Component prediction method for load of electronic device
By decomposing the load sequence into different components and adopting targeted prediction methods, the problem of low accuracy in electronic device load prediction in the existing technology is solved, and more accurate load change capture and improved reliability of prediction results are achieved.
Patent Information
- Application Number
- CN202510836797.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-22
- Publication Date
- 2025-09-16
AI Technical Summary
Existing electronic device load forecasting methods are difficult to accurately capture the periodicity and randomness of load changes, resulting in low prediction accuracy and complex models.
The load sequence is decomposed into initial daily component, daily cycle component, weekly cycle component, low-frequency component and high-frequency component. Different prediction methods are used to process each component, including extrapolation, averaging or smoothing of low-frequency components and linear compensation, to improve prediction accuracy and reliability.
Through decomposition and targeted forecasting, the accuracy and reliability of load forecasting are improved, and it is suitable for load forecasting at different time scales.
Smart Images
Figure CN120654895A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electronic systems, and in particular to a method for predicting the component of an electronic device load. Background Art
[0002] Electronic device load forecasting is a crucial foundation for electronic system planning, operation, and control. Accurate load forecasting is crucial for improving the safety, economy, and reliability of electronic systems. However, electronic device loads are affected by a variety of factors, resulting in significant periodic and random variations in load, posing a challenge to load forecasting.
[0003] Traditional electronic device load forecasting methods rely primarily on statistical methods such as time series analysis and regression analysis. These methods typically assume linear or nonlinear load variations, but struggle to capture the periodic and random nature of load variations. In recent years, with the development of artificial intelligence (AI), machine learning and deep learning methods have been applied to electronic device load forecasting, achieving some success. However, these methods still face challenges such as low prediction accuracy and complex models. Summary of the Invention
[0004] The purpose of the present invention is to provide a component prediction method for electronic device loads, which improves the accuracy and reliability of load prediction by decomposing the load sequence into different components and adopting different prediction methods for different components.
[0005] The technical solutions of the present invention are as follows:
[0006] Load sequence decomposition:
[0007] The load series is decomposed into an initial daily component A0, a daily periodic component Dt, a weekly periodic component Wt, a low-frequency component Lt, and a high-frequency component Ht. A0 represents the baseline load value, Dt has a period of 24 hours, Wt has a period of 7 days, Lt is extracted using a low-pass filter, and Ht reflects the randomness of load changes.
[0008] Component prediction:
[0009] First extrapolation method: If the proportion of daily cycle components and weekly cycle components in the load sequence is greater than a threshold (such as 98%), the influence of the remaining components is ignored, and the first extrapolation method is used, using the sum of the daily cycle components and the weekly cycle components directly as the basic data for future daily load forecasts.
[0010] Second extrapolation method: If the proportion of daily cycle components and weekly cycle components in the load sequence is less than or equal to the threshold, the second extrapolation method is used to predict the daily cycle components and weekly cycle components, and the low-frequency components and high-frequency components in the remaining components are estimated and predicted.
[0011] Low frequency component prediction:
[0012] Averaging method: The low-frequency component of the forecasted day is taken as the average of the low-frequency components of the historical load at the same time.
[0013] Smoothing method: The low-frequency component of the forecasted day is taken as the smoothed result of the low-frequency component of the historical load at the same time, and the low-frequency component is processed using the idea of large near and small far.
[0014] Linear compensation:
[0015] The preliminary forecast results are linearly compensated. It is assumed that the low-frequency component of the nth day can be modeled by a linear combination of the two previous days. The parameters c1 and c2 are obtained by least squares estimation, and the forecast results are compensated.
[0016] The beneficial effects of the present invention are as follows:
[0017] Improve forecasting accuracy: By decomposing the load series into different components and using different forecasting methods for different components, the law of load changes can be captured more accurately and the forecasting accuracy can be improved.
[0018] Enhanced reliability: Through methods such as linear compensation, the prediction error can be reduced and the reliability of the prediction results can be enhanced.
[0019] Wide applicability: The method of the present invention is applicable to load forecasting at different time scales, including short-term, medium-term and long-term load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 The present invention is a flowchart of a method for predicting the component of an electronic device load. DETAILED DESCRIPTION
[0021] Electronic device load sequence:
[0022] Pt=A0+Dt+Wt+Lt+Ht
[0023] The load sequence Pt is equal to the sum of the initial daily component A0 plus the daily cycle component Dt, the weekly cycle component Wt, the low-frequency component Lt and the high-frequency component Ht.
