The mean value of the neighborhood trend distance of all trend component data points in the first trend sequence is denoted as a neighborhood trend distance mean value, the accumulation result of the neighborhood slope difference values of all inflection points in the first trend sequence is denoted as a comprehensive neighborhood slope difference value, and the result of multiplying the neighborhood trend distance mean value and the comprehensive neighborhood slope difference value and then performing negative correlation mapping is denoted as a trend item stability characteristic value of the denoised left neighborhood electric energy data point sequence.

6. The concentrator with power down intelligent recovery functionality of claim 1, wherein, The method for obtaining the periodic item stability characteristic value of the denoised left neighborhood electric energy data point sequence comprises the following steps: For a fth periodic component data point in a dth sub-sequence on the first periodic sequence, a neighborhood periodic component difference value of the fth periodic component data point in the dth sub-sequence is obtained according to the difference between the fth periodic component data point in the dth sub-sequence and the fth periodic component data point in a sub-sequence adjacent to the dth sub-sequence; The accumulation sum of the neighborhood periodic component difference values of all periodic component data points in each sub-sequence is denoted as a comprehensive neighborhood periodic component difference value of the corresponding sub-sequence, and the result of performing negative correlation mapping on the mean value of the comprehensive neighborhood periodic component difference values of all sub-sequences on the first periodic sequence is denoted as a periodic item stability characteristic value corresponding to the denoised left neighborhood electric energy data point sequence.

7. A concentrator with power down intelligent recovery functionality as recited in claim 6, wherein, The method for obtaining the neighborhood periodic component difference value of the fth periodic component data point in the dth sub-sequence comprises the following steps: If d is 1, the absolute value of the ordinate difference value between the fth periodic component data point in the dth sub-sequence and the fth periodic component data point in a (d+1) th sub-sequence on the first periodic sequence is taken as the neighborhood periodic component difference value of the fth periodic component data point in the dth sub-sequence; if d is D, the absolute value of the ordinate difference value between the fth periodic component data point in the dth sub-sequence and the fth periodic component data point in a (d-1) th sub-sequence on the first periodic sequence is taken as the neighborhood periodic component difference value of the fth periodic component data point in the dth sub-sequence; if d is neither 1 nor D, the result of adding the absolute value of the ordinate difference value between the fth periodic component data point in the dth sub-sequence and the fth periodic component data point in the (d+1) th sub-sequence to the absolute value of the ordinate difference value between the fth periodic component data point in the dth sub-sequence and the fth periodic component data point in the (d-1) th sub-sequence is taken as the neighborhood periodic component difference value of the fth periodic component data point in the dth sub-sequence, the abscissa of the periodic component data point is time and the ordinate is periodic component, and D is the number of sub-sequences on the first periodic sequence.

8. The concentrator with power down intelligent recovery functionality of claim 1, wherein, The prediction weight value of the denoised left neighborhood electric energy data point sequence and the prediction weight value of the denoised right neighborhood electric energy data point sequence are the results of sum-normalization of the prediction credibility of the denoised left neighborhood electric energy data point sequence and the prediction credibility of the denoised right neighborhood electric energy data point sequence.

9. The concentrator with power down intelligent recovery functionality of claim 1, wherein, The method for obtaining target predicted electric energy data comprises the following steps: According to the de-noised left-neighborhood power data point sequence, the power of the power-off time stage is predicted to obtain first predicted power data of the power-off time stage; and according to the de-noised right-neighborhood power data point sequence, the power of the power-off time stage is predicted to obtain second predicted power data of the power-off time stage; For any power-off moment in the power-off time stage, the product of the first predicted power data at the power-off moment and the prediction weight value of the de-noised left-neighborhood power data point sequence is recorded as first weighted data, the product of the second predicted power data at the power-off moment and the prediction weight value of the de-noised right-neighborhood power data point sequence is recorded as second weighted data, and the sum of the first weighted data and the second weighted data is recorded as target predicted power data at the power-off moment.

Images