[0024] A0 is the initial daily component, representing the baseline load value (unit: watts). Dt has a period of T and is the component of the load that varies with a 24-hour cycle. A0 + Dt is the daily load component. Wt has a period of 7T and is the weekly load component. After deducting A0, Dt, and Wt, the remaining components can be divided into Lt and Ht. Lt is the sum of the low-frequency components in the remaining components. It is extracted using a low-pass filter (such as a moving average or Butterworth filter) and reflects the impact of slower-changing related factors on the load. Ht is the sum of the high-frequency components in the remaining components and primarily reflects the randomness of load changes.
[0025] In order to conveniently represent the periodicity of short-term load, n is used to represent the date number and t is used to represent the time period number within the day, so:
[0026] Pnt=A0+Dnt+Wnt+Lnt+Hnt
[0027] The load sequence Pnt on the nth day is equal to the initial daily component A0 plus the sum of the daily cycle component Dnt, weekly cycle component Wnt, low frequency component Lnt and high frequency component Hnt on the nth day.
[0028] If the sum of the daily component A0+Dnt and the weekly component Wnt obtained by Fast Fourier Transform (FFT) frequency domain decomposition accounts for a high proportion of the load sequence, when it is greater than a certain threshold condition such as 98%, then we can completely ignore the influence of the remaining components and use the first extrapolation method to use the sum of the daily and weekly components directly as the basic data for future daily load forecasts:
[0029] P`nt=A0+Dnt+Wnt,n=Cl,C-2,…,CN; t=1,2,…,T
[0030] The predicted load sequence P'nt is equal to the sum of the initial daily component A0 plus the daily cycle component Dnt and the weekly cycle component Wnt on the nth day, where n ranges from Cl to CN and t ranges from 1 to T.
[0031] Among them, C is the base date and N is the number of historical days.
[0032] If the forecast object is day C+F, where F is the number of days in the future forecast, then T consecutive points that match the day of the week of day C+F are taken as the forecast result for that day. It should be emphasized that the above condition is only met when the proportion of the residual component is very small and can be ignored.
[0033] If the proportion of the daily cycle component and the weekly cycle component in the load sequence is less than or equal to the threshold condition, the remaining components need to be calculated. In this case, the daily cycle component and the weekly cycle component are obtained by the second extrapolation method, and the prediction process is relatively simple; therefore, after the frequency domain decomposition, the low-frequency component and the high-frequency component in the remaining components are estimated and predicted. Figure 1 shown.
[0034] 1. Preliminary Forecast
[0035] Here are two preliminary prediction methods for calculating the remaining components through the second extrapolation:
[0036] (1) Frequency domain component method using low-frequency component averaging. The low-frequency component of the day to be predicted is the average of the low-frequency components of the historical load at the same time, while ignoring the high-frequency components that are more difficult to predict. Therefore, the approximate load on the nth day in the historical date can be approximated by the following formula:
[0037]
[0038] The recommended value for the number of historical days N is 7<=N<=14.
[0039] The predicted low-frequency component L'nt is equal to the average value of the historical low-frequency component Lnt from CN to C-1.
[0040] P'nt=A0+Dnt+Wnt+L'nt
[0041] If the prediction object is the C+F day, then take T consecutive points that are consistent with the weekday type of the C+F day as the prediction result for that day:
[0042] P(C+F)t=P`(C+F-7)t
[0043] Among them, the week type of day C+F is the same as that of day C+F-7.
[0044] (2) Frequency domain component method of low-frequency component smoothing. The low-frequency component of the forecast day is taken as the smoothed result of the low-frequency component of the historical load at the same time. This method treats the low-frequency component with the idea of near-large and far-small, but also ignores the influence of high-frequency components.
[0045] Take the smoothing coefficient as a, a∈[0,1], it is recommended to take 0.8~0.95, adjust according to the load fluctuation, the weight of the near big and far small on the nth day in history is a-(nC), n=C-1,C-2,…,CN, the weight of each day is the sum of the weights Normalize. Then, the low-frequency component of the day to be predicted can be predicted using the following formula:
[0046]
[0047] The predicted low-frequency component L`(C+F)t is equal to the weighted average of the historical low-frequency components Lnt from CN to C-1, with a weight of a-(nC), and is normalized.
[0048] Then, if the prediction object is day C+F, the daily cycle component and weekly cycle component of day C+F-7 in the historical date that is consistent with the weekday type of day C+F are superimposed with the low-frequency component of the day to be predicted as the prediction result for that day:
[0049] P(C+F)t=A0+D(C+F-7)t+W(C+F-7)t+L`(C+F)t
[0050] Among them, the week type of day C+F is the same as that of day C+F-7.
[0051] 2. Linear compensation
[0052] After the preliminary prediction, linear compensation is performed on the preliminary prediction results;
[0053] Assume that the low-frequency component of day n can be modeled by a linear combination of the two previous days:
[0054] Lnt=c1*L(n-1)t+c2*L(n-2)t
[0055] The low-frequency component Lnt of the nth day can be modeled by a linear combination of the low-frequency components L(n-1)t and L(n-2)t of the previous two days, where c1 and c2 are the coefficients of the linear combination.
[0056] Therefore, for the analysis of historical load, the parameters c1 and c2 can be obtained by the least squares estimation of historical data, and the objective function is:
[0057] Min∑[Lnt-c1*L(n-1)t-c2*L(n-2)t]2
[0058] The fitting results may be c1=0.3, c2=0.7.
[0059] If the prediction object is day C+F, the low-frequency component of the day to be predicted can be predicted using the following formula:
[0060] L`(C+F)t=c1*L(C+F-1)t+c2*L(C+F-2)t
[0061] The predicted low-frequency component L`(C+F)t is equal to the linear combination of the low-frequency components L(C+F-1)t and L(C+F-2)t of the previous two days, where c1 and c2 are the coefficients of the linear combination.
[0062] Then, take the daily cycle component and weekly cycle component of day C+F-7 in the historical date that is consistent with the week type of day C+F and superimpose them with the low-frequency component of the day to be predicted as the prediction difference result for that day:
[0063] Q(C+F)t=A0+D(C+F-7)t+W(C+F-7)t+L`(C+F)t
[0064] Among them, the week type of day C+F is the same as that of day C+F-7.
[0065] The linear compensation afterload sequence is:
[0066] Px(C+F)t=P(C+F)t+r*|P(C+F)tQ(C+F)t|
[0067] Wherein r is the compensation coefficient, for example, r=0.5.
[0068] The load sequence after linear compensation Px(C+F)t is equal to the predicted load sequence P(C+F)t plus the compensation coefficient r multiplied by the absolute value of the predicted difference |P(C+F)tQ(C+F)t|.
[0069] The above-mentioned electronic components include Ishizuka CRD constant current diodes, Xingqin series transient suppression diodes, MMBT3906TSOT523 transistors and Juding series electrostatic suppression tubes.
[0070] The purpose of the present invention is to provide a component prediction method for electronic device loads, which improves the accuracy and reliability of load prediction by decomposing the load sequence into different components and adopting different prediction methods for different components.
[0071] The beneficial effects of the present invention are as follows:
[0072] Improve forecasting accuracy: By decomposing the load series into different components and using different forecasting methods for different components, the law of load changes can be captured more accurately and the forecasting accuracy can be improved.
[0073] Enhanced reliability: Through methods such as linear compensation, the prediction error can be reduced and the reliability of the prediction results can be enhanced.
[0074] Wide applicability: The method of the present invention is applicable to load forecasting at different time scales, including short-term, medium-term and long-term load forecasting.
[0075] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the scope of the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the spirit and scope of protection of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present invention.
Claims
1. A method for predicting the component of electronic device load, characterized in that: The following steps are involved: Decompose the load sequence into the initial daily component A0, daily periodic component Dt, weekly periodic component Wt, low-frequency component Lt and high-frequency component Ht; According to the proportion of daily cycle component and weekly cycle component in the load sequence, the first extrapolation method or the second extrapolation method is selected for prediction; Use averaging or smoothing to predict low-frequency components; Perform linear compensation on the preliminary prediction results.
2. The method according to claim 1, characterized in that In the load sequence decomposition step, the daily cycle component and the weekly cycle component are extracted by fast Fourier transform (FFT), and the low-frequency component is extracted by a low-pass filter.
3. The method according to claim 1, characterized in that The first extrapolation method includes: Calculate the sum of the daily cycle component and the weekly cycle component as the basic data for future daily load forecasting; Take T consecutive points that are consistent with the week type of the forecast object as the forecast result for that day.
4. The method according to claim 1, wherein The second extrapolation method includes: Predict daily and weekly cycle components; Use averaging or smoothing to predict low-frequency components.
5. The method according to claim 4, characterized in that The averaging methods include: Calculate the average value of the low-frequency component of the historical load as the low-frequency component of the day to be predicted.
6. The method according to claim 4, characterized in that The smoothing method includes: The smoothed value of the low-frequency component of the historical load is calculated as the low-frequency component of the day to be predicted, where the value range of the smoothing coefficient a is 0.8<=a<=0.
95.
7. The method according to claim 1, characterized in that In the linear compensation step, the low-frequency component of the nth day can be modeled by a linear combination of the two previous days, and the parameters c1 and c2 are obtained by least squares estimation, and the prediction results are compensated.